The Bureau of Economic Analysis (BEA) recently reported that U.S. real GDP expanded at an annualized rate of 3.4% in the third quarter of 2026, substantially exceeding Atlas Analytics’ forecast of approximately 1.5%.

At first glance, this represents a meaningful discrepancy between our forecast and the government’s initial estimate. However, we believe the advance release overstates the underlying pace of economic growth.

Our expectation is that Q3 GDP will ultimately be revised downward by approximately 0.7–1.0 percentage points, settling in the mid-2% range.

While this would still represent a miss relative to our original forecast, it would suggest that the underlying economy was considerably weaker than the advance estimate implies.

Our expectation is driven by three principal considerations: continued resilience in fixed investment, weaker-than-reported personal consumption, and deterioration in international trade.

Fixed Investment: A Source of Continued Strength

We expect fixed investment to remain a relatively strong component of economic activity.

Investment in equipment, intellectual property, and structures continues to support economic growth, even as other areas of the economy exhibit signs of moderation.

Our models do not suggest that fixed investment will be the principal source of downward revisions to Q3 GDP.

This distinction matters. We are not forecasting a broad-based collapse in economic activity. Instead, we see an economy in which certain investment-related sectors remain resilient while other components of aggregate demand are less robust than the headline GDP figure suggests.

One particularly important source of investment growth has been the expansion of artificial intelligence infrastructure, including data centers, computing equipment, and associated energy infrastructure.

While this investment has helped support economic activity, its growing importance also raises questions about the sustainability of growth in 2027, an issue we will explore in greater detail next week.

Personal Consumption: Is the Consumer Really That Strong?

Personal consumption expenditures represent approximately two-thirds of U.S. GDP, making household spending the single most important determinant of overall economic growth.

We believe the advance estimate overstates the underlying strength of consumer spending during the third quarter.

Advance GDP estimates necessarily rely on incomplete source data, including preliminary information on retail sales, services consumption, and other measures of household expenditures.

As more comprehensive data become available, these estimates can change materially.

Our expectation is that subsequent releases will reveal weaker consumption growth than initially reported, contributing to a downward revision in headline GDP.

Importantly, this does not necessarily imply that consumer spending is contracting. Rather, we believe the rate of consumption growth is more consistent with a moderating economy than with the strength implied by the advance estimate.

Net Exports: What Satellite Imagery Is Telling Us

Perhaps the most distinctive element of our forecast concerns international trade.

At Atlas Analytics, we developed JACK, a proprietary computer-vision algorithm that analyzes satellite imagery to monitor international trade activity.

By observing physical economic activity through satellite imagery, JACK provides an independent perspective on trade flows that complements conventional government statistics.

Our satellite-based indicators continue to suggest weakness in net exports.

Recent official trade data reinforce this concern. According to the Census Bureau and BEA, the monthly U.S. goods-and-services trade deficit widened from approximately $71 billion in June to $106 billion in August.

That represents a substantial deterioration in the monthly trade balance.

Although monthly nominal trade deficits do not translate mechanically into quarterly real GDP contributions, these developments are consistent with the broader weakness identified by our satellite-based models.

We therefore expect that more complete trade data will support a weaker net-export contribution than currently reflected in the advance GDP estimate.

This is precisely the type of application for which satellite-based economic measurement can be valuable: providing an independent assessment of economic activity when conventional statistical releases are incomplete or subject to revision.

GDP Revisions Are Larger Than Many Investors Realize

One of the challenges in evaluating GDP forecasts is that the first government estimate is not necessarily the final measure of economic growth.

The BEA releases multiple estimates of quarterly GDP as additional source data become available. Longer-term annual revisions can further change the historical record.

Our analysis of historical BEA revisions indicates the following average absolute differences:

These figures represent absolute revisions, meaning they capture changes in either direction.

They should not be interpreted as evidence that the BEA systematically revises GDP downward.

Nevertheless, they illustrate an important point: the difference between an advance GDP estimate and a later estimate can be economically meaningful.

A revision of 0.7–1.0 percentage points, while substantial, is not unprecedented in the historical GDP revision process.

The relevant question is therefore not simply whether our forecast differs from the advance estimate, but whether the underlying economic evidence supports that difference.

We believe it does.

A Familiar Pattern: What Happened Last Quarter?

Frequent readers may recall that we encountered a similar discrepancy just last quarter.

The BEA initially reported GDP growth of 1.5%, compared with Atlas Analytics’ forecast of 1.95%. Subsequent revisions raised the official estimate to 2.2%, bringing it within just 0.25 percentage points (25 basis points) of our original forecast.

In other words, what initially appeared to be a 0.45-percentage-point forecasting error narrowed considerably as more complete economic data became available.

This experience reinforces an important distinction between forecasting the BEA’s advance estimate and forecasting the underlying pace of economic growth.

To be clear, one quarter does not establish a systematic pattern, nor does it guarantee that Q3 GDP will be revised downward. Indeed, last quarter’s revision was upward, whereas we anticipate a downward revision this quarter.

Nevertheless, it illustrates why evaluating economic forecasts solely against preliminary government statistics can be misleading.

We anticipate a similar convergence this quarter, although toward our updated assessment of approximately 2.4%, rather than our original forecast of 1.5%.

What Would Prove Our Forecast Wrong?

Economic forecasting requires intellectual discipline, particularly when an initial official release diverges from a model’s expectations.

Our original Q3 forecast was 1.5%, while our current assessment is that GDP will ultimately settle near 2.4%. These are distinct estimates, and we intend to evaluate our original forecast against the eventual official figures. If growth remains near 3.4%, it would represent a substantial forecasting error and warrant a careful examination of our underlying models.

However, this is a testable prediction, not a certainty.

If subsequent BEA estimates continue to show growth near 3.4%, that would represent a meaningful forecasting error for Atlas Analytics.

In that scenario, we will examine the discrepancies at the component level, evaluate the performance of our underlying indicators, and identify opportunities to improve our models.

Conversely, if consumption and net exports are revised downward as anticipated, it would provide further evidence of the value of independent, satellite-based economic measurement.

Either outcome offers useful information about the economy and the performance of our forecasting methodology.

Looking Ahead: Can Investment Continue to Carry the Economy?

While we expect consumption and net exports to weigh on Q3 GDP revisions, fixed investment remains an important source of economic resilience.

One particularly consequential development is the extraordinary expansion of investment in artificial intelligence infrastructure, including data centers, computing equipment, and related power infrastructure.

This investment has helped support economic growth even as other components of aggregate demand have shown signs of moderation.

But it raises an important question: How sustainable is this investment boom?

As data-center development becomes increasingly capital-intensive, financing conditions, debt obligations, and expectations for future AI-related revenues will become critical determinants of investment activity.

If financing conditions tighten or expected returns disappoint, a slowdown in data-center construction and equipment spending could become a meaningful headwind for GDP growth in 2027.

We are not yet forecasting such a reversal. However, the concentration of investment in AI-related infrastructure warrants closer examination, particularly if consumer spending continues to moderate.

Next week, we’ll examine the growing role of debt in financing the data-center boom, estimate how much AI infrastructure investment is contributing to U.S. GDP, and consider what a pullback could mean for economic growth in 2027.

The Bottom Line

The advance estimate suggests that the U.S. economy expanded at a robust 3.4% annualized rate in Q3 2026.

Our satellite-based indicators tell a more cautious story.

We see continued resilience in fixed investment, but weaker underlying momentum in personal consumption and international trade.

Taken together, these signals suggest that the economy grew closer to 2.4% than 3.4% during the third quarter.

The coming BEA revisions will provide an important test of that assessment.

At Atlas Analytics, our objective is not to reproduce the government’s preliminary estimates. It is to develop independent, forward-looking measures of economic activity that help investors, businesses, and policymakers better understand the economy.

Sometimes those measures will agree with the initial official statistics. Other times, as in this quarter, they will diverge.

What matters is whether they ultimately provide a more accurate picture of economic reality.

Atlas Analytics develops proprietary macroeconomic indicators and forecasts using satellite imagery, computer vision, and advanced statistical modeling.

Atlas Analytics | Atlas Analytics Research

Disclaimer: The forecasts and opinions expressed here are subject to uncertainty and revision and do not constitute investment advice.

Last week, we argued that inflation was proving more persistent than many economists had anticipated, strengthening the case for interest rates remaining elevated for longer. This week, we’re taking that analysis one step further: How much higher might interest rates actually go, and is the market getting ahead of itself?

To answer that question, we compared three different perspectives on the trajectory of U.S. interest rates through the end of 2027: the Federal Reserve’s own projections, market-implied expectations derived from CME FedWatch, and Atlas Analytics’ econometric forecasting model.

The results suggest an interesting divergence. While all three perspectives point toward a relatively restrictive interest-rate environment, the market anticipates substantially more tightening than our model indicates is necessary.

What the Market Is Pricing

The CME FedWatch Tool uses prices in federal funds futures markets to estimate the probabilities of future Federal Reserve policy decisions. These probabilities provide a useful window into how traders collectively assess the likely path of monetary policy.

We compared those market expectations with the Federal Reserve’s projected interest-rate path and forecasts generated by Atlas Analytics.

Source: Atlas Analytics analysis, Federal Reserve Summary of Economic Projections (SEP), and CME FedWatch, October 2026. CME FedWatch estimates are calculated as probability-weighted averages of potential federal funds rate outcomes for each quarter. Federal Reserve projections are based on year-end SEP estimates, interpolated to quarterly intervals.

Three observations stand out.

First, the market expects a relatively rapid increase in interest rates. CME FedWatch estimates imply rates rising from 4.06% in Q4 2026 to 4.67% by Q4 2027, an increase of approximately 61 basis points.

Second, Atlas anticipates a much more gradual adjustment. Our model projects rates increasing from 3.96% to 4.28% over the same period, a rise of just 32 basis points.

Third, the divergence is most pronounced in the middle of 2027. By Q2 2027, the market-implied rate exceeds the Atlas forecast by 45 basis points. Although the gap narrows slightly toward year-end, it remains substantial at 39 basis points in Q4.

The implication is straightforward: the market may be pricing a more aggressive Federal Reserve response than underlying economic fundamentals ultimately justify.

What Is Driving the Atlas Forecast?

Unlike market-implied expectations, which reflect the collective positioning and expectations of futures traders, our econometric model estimates the path of interest rates based on relationships between monetary policy and broader macroeconomic conditions.

For this exercise, we hold two important assumptions constant:

• Real GDP growth of 2.5% annually. This represents a relatively resilient economic expansion, rather than a recessionary or sharply slowing economy.

• Inflation of 2.35%. This assumption is based on the approximately 2.35% inflation breakeven rate on 10-year Treasury securities, which provides a market-based measure of expected average inflation over the coming decade.

Under these assumptions, the model suggests that monetary policy would remain relatively restrictive, but that the Federal Reserve would not need to tighten as aggressively as current market pricing implies.

This distinction matters.

If economic growth remains steady and inflation gradually stabilizes near market-implied long-term expectations, the case for rapid additional tightening becomes less compelling. Interest rates could continue to rise, but at a measured pace rather than through a series of more aggressive policy adjustments.

Of course, these forecasts are conditional. Stronger-than-expected inflation, accelerating economic growth, or changes in the Federal Reserve’s reaction function could produce a different outcome.

Equally, a slowdown in economic activity could reduce the need for further tightening.

Our forecast should therefore be interpreted not as a prediction of precisely what the Federal Reserve will do, but as an estimate of what historical economic relationships suggest it might do under a specified macroeconomic scenario.

Higher for Longer Doesn’t Necessarily Mean Higher Faster

The distinction between the direction of interest rates and the speed of adjustment is particularly important for investors.

Last week’s analysis emphasized the persistence of inflation and the resulting challenges for a return to accommodative monetary policy.

That conclusion remains intact.

But there is an important difference between a Federal Reserve that gradually raises rates toward 4.25% and one that moves relatively quickly toward 4.75%.

For financial markets, even a difference of 40–50 basis points can have meaningful implications for Treasury valuations, equity discount rates, corporate borrowing costs, and the relative attractiveness of different asset classes.

If Atlas is correct and the Federal Reserve ultimately follows a more gradual path than futures markets currently anticipate, interest-rate-sensitive assets could benefit from a repricing of policy expectations.

Longer-duration bonds, in particular, may become more attractive if investors begin to revise their expectations for future short-term interest rates downward. Growth equities and other assets sensitive to discount rates could also benefit, although their performance would depend on the broader economic outlook.

Conversely, if inflation remains stubbornly elevated and the market’s more aggressive trajectory proves correct, those same assets could face continued pressure.

This is precisely why comparing macroeconomic forecasts with market-implied expectations is valuable. The question for investors is not simply whether interest rates will rise, but whether they will rise by more or less than the market already expects.

The Bottom Line

Our analysis reinforces last week’s central conclusion: the era of persistently low interest rates appears unlikely to return anytime soon.

But it also introduces an important qualification.

Higher for longer? Probably. Higher as quickly as the market expects? Perhaps not.

Under a scenario of 2.5% real GDP growth and 2.35% inflation, Atlas Analytics forecasts interest rates reaching approximately 4.28% by the end of 2027, compared with the 4.67% implied by CME FedWatch estimates.

That difference is not enormous in absolute terms, but it could be consequential for financial markets.

As we continue monitoring incoming economic data, the key question will be whether inflation and growth evolve in ways that justify the market’s relatively aggressive expectations—or whether the Federal Reserve ultimately has room to move more gradually.

For now, our models suggest that the market may be getting a little ahead of itself.

Atlas Analytics develops macroeconomic forecasts using alternative data, satellite imagery, machine learning, and econometric modeling. The analysis presented here is for informational purposes only and does not constitute investment advice.

September’s Consumer Price Index (CPI) rose 3.7% year over year, accelerating from 3.4% in August and exceeding economists’ expectations of 3.6%. The market reaction was swift. The Nasdaq fell 1.9%, the S&P 500 declined 1.1%, and small-cap stocks suffered a similarly sharp selloff.

The catalyst was straightforward. Higher inflation increases the likelihood that the Fed will maintain restrictive monetary policy—or raise interest rates further—to bring inflation back toward its 2% target.

But at Atlas Analytics, we believe the market is missing a bigger story.

The era of near-zero interest rates is over, and investors have yet to fully adjust to what comes next.

The chart above should look familiar to longtime readers. We first published it on February 1 of this year, when we forecast that the Federal Reserve would embark on another rate-hiking cycle rather than deliver the cuts anticipated by consensus.

Eight months later, our conviction remains unchanged. We continue to expect the federal funds rate to approach 5% over the medium term, substantially above the Fed’s projected path.

Higher for Longer: A New Interest Rate Reality

We’ve been anticipating persistent inflation and a more hawkish Federal Reserve for some time.

Our thesis is not simply that the Fed might raise rates again this year. Rather, we believe investors have become accustomed to an interest-rate environment that was historically exceptional.

Following the 2008 financial crisis, the United States experienced an extraordinary period of monetary accommodation. Interest rates hovered near zero, quantitative easing suppressed long-term borrowing costs, and investors became accustomed to cheap capital.

But zero interest rates were the anomaly, not the norm.

For much of modern American economic history, nominal interest rates of 4–5% would hardly have been considered unusual. We believe the economy may be returning to something closer to that historical reality.

Our expectation is that the federal funds rate will remain closer to 5% over the medium term, rather than returning to the near-zero levels that characterized much of the 2010s.

Structural forces, including persistent fiscal deficits, changing global supply chains, and a potentially higher neutral rate of interest, could reinforce this environment.

This does not mean rates must rise indefinitely. It means that even after the Fed finishes tightening, the destination may be considerably higher than investors once expected.

And that has profound implications for asset valuations.

What Does This Mean for Equities?

The relationship between interest rates and equities is often understood through the equity risk premium (ERP): the additional return investors demand for owning stocks rather than relatively safe government bonds.

As Treasury yields rise, investors can earn higher returns without assuming equity risk. All else equal, this reduces the relative attractiveness of stocks and places downward pressure on valuations.

The relationship is imperfect. Corporate earnings growth and changes in risk appetite can offset higher discount rates. But the underlying mathematics is difficult to escape.

Growth companies are particularly sensitive because a disproportionate share of their expected value comes from earnings projected far into the future. Higher discount rates reduce the present value of those earnings.

This is especially important for technology companies trading at elevated multiples.

Conversely, value-oriented businesses with established cash flows and more conservative valuations may become relatively attractive.

We therefore anticipate a period of increasing differentiation across equities, with expensive growth stocks facing valuation pressure and selected value opportunities potentially outperforming.

The recent selloff in technology and small-cap equities may be an early illustration of this repricing.

Where We See Opportunity: Four ETF Convictions

Our macroeconomic outlook informs four distinct investment views: bearish on technology and energy, and bullish on long-duration Treasuries and communication services.

Short XLK: Technology Has Priced In Too Much Optimism

We’ve been skeptical of elevated technology valuations for some time.

The AI revolution is real, and many leading technology companies are exceptionally profitable. But a transformative technology does not automatically make every investment attractive at every price.

In our view, investors have extrapolated extraordinary growth expectations too far into the future, leaving valuations vulnerable to even modest disappointments.

As interest rates remain elevated, the discount applied to those future earnings becomes increasingly consequential.

We expect this pressure to weigh on the Technology Select Sector SPDR Fund (XLK), particularly where valuations already assume years of exceptional growth.

Smaller-cap companies may face additional pressure from higher borrowing costs and refinancing risks, although their valuations and balance sheets vary considerably.

Our concern isn’t the quality of the technology. It’s that investors are over paying for it.

Long TLT: Treasury Yields May Have Overshot

Our bullish view on long-duration Treasuries may initially appear contradictory. We were after all short long-bonds for most of last year.

If interest rates are going to remain higher for longer, why favor the iShares 20+ Year Treasury Bond ETF (TLT), which typically loses value when long-term yields rise?

The answer is that markets can overshoot.

Economist Rudiger Dornbusch famously demonstrated how exchange rates can overshoot their long-run equilibrium following monetary shocks. While his original model concerned foreign exchange, the underlying intuition applies more broadly: financial prices can move beyond levels ultimately justified by fundamentals.

We believe something similar may be occurring in the Treasury market.

Long-term yields reflect not only expectations for future Fed policy but also inflation expectations, real interest rates, and the term premium investors demand for holding longer-maturity bonds.

Our view is that the recent rise in long-term yields may have gone too far, even accounting for a structurally higher federal funds rate.

This distinction is critical. We can believe the Fed will remain hawkish while simultaneously believing long-term Treasury bonds have become oversold.

TLT could certainly decline further, particularly if inflation remains stubborn or Treasury risk premiums continue rising. But we see an increasingly attractive opportunity for a reversal in long-duration yields as markets reassess their longer-term expectations.

Long XLC: AI Disruption Fears May Be Overdone

While we’re cautious on technology valuations broadly, we see a more attractive opportunity in the Communication Services Select Sector SPDR Fund (XLC).

The fund is heavily concentrated in Alphabet and Meta, two companies facing considerable uncertainty surrounding artificial intelligence.

Investors worry that generative AI and alternative search platforms could erode Google’s dominance in search, while increasing competition and capital expenditure could pressure Meta’s profitability.

These are legitimate concerns. But we believe the market may be overestimating the speed and magnitude of the disruption.

Alphabet and Meta retain enormous advantages in distribution, advertising infrastructure, user networks, and financial resources. Both also have opportunities to deploy AI to improve their existing products and monetization.

Importantly, we see a distinction between companies whose valuations depend on extraordinary future AI growth and established businesses whose current earnings may be undervalued because of fears surrounding AI.

We believe XLC offers a compelling relative-value opportunity, provided its earnings durability remains intact.

Short XLE: Energy’s Geopolitical Premium Looks Vulnerable

Our bearish outlook on the Energy Select Sector SPDR Fund (XLE) is driven less by conventional valuation concerns and more by commodity-price expectations.

The conflict involving Iran and uncertainty surrounding Russian energy supplies have contributed to elevated oil and gasoline prices. Indeed, gasoline prices rose 9.1% in September alone, accounting for more than half of the monthly increase in headline CPI.

Energy companies have benefited from this environment, and some may be reasonably valued based on prevailing commodity prices.

But that is precisely the risk.

If geopolitical tensions ease, supply disruptions moderate, or arrangements involving Russian energy exports improve market availability, today’s elevated prices could reverse relatively quickly.

In that scenario, energy-sector earnings expectations could weaken, putting downward pressure on XLE.

We acknowledge that further geopolitical escalation could produce the opposite outcome. Nevertheless, we believe the balance of risks increasingly favors a normalization in energy prices rather than a sustained continuation of the recent surge.

The Bigger Picture: Capital Is No Longer Free

For more than a decade, investors operated in an environment where capital was exceptionally cheap.

Low discount rates supported elevated valuations, encouraged speculation, and rewarded businesses whose profits lay far in the future.

That environment is changing.

A world of persistently higher interest rates places greater importance on current earnings, sustainable cash flows, balance-sheet strength, and valuation discipline.

It also creates opportunities.

Some assets remain priced for a return to cheap money. Others, including long-duration Treasuries, may have already overreacted to the prospect of permanently higher rates.

At Atlas Analytics, we believe investors should distinguish between the direction of monetary policy and the extent to which its consequences are already reflected in market prices.

Our view is that the Fed will remain more hawkish than many market participants anticipate. But that doesn’t mean every asset exposed to higher rates is necessarily a bad investment.

The next phase of the market will not simply reward those who correctly predict interest rates. It will reward those who identify where expectations have gone too far.

And in a world where the cost of capital matters again, valuation discipline may be more important than ever.

Atlas Analytics provides independent macroeconomic research and forecasting. The views expressed are forward-looking opinions as of publication and are subject to change. ETF positions discussed represent directional market views and are not individualized investment recommendations. All investments, including short positions and long-duration bonds, involve risk.

The U.S. economy continues to expand, but international trade is weighing on headline growth. Atlas Analytics forecasts that real U.S. GDP grew at a 1.5% annualized rate in the third quarter of 2026, down from 2.2% in Q2 and 2.5% in Q1.

Our latest forecast suggests that underlying domestic economic activity remains relatively healthy. However, a widening trade deficit is expected to subtract substantially from headline GDP growth.

This forecast follows two consecutive quarters in which Atlas’s satellite-based GDP estimates came remarkably close to the Bureau of Economic Analysis (BEA) figures.

Our Q3 2026 Forecast: 1.5% Growth

Atlas Analytics’ final estimate for Q3 2026 is 1.5% annualized real GDP growth, with three principal components:

• Core GDP: +2.6 percentage points. Private consumption, fixed investment, and government expenditures continue to support economic growth.

• Private inventories: +0.5 percentage points. Inventory accumulation is expected to make a modest positive contribution.

• Net exports: −1.5 percentage points. A widening trade deficit is expected to significantly offset domestic economic activity.

Together, these components produce our headline forecast of 1.5%.

The distinction between headline GDP and underlying domestic activity is particularly important this quarter.

Our forecast suggests that the U.S. economy is growing more strongly domestically than the headline figure implies. However, rising imports relative to exports are expected to weigh heavily on measured GDP.

Recent trade data reinforce this interpretation. The U.S. trade deficit widened sharply in August, with imports reaching record levels amid continued demand for industrial supplies, capital equipment, and technology-related investment.

In other words, weaker headline GDP growth does not necessarily indicate a comparable deterioration in domestic demand.

Two Consecutive Quarters of Accurate Forecasting

Our Q3 forecast builds on encouraging results from the first half of 2026.

In Q1, Atlas forecast 2.04% GDP growth, compared with the BEA’s latest estimate of 2.49%.

In Q2, Atlas again forecast 1.95%, compared with the BEA’s latest estimate of 2.22%—a difference of just 0.27 percentage points.

Across these two quarters, our average absolute forecasting error was just 0.36 percentage points.

While two quarters represent a relatively small sample, these results are encouraging evidence of the potential for satellite-based economic forecasting.

Our objective is not simply to replicate traditional economic forecasts. It is to develop an independent perspective on economic activity using alternative data sources that can complement conventional government statistics.

Forecasting the Economy from Outer Space

At Atlas Analytics, we combine satellite imagery, geospatial information, and conventional economic data to develop timely estimates of economic activity.

Our approach seeks to identify real-world signals of production, transportation, consumption, and trade—often before these activities are fully reflected in official statistics.

This is particularly valuable when conventional economic indicators are incomplete, delayed, or subject to substantial revisions.

The third quarter illustrates why examining the individual components of GDP matters. Our models suggest that domestic demand continues to expand, even as international trade reduces the overall growth rate.

What Comes Next?

The BEA is scheduled to release its advance estimate of Q3 2026 GDP on October 29, 2026.

Our final forecast of 1.5% will provide another opportunity to evaluate how well our satellite-based approach anticipates official economic statistics.

Regardless of the outcome, our goal remains the same: to make economic forecasting more timely, transparent, and grounded in observable economic activity.

Thank you for following Atlas Analytics as we continue developing a new approach to understanding the economy—from outer space.

Atlas Analytics | Macroeconomic Forecasting from Outer Space

Forecasts are estimates, not official BEA statistics. All growth rates are annualized; component contributions are expressed in percentage points. Historical BEA estimates are subject to revision.

On July 26, Atlas Analytics published our forecast for second-quarter U.S. GDP.

Our estimate: 1.95% annualized real GDP growth — effectively 2.0%.

Four days later, the Bureau of Economic Analysis released its advance estimate.

BEA Advance Estimate: 1.5%.

A month later, the government received additional data and published its second estimate.

BEA Second Estimate: 1.5%.

And today, September 30, the BEA released its third estimate, incorporating substantially more complete information about what actually happened across the U.S. economy during the second quarter.

BEA Third Estimate: 2.2%.

The progression looks like this:

Estimate

Release Date

Q2 2026 GDP

Atlas Analytics

July 26

1.95%

BEA Advance Estimate

July 30

1.5%

BEA Second Estimate

August 26

1.5%

BEA Third Estimate

September 30

2.2%

The government estimate moved 0.7 percentage point upward between its initial and latest estimates.

Atlas’s forecast didn’t.

Forecasting Before the Data Are Complete

This distinction gets at the fundamental challenge of economic forecasting.

The BEA is tasked with measuring an enormous and extraordinarily complex economy. Its estimates become more accurate as additional surveys, administrative records, and other source data become available.

That’s why GDP is revised.

Today’s revision primarily reflected stronger estimates of investment, consumer spending, and government spending than were captured in the previous estimate.

There is nothing unusual about that process. In fact, it illustrates the problem Atlas is trying to solve.

Investors, businesses, and policymakers need to understand what is happening in the economy before all of the traditional data have arrived.

By the time the statistical picture becomes complete, the economic activity being measured may have occurred months earlier.

Seeing the Economy Earlier

Atlas approaches this problem differently.

Rather than relying exclusively on traditional economic releases, we use satellite imagery, computer vision, and machine learning to observe physical economic activity as it happens.

For Q2, those models pointed to an economy growing at roughly 2.0%.

The government’s first two estimates suggested something weaker: 1.5%.

Today, with substantially more information available, the BEA estimates growth at 2.2%.

Our forecast was 0.25 percentage point away from today’s estimate.

More importantly, it was available before the government’s first estimate was published.

This Is Why Timing Matters

A forecast isn’t particularly valuable if it tells you what happened after everyone else already knows.

The objective is to produce reliable information while decisions can still be made.

That is especially important because the economic statistics we rely on are estimates themselves. They evolve as new information arrives.

Q2 provides a particularly clear example:

July 26 — Atlas: 1.95%

July 30 — BEA: 1.5%

August 26 — BEA: 1.5%

September 30 — BEA: 2.2%

The official data eventually painted a picture remarkably close to the one our models were seeing months earlier.

GDP Is the Score. The Economy Is the Game.

We’ve said before that Atlas isn’t fundamentally a GDP forecasting company.

GDP is our scorecard.

Every quarter gives us another opportunity to test whether signals derived from the physical economy can tell us something useful before conventional statistics become available.

Sometimes we’ll be right. Sometimes we’ll be wrong. What matters is building a transparent track record and continuously testing the models against the official data as they arrive.

The bigger opportunity extends well beyond GDP.

The same infrastructure that observes economic activity from space can measure ports, manufacturing facilities, logistics hubs, construction sites, commercial properties, data centers, and other physical assets.

Instead of waiting for surveys and financial statements to tell us what happened, we increasingly have the ability to observe economic activity directly.

That is the future Atlas is building toward:

Economic intelligence based not only on what the economy reported yesterday, but on what the physical economy is doing today.

The September Decision

On September 16, the Federal Reserve raised its target range for the federal funds rate by 25 basis points, to 3.75%–4%. The vote was unanimous. The Fed described an economy expanding at a solid pace, supported by resilient spending and robust capital investment. Inflation remained elevated.

The decision depended on a familiar challenge: judging where the economy is headed with information that is incomplete at the time. Some indicators arrive quickly but cover only part of the picture. Broader measures take longer to publish and may later be revised.

That is why the quality and timing of economic data matter for policy. A clearer view of current activity could help the Fed tell a lasting change from a temporary one.

What the Five Task Forces are Examining

In July, Fed Chair Kevin Warsh announced five task forces to review how the central bank conducts monetary policy. Each will examine current practices and provide findings to the Federal Open Market Committee.

1. Communications will review how the Fed explains its decisions and deliberations when the outlook is uncertain.

2. Balance sheet policy will examine the costs and benefits of the Fed’s current system for maintaining ample reserves, along with other ways to implement monetary policy.

3. Data sources will look for more accurate and timely signals about the economy, including new sources of information and changes to data gathering methods. This is the task force most relevant to Atlas: its work concerns what policymakers can know before official statistics provide a fuller account.

4. Productivity and jobs will assess how technologies such as AI affect output, investment, employment, and the economy’s capacity to grow.

5. Inflation frameworks will revisit how the Fed identifies and responds to the forces behind rising prices.

The task forces have not announced policy changes. They are examining the tools the Fed uses and the evidence behind its decisions.

Where Atlas Fits

Atlas Analytics studies what satellite imagery can reveal about economic activity as it happens. ROY analyzes changes in the built environment and land use to estimate economic activity. JACK uses computer vision to measure container activity at ports and inform trade estimates.

Those observations can add detail to a national figure. Strong domestic activity, for example, can coincide with a drag from trade. Growth in one region can mask weakness in another. Understanding the components and geography of a forecast can be as useful as the headline number.

Satellite imagery requires careful interpretation. A visible change at a site does not, by itself, establish its economic meaning. The resulting signals need to be modeled, compared with other evidence, and tested over time. The Fed has not endorsed Atlas or identified satellite imagery as a preferred source. Its data task force is asking a broader question: whether additional sources can give policymakers a more timely and reliable picture of the economy.

Learn More

The Question After the Rate Hike

The September increase reflected the Fed’s judgment that inflation remains too high while the economy continues to expand. Future decisions will require the same judgment with a changing and incomplete set of facts.

Official statistics remain the benchmark. The opportunity for Atlas is to provide additional, testable observations during the period before those statistics arrive.

For most of modern economic history, measuring the economy has meant asking people what happened.

Government agencies survey households and businesses. Companies report financial results. Customs authorities record shipments. Statistical agencies collect, clean, aggregate, revise, and eventually publish the data.

This system is extraordinarily valuable. It is also, almost by construction, backward-looking.

GDP is the canonical example. We spend enormous amounts of time debating what the economy is doing right nowusing statistics that describe activity weeks or months in the past. Even the first estimate of quarterly GDP arrives after the quarter has ended, and that estimate is subsequently revised as better information becomes available.

But something fundamental has changed.

Increasingly, we don’t have to ask the economy what happened.

We can observe it.

Satellites can see construction sites expand. Ships can be tracked entering and leaving ports. Aircraft movements reveal changes in travel and commerce. Nighttime lights provide information about economic activity. Vegetation indices tell us about agricultural conditions. Digital exhaust—from searches, transactions, mobility, and other sources—provides another layer of information about what households and businesses are doing.

The interesting question is no longer whether these datasets contain economic information.

They do.

The harder question is how to turn millions of noisy physical observations into economically meaningful signals.

From economic statistics to economic sensing

Traditional economic measurement generally starts with a conceptual framework.

Define consumption. Define investment. Define exports. Then construct systems for measuring those quantities.

Alternative data often reverses that process.

You begin with something observable: a parking lot, a container terminal, a factory roof, a crop field, a ship.

Then you ask: what economic variable does this observation tell us something about?

That distinction sounds subtle, but I think it represents a meaningful change in how economists can approach measurement.

Consider a port.

Traditional statistics might eventually tell us the dollar value of goods imported through it. But before those statistics are released, satellites and other sensors can observe vessels, containers, vehicles, storage areas, and physical changes in the facility.

None of those observations is imports.

But collectively they contain information about imports.

The job of the economist becomes translating physical activity into an economic concept.

That is economic sensing.

Satellite imagery isn’t magic

There is an understandable temptation to think that sufficiently sophisticated AI applied to satellite imagery can simply “see GDP.”

It can’t.

A satellite observes reflected electromagnetic radiation. GDP is an accounting construct.

There is an enormous modeling problem between the two.

In fact, one of the biggest lessons I’ve learned from working with geospatial data is that the economic model matters at least as much as the imagery.

Suppose a satellite observes significantly more activity around a manufacturing facility.

Is production increasing?

Maybe.

Or perhaps inventory is accumulating because demand has collapsed. Perhaps construction is occurring. Perhaps the imagery reflects a seasonal pattern. Perhaps atmospheric conditions changed the measurement.

Remote sensing provides observations. Economics provides interpretation.

That is why I suspect the most valuable applications of alternative data will come not from replacing economists with computer vision models, but from combining domain knowledge, econometrics, machine learning, and increasingly powerful observation systems.

The evidence is accumulating

This isn’t merely theoretical.

Researchers have been using nighttime lights as proxies for economic activity for years. More recently, improvements in satellite coverage, computing power, machine learning, and data accessibility have made much richer applications possible.

An IMF working paper published this year, for example, examined satellite measures including vegetation, nighttime and daytime lights, and nitrogen dioxide alongside machine-learning techniques for estimating Venezuelan GDP. The authors found that incorporating satellite information improved the performance of their random-forest model.

Another IMF study applied satellite indicators, including nighttime lights, NO₂ emissions, precipitation, and vegetation measures, to GDP nowcasting in Cambodia, where timely conventional economic statistics are more limited.

And research published this summer on Morocco combined nontraditional information including satellite imagery, Google Trends, and flight-tracking data. The enriched models reduced out-of-sample errors relative to conventional benchmarks across measures of agricultural activity, tourism revenues, and unemployment.

These applications are particularly powerful in countries where official statistics are delayed or incomplete.

But I don’t think their relevance ends there.

The real opportunity is granularity

The most interesting advantage of alternative data may not actually be speed.

It may be resolution.

GDP tells us what happened to the United States.

But investors, companies, and policymakers frequently care about much smaller units.

What is happening around a particular logistics hub?

Is construction activity accelerating in one metropolitan area but slowing in another?

Are factories in one industrial corridor behaving differently from the national manufacturing sector?

Is activity increasing around a particular asset?

Traditional economic statistics become progressively thinner as you move from country to state to county to company to individual physical asset.

Satellite imagery moves in almost exactly the opposite direction.

The Earth can increasingly be observed at extremely high spatial and temporal resolution.

That creates the possibility of building economic indicators from the ground up rather than merely disaggregating national statistics from the top down.

Official statistics aren’t going away

None of this means GDP statistics, surveys, or government statistical agencies become obsolete.

Quite the opposite.

Alternative data needs authoritative benchmarks.

A satellite-derived indicator may tell us that activity appears to be accelerating. Official statistics ultimately tell us what that activity meant economically.

The relationship is therefore complementary.

Traditional statistics provide carefully constructed measurements of the economy.

Alternative data provides another set of sensors capable of detecting changes before, or below the level at which, those measurements become available.

Put them together and economic measurement potentially becomes faster, more granular, and more resilient.

Economics is becoming an observational science

Astronomers don’t survey stars about their temperature.

Meteorologists don’t ask clouds whether it is raining.

They build instruments that observe the physical world and models that translate those observations into useful information.

Economics is obviously different. Human behavior cannot be reduced to remote sensing, and many of the things economists care about, expectations, prices, wages, welfare, institutions, cannot simply be photographed from space.

But a surprisingly large portion of economic activity leaves a physical footprint.

Factories produce emissions.

Consumers drive to stores.

Developers pour concrete.

Farmers grow crops.

Ships move goods.

Cities consume electricity.

Aircraft carry passengers.

Mines move earth.

For the first time, we are developing the technological infrastructure to observe much of that activity continuously and at enormous scale.

The next generation of economic data may therefore look very different from the last one.

It won’t eliminate surveys, national accounts, or traditional econometrics.

It will add something economists historically haven’t had:

the ability to watch parts of the economy happening in real time.

Every quarter, economists, investors, policymakers, and journalists wait for the same number: U.S. GDP growth.

Within minutes of its release, the number is dissected. Did the economy beat expectations? Is growth accelerating or slowing? What does it mean for the Federal Reserve? Markets can move, forecasts are updated, and narratives about the economy begin to take shape.

But there is an important caveat: the first GDP estimate is not the final word on how fast the economy grew.

It can’t be.

Measuring an economy as large and complex as the United States in something close to real time is an extraordinary statistical undertaking. When the Bureau of Economic Analysis (BEA) publishes its initial estimate, some of the information needed to calculate GDP is incomplete or not yet available. As more comprehensive data arrive, the BEA updates its estimates.

Those revisions can be meaningful.

As the chart above shows, BEA estimates are revised by an average of approximately 0.3 percentage points in the first revision, 0.7 percentage points in the second, and 1.2 percentage points over a five-year period.

For an economy whose long-run growth rate is measured in just a few percentage points per year, those aren’t trivial differences.

Imagine an initial estimate showing the economy growing at a 2.0% annualized rate. A revision of even several tenths of a percentage point can meaningfully change our assessment of the quarter. Around economic turning points, revisions can be even more consequential for understanding whether the economy was accelerating, stagnating, or contracting.

Why GDP gets revised

Revisions aren’t a flaw in the system. They reflect a fundamental tradeoff in economic measurement: speed versus completeness.

If we wanted to know exactly what happened in the economy, we could wait until nearly every business survey, tax record, trade report, construction estimate, and other underlying dataset was complete.

But a GDP estimate arriving years after the fact wouldn’t be particularly useful to a policymaker deciding what to do today or an investor deciding where to allocate capital.

So statistical agencies produce the best estimates they can using the information available at the time. As better and more complete information arrives, those estimates improve.

That raises an interesting question: Can we expand the information available to us earlier?

A New Economic Information Set

For most of modern economic history, measuring the economy has depended heavily on businesses, households, and governments reporting what happened.

Today, we have another possibility: observing portions of economic activity directly.

Satellite imagery can reveal changes in construction, industrial facilities, transportation networks, ports, and other physical infrastructure. Geospatial datasets can help identify where economic activity is occurring. Shipping and logistics data can provide signals about the movement of goods. Other high-frequency datasets can capture activity long before traditional statistics are finalized.

None of these sources replaces official economic statistics.

A satellite cannot tell us how much households spent on healthcare last quarter. It cannot directly measure wages, software investment, or the enormous service economy. And alternative datasets introduce their own measurement challenges, biases, and noise.

But they can provide independent signals about parts of the economy while the traditional data picture is still incomplete.

That distinction matters.

Learn More

Forecasting the Present

We normally think of economic forecasting as trying to answer a question about the future: What will the economy look like next quarter or next year?

But there is another problem that is arguably just as important:

What is happening in the economy right now?

Economists call this nowcasting.

The difficulty of nowcasting is easy to underestimate. By the time we learn the official GDP estimate for a quarter, that quarter has already ended. And, as the history of revisions demonstrates, our understanding of what happened can continue changing long afterward.

This means that nowcasting is not simply a forecasting problem. It is also a measurement problem.

The opportunity presented by new forms of data is therefore not to replace the national accounts. It is to complement them: to add information to the economic picture sooner.

At Atlas Analytics, that’s the problem we’re working on. We combine satellite imagery, geospatial data, and traditional economic information to measure economic activity before the official picture is complete.

Better real-time measurement won’t eliminate uncertainty. No dataset can.

But if we can observe more of the economy as it happens, we can make better estimates of the present, and potentially make better decisions about the future.

Work With Us

Satellite-based macroeconomic forecasting isn’t just about GDP.

Atlas uses proprietary machine-learning and satellite-imagery models to measure economic activity from above. We are currently taking on a select number of projects to explore how these capabilities can be applied to specific industries, assets, and investment questions.

Have a use case where better real-time economic visibility could matter? Reach out to connect.

In May, Atlas wrote that the dollar was primed for a breakout, arguing that persistent inflation would keep U.S. interest rates higher than markets expected and, in turn, support a stronger dollar.

The Federal Reserve hasn’t raised rates yet. The dollar has moved anyway.

The Invesco DB U.S. Dollar Index Bullish Fund (UUP), our preferred proxy for the dollar, is now up approximately 4% year-to-date.

And following Federal Reserve Chairman Kevin Warsh’s hawkish Jackson Hole speech last week, the case for higher U.S. interest rates, and a stronger dollar. has only strengthened.

The question now isn’t whether the May thesis has begun to play out. It’s whether the move has further to run.

We think it does.

Why Do Higher Interest Rates Likely Mean a Stronger Dollar?

At its core, the relationship between interest rates and currencies is about where investors can earn the highest return on their money.

When U.S. interest rates rise relative to those in other major economies, dollar-denominated assets become more attractive. An investor choosing between otherwise comparable U.S. and European assets, for example, has a greater incentive to hold the U.S. asset when it offers a higher yield.

But to buy U.S. Treasuries, corporate bonds, or other dollar-denominated assets, investors generally need dollars.

That creates additional demand for the currency.

The key, therefore, isn’t simply whether the Federal Reserve raises interest rates. It’s how U.S. rates evolve relative to rates elsewhere, and how investors expect that gap to change.

This is why expectations matter so much. Currency markets don’t wait for the Federal Reserve to actually change the federal funds rate. If investors become convinced that U.S. rates will remain higher for longer, or rise while other central banks hold steady or cut, the expected return from holding dollar-denominated assets increases today.

That’s precisely the dynamic we’re watching now.

The Fed has yet to raise its benchmark rate, but expectations have shifted toward tighter U.S. monetary policy. If Chairman Warsh follows through on the hawkish message delivered at Jackson Hole, widening interest-rate differentials could provide another source of support for the dollar.

Why Should Warsh Hike?

Two weeks ago, we argued that the bond market is already sending the Federal Reserve a message: monetary policy may not be tight enough.

Despite the Fed holding its benchmark rate at 3.50%–3.75% at its July meeting, long-term borrowing costs have continued to rise. The 10-year Treasury yield recently climbed to roughly 4.7%, its highest level since January 2025. Mortgage rates have risen, corporate borrowing has become more expensive, and financial conditions have tightened even without another move from the Fed.

Why?

Part of the answer is inflation. Core PCE inflation remains above the Federal Reserve’s 2% target, while renewed energy-price pressures threaten to keep inflation elevated. Investors increasingly appear to be pricing the risk that inflation will remain persistent, and that today’s policy rate may ultimately prove insufficient to bring it back to target.

But inflation isn’t the only concern. Large federal deficits have increased the supply of Treasury securities just as foreign investors account for a smaller share of the market. At the same time, investors are demanding greater compensation for the uncertainty associated with holding long-term government debt.

The result is an important divergence: the Fed has paused, but the bond market hasn’t.

As we wrote in The Bond Market Is Telling You the Fed Should Raise Rates, Treasury markets continuously incorporate expectations about inflation, growth, fiscal policy, and monetary policy. Rising long-term yields therefore suggest that investors are demanding tighter financial conditions than they were only a few months ago.

Chairman Warsh now faces a choice. He can wait for inflation to return convincingly toward 2%, or he can respond to a market that is already signaling that rates may need to remain higher, and perhaps move higher still.

We think the balance of risks favors another hike.

What Could Make Us Wrong

No macroeconomic forecast is certain. While we believe the balance of risks points toward higher interest rates and a stronger dollar, there are several developments that could undermine our thesis.

• First, economic growth could deteriorate faster than expected. Atlas’s satellite-based macroeconomic forecasts are already pointing toward a slowdown in U.S. growth. So far, however, we don’t believe that deterioration is severe enough to mandate a rate cut. If incoming data were to show a much sharper contraction, particularly alongside a substantial deterioration in the labor market, the Fed could be forced to prioritize growth over inflation. Rate cuts would substantially weaken our dollar thesis.

• Second, inflation could surprise meaningfully to the downside. August Core PCE will be an important test. A particularly soft reading could give the Fed greater confidence that inflation is returning toward its 2% target and reduce the case for another hike. We think that’s possible, but unlikely enough that it doesn’t change our base case today.

• Third, even a Fed hike may not be enough. Currency values depend on relative interest rates, not simply the direction of U.S. rates. If other major central banks tighten more aggressively than the Federal Reserve, the yield advantage of dollar-denominated assets could narrow, or even move against the United States. In that scenario, we could be right about the Fed and still be wrong about the dollar.

• Finally, politics and Treasury policy could complicate the signal coming from bond markets. The administration has sought to influence longer-term borrowing costs through changes in Treasury issuance and other policy levers. So far, those efforts have not been sufficient to prevent long-term yields from remaining elevated. But a more successful effort to push down the long end of the Treasury curve could weaken one of the pillars supporting our outlook.

For now, none of these risks is enough to overturn our base case. We continue to expect tighter U.S. monetary policy, and thus believe the dollar has further room to run.

Work With Us

Satellite-based macroeconomic forecasting isn’t just about GDP.

Atlas uses proprietary machine-learning and satellite-imagery models to measure economic activity from above. We are currently taking on a select number of projects to explore how these capabilities can be applied to specific industries, assets, and investment questions.

Have a use case where better real-time economic visibility could matter? Reach out to connect.

Over the past several years, the AI investment story has largely centered on chips, models, and technology stocks. Today, the scale of the AI buildout is shifting attention toward the physical infrastructure required to support and power it.

Hundreds of billions of dollars are flowing into data center buildouts, land acquisition, servers, power generation, transmission, cooling systems, legal, and construction. As that capital moves into the physical economy, AI is beginning to reshape not only technology markets, but energy demand, infrastructure development, and regional economies across the United States.

The gap between what is announced and what is actually built may therefore become an increasingly important signal for understanding the next phase of AI investment.

ICYMI: Atlas AI Bubble Recap

Follow the Capital

The spending numbers alone illustrate how quickly the AI economy is changing.

Amazon increased its expected 2026 capital spending to approximately $220 billion following strong second-quarter cloud growth. AWS revenue increased 37% year over year, its fastest growth in more than four years, even as Amazon said demand continued to exceed available computing capacity.

Microsoft expects to invest roughly $190 billion in capital expenditures during calendar year 2026. The company has said that much of its recent spending is going toward GPUs, CPUs, and long-lived assets such as large data center sites. Microsoft also expects to remain capacity constrained as it continues expanding AI infrastructure.

Alphabet has guided toward $175 billion to $185 billion of 2026 capital expenditures, primarily to expand AI computing capacity and meet growing Google Cloud demand.

Meta expects another $130 billion to $145 billion of capital expenditures in 2026. In the second quarter alone, Meta spent approximately $31 billion, nearly double the level of the same quarter a year earlier.

Taken together, these four companies alone are planning more than $700 billion of capital spending in 2026, with AI and cloud infrastructure among the major drivers.

That is no longer simply a technology investment cycle.

It is becoming one of the largest physical capital deployment cycles in the U.S. economy.

The Investment Is Showing Up on the Ground

Recent announcements make the scale even more tangible.

In Kentucky, NextEra Energy announced plans for a more than $100 billion privately funded development at the U.S. Department of Energy’s Paducah Site in Kentucky.

Kentucky is not an isolated case.

In Ohio, Nvidia has agreed to provide guarantees of up to $105 billion supporting OpenAI’s long-term lease of a data center development that could ultimately reach 8 gigawatts of capacity. The project’s first 800 megawatts are expected online in 2028.

In West Virginia, Anthropic is expected to commit $45 billion over six years to rent computing capacity from an Nscale data center, representing roughly 460 megawatts of capacity.

In Texas, Meta is also bringing outside infrastructure capital into the buildout. In July, the company announced a partnership with BlackRock to finance a roughly $14 billion data center campus. Funds managed by BlackRock will own 80% of the venture, with Meta retaining 20%.

These projects represent different types of financial commitments, but together they point to the same structural shift. AI investment is becoming geographically specific, asset intensive, and increasingly dependent on physical infrastructure.

Corporate Balance Sheets to Capital Markets

The financing is changing too.

For much of the AI boom, the largest technology companies could finance infrastructure largely from their enormous operating cash flows.As investment requirements have grown, outside capital and debt markets have become increasingly important.

According to Reuters, U.S. corporate debt issuance associated with AI infrastructure reached approximately $220 billion in 2026, compared with just $12.5 billion the previous year.

Private capital is moving in as well.

This matters because the AI buildout is beginning to resemble other major infrastructure cycles. As the sector becomes more capital intensive, the relationship between capital committed and productive infrastructure delivered will matter more.

The Tale of Two Buildouts

This creates a useful distinction for investors.

There is the announced AI buildout. It consists of capital expenditure guidance, financing commitments, data center announcements, power agreements, equipment orders, and projected capacity.

Then there is the physical AI buildout. Land gets cleared. Foundations appear. Buildings go vertical. Cooling systems and electrical infrastructure are installed. New generation comes online. Facilities expand, become operational, or encounter delays.

Over time, the two should converge.

If capital in planned investment translate into completed assets with strong utilization and attractive returns, the physical economy will provide evidence that the infrastructure cycle is delivering on its promise.

If deployment begins slowing, construction timelines slip, facilities remain underutilized, or announced capacity repeatedly fails to materialize, the physical economy may provide an early indication that expectations have moved ahead of reality.

That makes physical deployment an increasingly important piece of the AI investment puzzle.

Measuring What Actually Gets Built

At Atlas Analytics, our technology is designed to measure economic activity through changes that can be observed in the physical world.

Using satellite imagery, computer vision, AI/ML, and time-series analysis, Atlas can monitor activity across individual locations and assets over time.

For the AI infrastructure cycle, that creates an opportunity to track activity around specific data centers, construction projects, energy infrastructure, industrial facilities, and other economically significant assets.

Financial markets provide one view of the AI boom. Corporate announcements provide another.

Physical activity provides an independent layer of evidence.

As AI investment becomes increasingly tied to land, energy, construction, and individual facilities, understanding what is actually happening at those locations can help distinguish between capital announced and capital deployed.

The AI boom is increasingly leaving a physical footprint across the United States.

And that footprint can be measured.

Learn More on Atlas Asset Signals.

Recent movements in U.S. Treasury markets present a puzzle. At its July meeting, the Federal Open Market Committee left the federal funds rate unchanged at 3.50%–3.75% for a second consecutive meeting, despite three officials dissenting in favor of an immediate increase1.

Since then, long-term borrowing costs have continued to rise. The yield on the 10-year U.S. Treasury has climbed to roughly 4.7%, its highest level since January 20252. Thirty-year mortgage rates have increased to 6.65%, corporate borrowing has become more expensive, infrastructure projects face higher financing costs, and higher discount rates have weighed on equity valuations3. Financial conditions have therefore continued to tighten despite no change in the Fed’s policy rate.

If the Fed has stopped raising interest rates, why have borrowing costs continued to increase?

One Rate, Many Yields

The Federal Reserve directly controls the federal funds rate, the overnight interest rate at which banks lend reserves to one another. Most borrowing decisions in the economy, however, are made over much longer horizons. Households borrow through thirty-year mortgages. Firms issue debt with maturities of five, ten, or even twenty years. Governments finance infrastructure whose lives extend over decades.

The interest rates attached to these decisions are not set directly by the Federal Reserve. They are determined in financial markets, where investors decide the return they require to lend over different horizons. U.S. Treasury securities play a central role in this process because they provide the benchmark against which many other borrowing costs are priced4.

This distinction is important. The federal funds rate reflects the Federal Reserve’s current policy stance. Treasury yields reflect investors’ expectations about the future. Every day, investors reassess inflation, economic growth, fiscal policy, and risk, and those expectations are incorporated into Treasury prices and yields.

This helps explain why long-term interest rates can continue to rise even when the Federal Reserve leaves its policy rate unchanged. The recent increase in Treasury yields is unlikely to have a single explanation. Instead, it reflects several forces that have influenced how investors value long-term government debt.

Four Explanations for Rising Yields

1️⃣ The first explanation is that investors may doubt the durability of the pause itself. Core PCE inflation rose to roughly 3.4% year over year in May before easing only modestly, to about 3.3%, in June, still well above the Federal Reserve’s 2% objective.

Fed Chair Kevin Warsh made this point directly at the July press conference, stating that “there’s only a target and it’s 2%,” and citing already-tight financial conditions, rather than confidence that inflation had been contained, as the rationale for holding rather than raising rates.

Markets responded by pricing in a higher probability of a hike later in the year. Under this interpretation, Treasury yields are not a reaction to today’s policy rate; they reflect expectations about the future policy path.

2️⃣ A second explanation is closely related but distinct: it concerns the persistence of inflation risk rather than the near-term policy path. Long-term bonds promise fixed payments many years into the future, and those payments lose real value if inflation proves more persistent than currently priced. Energy prices have re-emerged as a source of concern amid ongoing tensions in the Middle East, feeding directly into both core inflation readings and mortgage pricing5. Under this reading, investors are not necessarily forecasting another rate increase; they are demanding higher compensation for the risk that inflation does not return smoothly to target.

3️⃣ A third explanation is fiscal rather than monetary. The Committee for a Responsible Federal Budget noted in July that the 10-year yield had moved roughly 45 basis points above the level implied by Congressional Budget Office projections, and warned that yields sustained at this level could add an estimated $1.7 trillion to the national debt over the next decade.

The Treasury continues to issue substantial volumes of debt to finance persistent deficits, and, as in any market, a growing supply of securities requires sufficient demand to absorb it at unchanged prices.

That demand has become harder to secure. Foreign investors, who held nearly half of publicly traded federal debt in the early 2010s, now hold closer to 30%, even as debt outstanding has grown more than fourfold since 20086. A $16 billion Treasury auction in May 2025 drew unusually weak bidding, and, taken together with the S&P, Fitch, and Moody’s sovereign credit downgrades of the past fifteen years, has reinforced a narrative in which federal borrowing needs are beginning to outrun the market’s willingness to absorb them at prevailing yields7.

4️⃣ The fourth explanation is more structural: investors may simply require greater compensation for bearing long-horizon uncertainty. Economists refer to this as the term premium, defined as the portion of a bond’s yield not explained by the expected path of short-term rates, but instead by the additional risk of committing capital for ten or thirty years. The Federal Reserve Bank of New York’s ACM model, developed by economists Tobias Adrian, Richard Crump, and Emanuel Moench, decomposes Treasury yields into precisely these two components, and has repeatedly shown the term premium rising during periods of heightened fiscal and inflation uncertainty8.

A rising term premium requires neither an expectation of a further Federal Reserve rate increase nor a deterioration in current economic data. It reflects, more simply, a higher price placed on the unknown. These four explanations are not competing hypotheses so much as complementary ones.

Treasury yields represent the aggregated judgment of a large number of investors simultaneously reassessing inflation, growth, fiscal sustainability, and monetary policy. The resulting price reflects a distribution of probabilities, not a single certainty.

A Precedent: 2013 and the Return of the Bond Vigilantes

This pattern, in which yields move well ahead of, or independently of, the Federal Reserve’s own decisions, is not new.

During the 2013 episode known as the Taper Tantrum, the 10-year Treasury yield rose by 137 basis points between early May and early September, after then-Chair Ben Bernanke signaled only that the Federal Reserve might eventually slow its asset purchases, not that it intended to raise rates.

Emerging-market currencies absorbed much of the resulting adjustment: a basket of eighteen floating currencies depreciated by roughly 8.8% against the dollar over four months, even though tapering did not begin until December of that year and rates were not raised until December 20159. The market repriced on an expectation, not an action.

A comparable pattern has been visible more recently in what analysts have described as the return of the bond vigilantes, a term coined in the 1980s for investors who discipline government borrowing by demanding higher yields rather than waiting for policymakers to act.

In May 2025, the 30-year Treasury yield briefly exceeded 5% for the first time since 2007, coinciding with Moody’s downgrade of U.S. sovereign debt and growing concern that a pending tax bill could add several trillion dollars to projected deficits over the following decade. None of that repricing required action by the Federal Reserve. It required only a market reassessing how much risk it was prepared to absorb at prevailing prices.

The broader lesson is that markets rarely wait for official decisions before repricing economic risk. They act on the expected distribution of future outcomes, not merely on realized policy.

More on Predictive Yield Rates

Implications for Monetary Policy

This has direct implications for how monetary policy should be understood.

Central banks do not influence the economy solely through the federal funds rate. Monetary policy operates primarily by affecting financial conditions more broadly, and Treasury yields sit close to the center of that transmission mechanism. They influence mortgage rates, corporate financing costs, infrastructure investment, commercial real estate valuations, and equity pricing.

When long-term yields rise independently of the policy rate, financial conditions tighten even in the absence of further action by the Federal Reserve. The bond market can therefore perform part of the central bank’s tightening function on its own; symmetrically, a decline in yields can perform part of its easing function.

The relevant question is not only whether the Federal Reserve raises rates at its next meeting, but whether financial markets have already begun tightening conditions on the Federal Reserve’s behalf.

The bond market does not set monetary policy. It does, however, continuously assess whether current policy is consistent with the economy that investors expect to face in the years ahead.

The recent rise in Treasury yields is unlikely to reflect a single cause. It more plausibly reflects the combined influence of higher-for-longer policy expectations, persistent inflation risk, expanding Treasury supply against softening demand, and a higher premium for long-term uncertainty. Taken together, these forces indicate that investors are demanding tighter financial conditions than most expected only a few months ago.

The broader implication extends beyond the current episode. The Federal Reserve announces policy eight times a year. The bond market updates its assessment continuously. Central banks communicate their intentions through periodic policy statements, but Treasury markets aggregate dispersed information from a large number of investors into a single price in real time.

Recognizing this distinction helps explain not only why yields move, but why financial markets frequently begin adjusting well before official policy catches up.

Tell Us What You Think

This summer, those indicators moved in different directions.

According to the Bureau of Economic Analysis (BEA)’s advance estimate, the U.S. economy expanded at a 1.5% annualized rate during the second quarter. That headline was relatively soft, but underlying private domestic demand was stronger. Real final sales to private domestic purchasers, the sum of consumer spending and gross private fixed investment, increased 3.9%.

Then came the July jobs report.

The Bureau of Labor Statistics (BLS) estimated that nonfarm payroll employment declined by 23,000 in July. BLS described that change as “little changed,” not as a collapse in employment. The report nevertheless showed that hiring had become subdued. Payrolls increased by an average of only 34,000 per month over the prior 12 months, and the May and June estimates were revised down by a combined 103,000 jobs.

How can the economy continue producing more while payroll growth stalls?

Timing is part of the answer. GDP measures activity from April through June, while the latest employment report covers July. The soft July payroll estimate could reflect conditions that developed after the second quarter ended.

Productivity offers another possible explanation.

More Output Does Not Necessarily Require More Workers

The latest productivity report provides the clearest bridge between the GDP and employment data.

During the second quarter, nonfarm business output increased at a 1.7% annualized rate. Hours worked increased only 0.3%. The difference produced a 1.4% increase in labor productivity, which BLS defines as output per hour worked.

The year-over-year comparison is even more striking. Nonfarm business output increased 2.5%, while hours worked rose just 0.2%. Recent output growth has therefore required very little growth in total labor hours.

That does not mean artificial intelligence eliminated 23,000 jobs in July. The data cannot support that conclusion. Productivity can improve for many reasons, including new equipment, software, automation, changes in business processes, or more efficient use of existing workers.

The composition of second-quarter growth illustrates one possible mechanism. BEA reported that the increase in investment primarily reflected gains in equipment and intellectual property products, including software and research and development. These data do not prove that investment caused the weak hiring figures. They do show how capital investment can allow output to grow faster than payrolls.

The July payroll weakness was also concentrated rather than universal. Local government education lost an estimated 50,000 jobs, retail trade lost 19,000, and financial activities continued to trend down by 14,000. Health care added 22,000 positions. Employment in construction, manufacturing, transportation and warehousing, information, and most other major industries changed little.

That is not evidence of a broad-based employment contraction. It is evidence of a labor market that has lost momentum, with the latest weakness concentrated in several sectors.

The distribution of the gains also deserves attention. Labor’s share of nonfarm business output was 52.9% in the second quarter, the lowest level in a BLS series beginning in 1947. Price-adjusted hourly compensation declined at a 3.1% annualized rate during the quarter and was down 0.1% from a year earlier.

Together, these figures support a narrower conclusion. Output per hour increased, total hours barely grew, and the share of output accruing to workers as compensation reached a series low. They do not tell us whether the divergence will be temporary or structural.

Which Signal Should We Trust?

The answer is not to choose between GDP and employment. They measure three parts of the economy.

1. GDP measures output.

2. The payroll report estimates the number of jobs on employer payrolls.

3. Productivity measures output per hour worked.

Read together, the three reports describe an economy that continued to produce more in the second quarter without a comparable increase in labor input.

This is where measuring the physical economy can add another perspective.

Atlas Analytics’ algorithms, ROY and JACK, combine satellite-derived measures of the built environment with traditional economic data to forecast GDP. It provides an independent way to assess whether observable economic activity is expanding or slowing while official GDP and employment estimates are revised.

If physical activity remains firm while hiring weakens, the current divide may reflect a more productive or increasingly capital-intensive expansion. If physical activity and employment both begin declining, the July report may instead prove to have been an early warning of a broader slowdown.

Geography also plays a role.

A national payroll number can conceal meaningful differences between states and industries, just as national GDP can. Atlas produces economic forecasts for all 50 states and the District of Columbia, allowing us to examine whether growth remains broadly distributed or is becoming concentrated in fewer markets.

For now, the data do not establish either a recession or an AI-driven labor shock. They show that output grew in the second quarter, labor hours barely increased, and July payroll employment was essentially flat to slightly negative.

The next few months will show whether this is a temporary hiring pause or a more durable change in the relationship between output and work.

Get in Touch

For Atlas Analytics, our final forecast, published on July 13, more than two weeks before the release, projected:

• Headline GDP: 2.0%

• Core GDP: 3.2%

• Private Inventories: 0.3 percentage points

• Net Exports: -1.5 percentage points

The BEA’s Advance Estimate came in at:

• Headline GDP: 1.5%

• Core GDP: 3.2%

• Private Inventories: -0.7 percentage points

• Net Exports: -1.0 percentage points

While these comparisons are interesting, they raise a broader question that often gets overlooked:

What should we actually benchmark a GDP forecast against?

GDP Isn’t Measured Once

Unlike many economic indicators, GDP is not published once and left unchanged.

Instead, every quarter goes through a sequence of revisions:

• Advance Estimate

• Second Estimate

• Third Estimate

• Annual revisions

• Comprehensive benchmark revisions

The number reported on release day is simply the BEA’s best estimate using incomplete information. As additional surveys, trade statistics, inventory reports, tax records, and other source data become available, the estimate evolves.

In other words, even the government’s “official” GDP release is itself revised over time.

Why GDP Revisions Matter

This creates an important challenge when evaluating any GDP forecasting model.

Should a forecaster be judged against the BEA’s Advance Estimate? The Second Estimate? The Third Estimate? Or the benchmarked values published years later?

Each answers a slightly different question.

The Advance Estimate matters because it is the first number markets react to. But it is also produced using incomplete information. As additional source data become available, including more complete trade statistics, inventories, business surveys, and administrative records, the BEA updates its estimate to better reflect underlying economic activity.

Historically, these revisions have been far from trivial.

On average, the BEA revises quarterly GDP growth by approximately:

• 0.3 percentage points between the Advance and Second Estimates,

• 0.7 percentage points by the Third Estimate, and

• 1.2 percentage points following its five-year comprehensive benchmark revisions.

In other words, the “official” estimate released on GDP day is often only the beginning of the story.

Q2 2025 provides a good illustration. The BEA initially reported annualized GDP growth of 3.0%. One month later, the estimate was revised to 3.3%, and by the Third Estimate it had increased again to 3.8%, a cumulative upward revision of 0.8 percentage points.

This example highlights why evaluating a real-time forecasting model solely against the Advance Estimate can be misleading. The target itself continues to evolve as better information becomes available.

For that reason, we view every GDP release as the start of the evaluation process rather than the end. We’ll compare Atlas against the Advance Estimate because that’s what markets observe in real time, but we’ll also continue tracking the forecast as the BEA revises its estimate over the coming months and years.

What Atlas Is Actually Forecasting

Atlas was never designed simply to predict what the BEA will print.

Our objective is different.

Using satellite imagery and proprietary geospatial intelligence, we’re attempting to measure underlying economic activity directly, independently of the government statistical process.

That distinction matters.

The BEA builds GDP from thousands of surveys, administrative records, and statistical assumptions. Atlas begins with a completely different information set.

Sometimes those approaches converge almost perfectly.

Sometimes they don’t.

Understanding why they differ is often just as informative as the forecast itself.

Q2 2026: A Closer Look

One aspect of this quarter that stood out was the composition of growth.

Atlas matched the BEA’s estimate for Core GDP almost exactly. The primary differences came from the more volatile components of GDP: Private Inventories and Net Exports.

That result is perhaps not surprising. These are two of the most difficult components of GDP to measure in real time because they depend on complex physical supply chains that are only partially observed through traditional economic statistics.

Net Exports

Trade is fundamentally a logistics problem.

Every quarter, millions of containers move through U.S. ports carrying imported and exported goods. Traditional trade statistics are compiled from customs declarations and administrative reporting, meaning they become available only after considerable processing.

Atlas approaches this differently.

Using our Joint Algorithm for Containerized Knowledge (JACK), we identify and count individual shipping containers directly from commercial satellite imagery at major ports across the United States. Rather than waiting for reported trade statistics, we’re able to observe physical trade activity as it occurs.

By monitoring container flows over time, JACK provides an independent measure of import and export activity that feeds directly into our estimates of Net Exports.

Private Inventories

Inventories present a different challenge.

Unlike trade, there is no single observable location where inventories accumulate. Goods are dispersed across manufacturing facilities, warehouses, distribution centers, retail locations, and transportation networks throughout the country.

As a result, inventories remain one of the most difficult components of GDP to estimate in real time.

Atlas is actively developing new computer vision capabilities designed specifically to improve our measurement of inventory accumulation and depletion. While this work is still underway, we also closely monitor each month’s Census Bureau inventory releases to refine our estimates as additional information becomes available.

Over time, we believe combining satellite-derived observations with traditional inventory statistics will provide a more complete picture of inventory dynamics than either source can offer independently.

We’ll Continue Following the Story

One of the advantages of publishing forecasts in real time is that the evaluation doesn’t stop after release day.

Over the coming months, we’ll continue comparing Atlas against each successive BEA revision while also examining what drove any remaining differences.

Forecasting isn’t just about producing a single number: it’s about understanding the economy as new information becomes available.

As the Q2 data evolve, we’ll continue sharing what we learn.

Amazon announces a hyperscale data center. Regulators approve permits. Investors update forecasts. Utilities begin planning transmission upgrades.

But one question remains unanswered:

Is the construction actually happening?

How is the project actually developing once construction begins?

For much of the past century, this distinction mattered relatively little. Many infrastructure projects were local in both their benefits and their costs. Increasingly, however, large infrastructure investments interact with networks that extend well beyond the project site. Electricity, water, transport, and communications systems connect places that administrative boundaries separate. As a result, the economic consequences of infrastructure investments are not always experienced where the investment itself is located.

Data centers provide a useful illustration.

A proposed hyperscale data center in Alabama, for example, was projected to consume nearly two million gallons of water each day, approximately two-thirds of the daily water use of its host city1.

Electricity presents a similar challenge.

Under the traditional utility model, investments required to accommodate large new electricity demand are generally recovered through regulated rate structures. Whether existing approaches to cost allocation remain appropriate for exceptionally large electricity consumers, such as hyperscale data centers, has become an increasingly important regulatory question2.

The immediate economic benefits of a data center, including investment, employment, and local tax revenue, are concentrated where the facility is located. Some of the associated infrastructure costs, however, may extend across a much broader utility system.

The challenge is therefore not simply to understand these consequences after they occur. It is to observe infrastructure development while it is taking place.

This is where geospatial observation complements administrative information. A recent study by the Federation of American Scientists demonstrates how commercially available satellite imagery can independently track the physical construction of hyperscale data centers.

Comparing observable construction activity with publicly announced development timelines, the study shows that announcements and physical development do not always progress in parallel3.

The contribution of satellite imagery is not that it replaces administrative records or company disclosures. Instead, it provides an independent observation of physical development that can be compared with existing sources to verify how infrastructure projects evolve over time.

This distinction matters because many of the consequences of infrastructure investment emerge gradually. Electricity demand rises as facilities become operational. Supporting infrastructure expands as construction progresses. Administrative datasets, utility filings, and economic statistics eventually capture these developments, but often only after investment decisions have been made.

Geospatial observation cannot answer these questions on its own. Used alongside administrative records and economic analysis, however, it can reduce the interval between physical development and empirical observation.

Infrastructure Operates Through Networks

The reason is straightforward.

The economic effects of infrastructure are often assumed to be local because the infrastructure itself occupies a specific location. In practice, however, many infrastructure investments operate through interconnected systems whose boundaries differ from administrative jurisdictions.

Electricity is generated in one place, transmitted across regional networks, and priced through regulated utility systems. Water resources are shared across river basins and aquifers. Transport infrastructure influences economic activity well beyond the communities in which it is constructed.

Consequently, the geography of infrastructure investment does not necessarily coincide with the geography of its economic consequences.

Electricity illustrates this particularly well. Large data centers require substantial investments in generation, transmission, and distribution infrastructure before they become operational. Under traditional utility regulation, these investments are generally recovered through regulated rate structures that allocate costs across customers within a utility system. Whether existing regulatory frameworks remain appropriate for exceptionally large electricity consumers has therefore become an increasingly important policy question as electricity demand from hyperscale data centers continues to grow4.

The central issue is not that costs are necessarily shifted to other consumers, but that the economic consequences of new infrastructure depend on the structure of the utility network rather than the location of the facility alone.

Recent work by Lange, Bagwe, Murlidharan, and Rojas examines this question by mapping data-center development onto utility service territories rather than administrative boundaries. The study finds evidence consistent with residential electricity prices increasing in non-host counties connected through shared utility systems, while industrial electricity prices respond more strongly in host counties.

The authors emphasize that these estimates should be interpreted as evidence on electricity price incidence rather than as a full welfare analysis because the study does not directly observe utility rate cases, tariff provisions, or cost-allocation decisions. Their findings suggest that the geography of electricity price incidence may differ from the geography of infrastructure investment, reinforcing the importance of understanding the networks through which infrastructure operates.

The same principle extends beyond electricity prices. Muller estimates that the environmental damages associated with electricity consumption by U.S. data centers are geographically concentrated, shaped less by where a given facility is sited than by the emissions profile of the regional grid supplying it5. Although the outcomes are different, both studies point to the same conclusion. The consequences of major infrastructure projects frequently propagate through interconnected systems rather than remaining confined to the locations where projects are built.

This is precisely where geospatial intelligence becomes valuable. Satellite imagery identifies where physical infrastructure is emerging, while economic analysis explains how its consequences propagate through connected systems. Neither is sufficient on its own. Together, they provide policymakers with a more complete understanding of infrastructure development than either source could offer independently.

Learn More

From Observation to Decision

The examples discussed above point to a broader lesson.

Modern infrastructure projects generate information long before their full economic consequences appear in conventional statistics.

Administrative records document what has been approved. Satellite imagery observes what is physically happening on the ground. Economic analysis explains how those developments interact with electricity systems, supply chains, labor markets, or environmental outcomes. Each source answers a different question.

Viewed independently, each source also has clear limitations. Administrative records cannot verify whether announced projects are progressing as expected. Satellite imagery can observe physical development but cannot explain its economic significance. Economic indicators ultimately capture many of the consequences of infrastructure investment, but often only after substantial time has passed. Combining these sources provides a more complete picture than any one of them can offer individually.

This integrated approach becomes increasingly valuable as infrastructure projects grow larger and more interconnected. Whether the question concerns electricity systems, transport corridors, industrial facilities, logistics networks, or urban development, policymakers are increasingly required to make decisions before conventional datasets fully reflect changing economic conditions. Independent geospatial observation, interpreted alongside administrative information and economic analysis, can help reduce that information gap.

The value of geospatial intelligence therefore lies not simply in producing better images, but in improving economic understanding. Satellite imagery allows infrastructure to be observed as it develops. Economic analysis helps explain why those developments matter. Together, they provide policymakers with earlier, more comprehensive evidence for evaluating infrastructure investment and its broader consequences.

From Research to Practice

The questions discussed throughout this article are no longer purely academic.

At Atlas Analytics, we are in the early innings of applying geospatial intelligence to real-world infrastructure decisions alongside leading institutions in finance, consulting, and the public sector. Our work combines satellite imagery, administrative records, and economic modeling to help organizations monitor infrastructure development, understand its broader economic implications, and make better-informed decisions before conventional datasets fully reflect changing conditions.

We believe this is just the beginning. As satellite imagery becomes more frequent, computer vision continues to improve, and economic models become increasingly integrated with geospatial data, the opportunity to generate earlier, more actionable economic intelligence will continue to expand.

If your organization is evaluating major infrastructure investments, tracking industrial development, or exploring how geospatial intelligence can improve decision-making, we’d welcome the opportunity to start a conversation. We are actively partnering with a small number of organizations to develop new applications and would be delighted to explore how this approach could support your work.

Interested in collaborating? If your organization is exploring how geospatial intelligence can support infrastructure planning, economic analysis, or investment decision-making, we’d love to hear from you. Contact us at contact@atlasanalytics.com to discuss research collaborations, pilot projects, or bespoke geospatial intelligence solutions.

Let's Connect

Last quarter marked an important milestone for Atlas Analytics.

We publicly forecasted 2.0% annualized real GDP growth several weeks before the Bureau of Economic Analysis released its advance estimate. The government’s initial estimate also came in at 2.0%, and the final figure settled at 2.1% following subsequent revisions.

Atlas and the government ultimately arrived at nearly the same answer. The difference is that our forecast was available almost three months before the final estimate.

This is why we are building Atlas: to provide reliable economic intelligence while conditions are still unfolding, rather than months after the underlying activity has occurred.

So, what are we seeing for Q2?

The Q2 Forecast

Atlas Analytics is forecasting 1.95% annualized real GDP growth for the second quarter of 2026.

The headline appears soft, but the components tell a more balanced story.

Within our framework:

Core GDP + Net Exports + Private Inventories = Headline GDP

For the second quarter, Atlas estimates:

• Core GDP: +3.18 percentage points

• Private Inventories: +0.30 percentage points

• Net Exports: -1.52 percentage points

• Headline GDP: +1.95%

Core GDP and Private Inventories together contribute approximately 3.48 percentage points of growth. The difference between that figure and our headline forecast comes almost entirely from international trade.

Atlas estimates that Net Exports will subtract approximately 1.52 percentage points from second-quarter growth.

This composition is important. The forecast does not point to broad weakness across the domestic economy. Core activity remains healthy, and businesses continued rebuilding inventories. The primary source of weakness is the trade balance.

That distinction would be difficult to identify from the headline number alone.

What JACK Is Seeing

Over the past year, Atlas has been developing JACK, the Joint Algorithm for Containerized Knowledge.

As we discussed in our recent article, You Don’t Know JACK, JACK uses satellite imagery and artificial intelligence to monitor activity across major U.S. container ports.

The system identifies container ships and uses computer vision to count containers moving on and off vessels. These observations help Atlas estimate imports, exports, and Net Exports before the official trade statistics become available.

Historically, our models inferred trade activity from related economic indicators. JACK allows us to begin measuring much of that activity directly.

This quarter, the signal is clear. International trade represents the primary drag on second-quarter GDP growth.

There is also some encouraging news. Our data suggest that trade conditions improved modestly during June, indicating that the drag from Net Exports may have started to ease toward the end of the quarter.

That improvement does not offset the weakness recorded earlier in Q2, but it may provide an early indication of where conditions are heading as the third quarter begins.

GDP Is the Score. The Economy Is the Game.

People often ask whether Atlas is a GDP forecasting company. GDP is an important part of what we do, but it is not the end product.

GDP is our benchmark. Economic intelligence is our business.

We forecast GDP because it gives us an objective scorecard. Every quarter, our models are tested against one of the world’s most closely watched economic statistics. That comparison allows our clients and our team to evaluate where the system is performing well and where it needs to improve.

Government statistics eventually capture these changes, but usually after a delay. Atlas is building the infrastructure to observe them as they happen.

Ab Astris ad Terram et Mare

From the stars to the land to the sea.

Atlas signals observe the physical economy across land and sea, using satellite imagery and artificial intelligence to measure activity before traditional data becomes available.

At sea, JACK monitors import and export activity at major U.S. ports. On land, Atlas measures economic conditions across states, metropolitan areas, counties, and neighborhoods, providing earlier visibility into growth, market conditions, and valuations.

We are now extending that capability to individual assets, including manufacturing facilities, data centers, logistics hubs, commercial properties, and critical infrastructure. By measuring changes in production, utilization, construction, and activity, Atlas can support investment decisions, underwriting, asset monitoring, and operational planning with intelligence grounded in what is physically happening.

What Comes Next

The second quarter illustrates why the composition of growth matters.

A headline forecast of 1.95% might suggest that the entire economy is losing momentum. Our data point to a different conclusion. Domestic activity remains healthy, inventories are contributing to growth, and international trade accounts for nearly all of the weakness in the forecast.

We can make that distinction because Atlas is moving beyond traditional forecasting inputs and directly observing more of the physical economy.

Get in Touch

ICYMI

Last week, Our Founder and CEO Jake Schneider and Aditya Murlidharan, Economic Analytics Fellow, examined whether elevated market valuations extend beyond the technology sector. Their analysis found that the disconnect between equity prices and current economic activity has become increasingly broad-based.

Full Analysis Here

Over the next two trading days, technology stocks (led by the XLK and QQQ) fell another 5%.

By the end of Q2 2026, our estimated fair value for XLK had fallen to $145.39, while the ETF closed at $180.35. The valuation gap widened to 24%.

That should have been the story.

It was not.

Irrational Exuberance Extends Into the Broader Market

We applied our same fair-value framework to the remaining fifteen ETFs tracked by Atlas. We expected technology to stand apart. Instead, we found that elevated valuations were visible across much of the market. Technology remained the most expensive sector, but it was no longer an outlier.

One interpretation is that markets are pricing a future economy rather than the one that exists today. Investors may expect stronger productivity growth, higher future earnings, or a structural improvement in economic performance. The rapid diffusion of generative AI and the ease with which it is being adopted across industries have strengthened those expectations. If investors believe the gains will be economy-wide rather than confined to technology, higher valuations should naturally extend beyond technology and technology-adjacent firms. Our findings are consistent with that possibility.

Alan Greenspan, our founder Jake Schneider’s first boss, famously asked whether financial markets were exhibiting “irrational exuberance.” We do not use that phrase lightly. Nor do we argue that current valuations are necessarily irrational. Our claim is narrower. Relative to today’s level of economic activity, equity markets appear to be pricing a considerably stronger economy than the one reflected in current economic activity.

That raises a simple question:

How far have market prices moved ahead of the economy that exists today?

Our framework begins with an equally simple proposition. Over the long run, corporate revenues, profits, and cash flows are constrained by aggregate economic activity. Gross Domestic Product (GDP), despite its imperfections, remains the broadest measure of that activity. If GDP can be estimated accurately before the official releases, it becomes possible to compare market prices with the level of economic output they implicitly assume. That comparison forms the foundation of this publication.

What the Numbers Show

Table 1 compares our estimated fair values with market prices for seven widely followed ETFs at the end of Q2 2026.

The results are difficult to dismiss as an isolated technology phenomenon.

XLK and QQQ remain the most expensive, trading 24.0% and 19.3% above fair value respectively. That is unsurprising.

What is more surprising is everything else. SPY and DIA, the two most widely followed large-cap benchmarks, trade roughly ~16% above fair value. IWV, representing almost the entire investable US equity market, remains more than 11.1% above fair value. Even IWM, despite representing smaller companies that have largely stayed outside the enthusiasm surrounding the largest technology firms, trades above the level implied by current GDP.

Only one ETF stands apart. XLV, representing the health care sector, trades below our estimated fair value. That’s particularly interesting given that the majority of recent hiring from the Bureau of Labor Statistics’ monthly Jobs Reports have been accrued to this single sector.

Technology therefore appears to be the extreme case, but not the only case. The difference is increasingly one of magnitude rather than direction.

Get in Touch

Is This Just One Quarter?

One quarter proves very little.

Markets often move ahead of the economy and, at other times, fall behind it. Short-term deviations from fundamentals are therefore neither unusual nor especially informative. The more relevant question is whether the current valuation gap reflects a temporary episode or whether it forms part of a broader pattern.

To examine this, we extended the analysis beyond the latest quarter and estimated GDP-implied fair values across the historical sample for all sixteen ETFs using the current Atlas valuation models. Rather than asking whether markets appear expensive today, we ask a simpler question: how has the relationship between market prices and underlying economic activity evolved over time?

Chart 1 reports the results for XLK. We focus on technology because it provides a useful illustration of the broader exercise.

The chart shows that the relationship between market prices and GDP-implied fair values is far from constant. Through much of 2022 and early 2023, XLK traded below its GDP-implied fair value as markets adjusted to higher interest rates, slowing growth expectations, and a broad repricing of technology stocks. As macroeconomic conditions evolved, the gap gradually narrowed. By early 2024, market prices had moved above the level implied by realised economic activity and have, with few interruptions, remained there since.

This evolution is perhaps more informative than the latest observation itself. The valuation gap does not move in one direction. Periods of undervaluation are followed by periods of overvaluation, with prices repeatedly moving around their GDP-implied fair values as economic and financial conditions change. The current quarter is therefore better viewed as one point in a longer sequence than as an isolated result.

We repeated the same exercise for the remaining ETFs in our sample. While the timing and magnitude of the valuation gaps differ across sectors, the exercise reveals a similar feature across the broader market. The relationship between prices and underlying economic activity evolves over time, making it possible to distinguish temporary departures from more persistent divergences.

Two ETFs deserve equal attention for the opposite reason.

Health care (XLV) and small-cap equities (IWM) remain much closer to their GDP-implied values. Both alternate between modest overvaluation and modest undervaluation rather than drifting consistently in one direction.

That distinction matters.

If elevated valuations reflected only enthusiasm surrounding the largest technology companies, broad market benchmarks would not display the same pattern. Nor would the divergence persist across sectors with very different exposures to technological change.

Instead, the evidence suggests that optimism has become considerably more widespread.

Why GDP?

The framework rests on a simple proposition.

Corporate earnings do not grow independently from the economy forever. Household consumption, business investment, government expenditure and exports ultimately determine the revenues firms can generate. GDP is the broadest measure of that aggregate activity.

Importantly, no single macroeconomic variable explains equity prices completely, not even GDP. Interest rates matter. Credit conditions matter. Financial conditions matter. These variables influence valuation around a common anchor: the economy’s capacity to generate income.

That is why GDP sits at the center of our framework.

For the historical analysis presented above, we use final GDP after all official revisions. The exercise therefore evaluates the pricing relationship itself rather than the quality of any forecast.

Current-quarter estimates present a different challenge. Official GDP is released quarterly and revised twice afterwards. Markets therefore spend much of their time trading without knowing actual current economic conditions.

To address that gap, we combine the same valuation framework with Atlas’s live GDP nowcast, generated from proprietary satellite-based measures of economic activity.

The objective is straightforward:

Estimate today’s economy using today’s information rather than waiting for tomorrow’s official data.

More on Market Intelligence

What This Suggests

There are at least two plausible interpretations of these findings:

1. The Market Is Correct: The first is that markets are correctly looking beyond today’s economy. Expectations of stronger productivity growth, broader technological diffusion and higher future earnings may justify valuations that appear elevated relative to current GDP. Under this interpretation, prices are discounting future fundamentals rather than departing from them.

2. The Return of “Irrational Exuberance”: Markets may simply have moved ahead of themselves. Expectations have risen faster than realised economic activity, leaving valuations increasingly detached from the economy that ultimately supports corporate earnings.

Our analysis does not distinguish between these explanations.

It establishes something narrower, but also more measurable.

Key Takeaway

The gap between equity prices and current economic activity has become persistent, broad-based and larger over time. Technology remains the most expensive corner of the market, but it no longer appears exceptional. The same pattern is now visible across much of the market.

We do not know whether today’s valuations will ultimately prove justified. We do know that markets are pricing an economy considerably stronger than the one reflected in current GDP. Whether that gap reflects foresight or irrational exuberance is a question only time can answer.

Want to know our Atlas ETF estimates in-real time? Contact us to subscribe to an enterprise license and get the power of satellite-derived economic and equity estimates delivered through our proprietary portal to your inbox every week.

Get in Touch

What if investors could understand the economy as it unfolds, rather than months after the fact?

That was the central question Atlas Analytics, Founder & CEO, Jake Schneider explored on The Blunt Dollar, where he joined Ignacio Ramirez to discuss how satellite imagery, artificial intelligence, and macroeconomic modeling are changing the way economic activity can be measured.

For decades, markets have relied on official economic data that is delayed, revised, and often backward-looking. GDP is one of the most important measures of economic performance in the world, but by the time official releases arrive, investors, policymakers, and business leaders are often reacting to conditions that have already changed.

As Jake put it:

“GDP and other key economic indicators come out with a lag. Satellites continue overhead in real time.”

Get in Touch

Atlas Analytics algorithms were built around a different idea: the physical economy leaves visible traces in real-time.

Construction sites expand. Ports fill and empty. Land use changes. Containers move through supply chains. Factories, infrastructure, farmland, and transportation networks all generate observable signals that can be measured from above. At Atlas, those signals are transformed into real-time economic intelligence.

“We ingest satellite imagery from outer space and turn raw images of Earth into usable data for forecasting the economy in real time.”

The company’s core algorithm, ROY, short for Remote Orbital Yield, focuses on broad economic activity, including construction, land use, vegetation, and other physical signals tied to core GDP. By comparing decades of historical satellite imagery with official GDP data, Atlas has built models designed to identify changes in economic momentum as they happen.

One of the clearest examples came in Q2 2025. At the time, consensus expectations suggested the U.S. economy was slowing. Atlas saw something different. The company forecasted 4.3% GDP growth, well above consensus estimates near 2%. Months later, the Bureau of Economic Analysis revised GDP to 3.8%, much closer to Atlas’ original forecast.

“The consensus was 2%. We predicted 4.3%, and the final answer was 3.8% three months later.”

That time advantage matters. In markets, being directionally right is valuable. Being directionally right months early can be transformative.

The conversation also highlighted JACK, Atlas’ trade intelligence platform and second major algorithm. JACK, short for Joint Algorithm for Containerized Knowledge, uses satellite imagery over major U.S. ports to monitor container activity and estimate trade flows in near real time. By tracking TEUs, or twenty-foot equivalent units, JACK is designed to improve visibility into imports, exports, and net exports before official data is available.

A Global Forecast

Jake highlighted Atlas’ work with the International Monetary Fund and Johns Hopkins University to explore how satellite-based economic measurement can be applied beyond the United States, particularly in regions where official data can be delayed, incomplete, or released with long lags. In parts of Sub-Saharan Africa, for example, economic data can arrive up to a year late. Real-time measurement tools could help investors, policymakers, and institutions better understand economic conditions and allocate capital more effectively.

The conversation was also candid about what Atlas has learned. Jake discussed a recent quarter where Atlas’ headline GDP forecast diverged from the official release, particularly because of net exports. Rather than avoid the miss, Atlas has used it to sharpen its methodology and accelerate development of JACK.

“We are transparent about our wins and our losses — and that’s how you know you can trust us.”

That transparency is central to how Atlas thinks about trust.

The broader vision is simple but ambitious: move macroeconomic analysis from delayed reporting to real-time measurement. By observing the physical economy directly, Atlas is building tools that help investors and institutions see economic turning points sooner, understand regional and global activity more clearly, and make decisions with a faster, more grounded view of the world.

The economy moves in real time, economic intelligence can too with Atlas Analytics.

Let's Chat

On June 25th, the Bureau of Economic Analysis released its third estimate of U.S. GDP growth for Q1 2026: 2.1%.

For most economists, that’s the end of the story.

For us at Atlas Analytics, it was confirmation of something we already knew.

We published essentially the same forecast nearly three months ago.

The Problem with Economic Data

The U.S. economy moves in real time.

Economic statistics do not.

By the time official GDP is released:

• Investors have already repriced markets.

• Businesses have already made capital allocation decisions.

• Policymakers are already reacting to conditions that no longer exist.

GDP is arguably the single most important measure of economic activity in the world, yet it remains fundamentally backward-looking.

The advance estimate arrives roughly one month after the quarter ends.

The second estimate another month later.

The third estimate another month after that.

By the time the “final” number is published, nearly an entire quarter has passed.

In fast-moving markets, that’s an eternity.

Atlas Predicted It Months Earlier

Using Atlas Analytics’ ROY (Remote Orbital Yield) framework, which combines satellite imagery, geospatial intelligence, and economic modeling, we estimated first-quarter GDP long before the BEA released its official figures.

Our published estimate:

2.0%

BEA Third Estimate:

2.1%

A difference of just 0.1 percentage points.

That isn’t luck.

It’s the product of continuously measuring the real economy rather than waiting for surveys and delayed government reporting.

Real-Time Intelligence Changes Decisions

The goal isn’t simply to forecast GDP.

The goal is to understand the economy while it’s happening.

Because once you know where growth is today, you can begin answering much more valuable questions:

• Which industries are accelerating?

• Which regions are slowing?

• Which assets are becoming mispriced?

• Which companies are outperforming what investors currently believe?

Instead of reacting to history, decision-makers can respond to the present.

Learn More

Introducing the Atlas Portal

Today we’re also excited to share a preview of the next evolution of Atlas Analytics.

Our new client portal delivers our economic intelligence directly to customers through an interactive dashboard.

The executive overview provides continuously updated macroeconomic indicators, including Core GDP, inventories, net exports, and headline GDP, in a single interface.

Clients can also explore economic performance geographically through our new state-level forecasting dashboard.

Every U.S. state receives its own GDP forecast, allowing investors, corporations, and policymakers to identify regional divergence long before official statistics become available.

This is only the beginning.

Future releases will include:

• Historical forecasting revisions

• Weekly economic updates

• Asset-level intelligence

• Security-level insights

• Client-specific custom dashboards

Our objective is simple:

Move macroeconomic intelligence from quarterly reports to continuous decision support.

The Future of Economics

For decades, economists have accepted a simple constraint:

You cannot know the economy until after it has already happened.

Satellite imagery changes that.

Machine learning changes that.

Geospatial intelligence changes that.

The future of economics isn’t faster spreadsheets.

It’s observing economic activity directly.

Official statistics will always remain important.

But increasingly, they will become confirmation—not discovery.

At Atlas Analytics, that’s the future we’re building.

And this quarter, it arrived three months early.

Get in Touch

This week, the world lost a giant of economics.

As a young man, I had the extraordinary privilege of working directly for Dr. Alan Greenspan for nearly two years. Rather than write about his career, his legacy, or, quite frankly, his genius, I thought I’d simply share a few stories from our time together and the lessons they left with me.

Frequent readers of this Substack may know that I worked for Dr. Greenspan, but few know how I came to have the job.

Like many aspiring economists, I admired him from afar. One afternoon, while he was delivering a speech at the Brookings Institution, I waited in line afterward with a worn copy of The Age of Turbulence, hoping to have it signed. His Chief of Staff couldn’t arrange the signing that afternoon, but took my information.

A few weeks later, I arrived at his office carrying that same book. I assumed I would leave it with the receptionist for an autograph.

Before I could even introduce myself, the office doors opened.

Out walked Dr. Greenspan.

What I expected to be a thirty-second exchange became an hour-long conversation. We talked not about the econometric models or the Federal Reserve, but about what I considered one of the greatest accomplishments of his career: engineering the soft landing of 1994. I had studied that episode extensively and peppered him with questions about monetary policy, inflation expectations, and the delicate balance between acting too early and too late.

About a week later, my phone rang.

It was his Head of Research.

“Dr. Greenspan would like to know if you’d be interested in joining him as his research assistant.”

It remains one of the most unexpected and consequential phone calls of my life.

More important than how I was fortunate enough to land the job, however, was what I learned from him.

Search for today’s economy, not yesterday’s

Dr. Greenspan had enormous respect for official statistics, but he also understood their limitations. By the time GDP, employment, or inflation figures reached policymakers, the economy had often already changed.

He was constantly searching for signals that could tell him what was happening now, not what had happened three months ago.

That philosophy left an indelible mark on me.

Years later, it became the intellectual foundation for Atlas Analytics. Our work using satellite imagery to estimate GDP is, in many ways, an attempt to answer the same question he asked every day:

“What is the economy telling us today?”

Hold your convictions with humility

Despite being one of the most influential economists in history, Dr. Greenspan never pretended the economy could be known with certainty.

He held strong views, but lightly.

He understood that forecasting was not about proving yourself right. It was about continuously updating your beliefs as new information emerged.

Economics often rewards confidence.

Reality rewards intellectual honesty.

Never lose your sense of humor

The public knew Alan Greenspan as the famously reserved Chairman of the Federal Reserve.

I came to know someone with an unexpectedly dry wit.

One afternoon, while we were working on a research project together, I told him I’d make sure everything was “spick and span.”

Without looking up from his papers, he smiled and replied,

“I’m Greenspan. I’ll handle the span—you handle the first part.”

It was classic Greenspan: understated, clever, and perfectly timed.

On another occasion, during a trip to New York, the hotel pastry chef surprised Dr. Greenspan with an elaborate Star Wars-themed chocolate sculpture depicting him as Luke Skywalker. He admired it for a few moments, laughed, and then handed it to me.

That was him.

He never seemed particularly interested in the attention that came with being Alan Greenspan. To him, it was just chocolate.

When people think of Dr. Greenspan, they think of interest rates, inflation, and monetary policy.

I think of the young economist who nervously walked into an office carrying a book for an autograph and walked out with a mentor.

I think of the man who challenged me to look beyond the data everyone else was looking at.

I think of his quiet generosity, his relentless curiosity, and his wonderfully understated sense of humor.

Dr. Greenspan was a brilliant economist, but he was much more than that.

He represented an ideal that feels increasingly rare: intellectual curiosity, humility before uncertainty, respect for evidence, and a lifelong commitment to public service.

History will remember the Chairman.

I’ll remember Dr. Greenspan.

Rest in peace, Mr. Chairman.

We’re pleased to announce that Atlas Analytics’ economic intelligence is now available through Macrobond, one of the world’s leading platforms for macroeconomic research and analysis.

Macrobond is trusted by thousands of economists, investment professionals, central banks, government agencies, and research institutions around the world. By integrating Atlas’ data into the platform, institutional users can now access our real-time economic estimates alongside many of the world’s most widely used economic and financial datasets.

We’re also honored that Atlas Analytics was featured in Macrobond’s latest “Most Insightful Additions of the Past Two Weeks,” highlighting our satellite-derived GDP forecasts as one of the platform’s notable new data releases.

Learn More

Bringing Real-Time Economic Intelligence into Institutional Workflows

For decades, economists and investors have relied primarily on government statistical releases to understand the state of the economy. While these datasets remain indispensable, they are often published with meaningful delays and are subject to subsequent revisions.

Atlas Analytics takes a different approach.

Our proprietary Remote Orbital Yield (ROY) model combines satellite imagery with machine learning to estimate economic activity in near real time, providing a forward-looking view of the economy before many traditional indicators become available.

Through Macrobond, users can now incorporate Atlas’ economic intelligence directly into the research environment they already use every day.

Current Data Availability

The initial Macrobond release includes Atlas’ precision GDP forecasts for the United States, including several state-level economic forecasts such as:

• New York

• Illinois

• Arizona

These datasets provide timely, geographically granular measures of economic activity that complement traditional government statistics and may help investors identify changing economic conditions earlier.

More on State-level Forecasts

Looking Ahead

Our partnership with Macrobond represents an important step in Atlas’ mission to modernize macroeconomic measurement.

Over the coming months, we expect to expand our offerings with additional U.S. states, broader geographic coverage, and new datasets that leverage satellite imagery to measure economic activity, trade, and financial market intelligence.

We’re grateful to the Macrobond team for their partnership and excited to make Atlas’ economic intelligence available to a broader community of investors, economists, policymakers, and researchers.

As always, we’ll continue building the next generation of macroeconomic intelligence from outer space.

Learn More