Viewpoints: 2026

Artificial intelligence can lift productivity and sustain business investment. The scale and durability of the gains will depend on labour, energy, materials, finance and the policy response.

Introduction and summary messages The surge in artificial-intelligence investment is being driven by the technology's potential to raise productivity and corporate profitability. It is unfolding, however, in a world economy increasingly constrained by scarce labour, power, materials and capital. Those constraints make labour-saving technology more valuable, but they also limit how quickly the infrastructure supporting it can expand.

The initial macroeconomic effects can be favourable. If productivity rises faster than wages, unit labour costs fall. Firms can expand output without increasing employment at the same rate, supporting margins while containing underlying inflation. Stronger supply can therefore coexist with resilient demand and give monetary policy more room to accommodate growth.

The central risk comes later. Capacity and scarcity determine whether that favourable chain continues or reverses. Productivity gains can be diluted if electricity, grids, equipment, industrial materials or skilled labour become binding constraints. Inflation may then rise even before demand has weakened, leaving central banks to balance price stability against the investment needed to expand supply.

Backdrop Companies are buying computing power, software and equipment because they expect to produce more with fewer workers. Scarcity adds urgency to that calculation. Labour is expensive, trade is more fragmented and capital is no longer free. Governments are also investing in energy security, defence and domestic supply chains. Climate disruption and tighter immigration policies in some economies are adding to the pressure on energy systems, transport, construction and skilled labour.

Energy is where the tension is sharpest. Digital capital can be installed quickly; power stations, transmission lines and mines cannot. The most important price in the AI boom may therefore be the price of electricity rather than the price of a semiconductor.

The investment response is already visible. Orders for capital goods across the United States, Germany and Japan have risen for seventeen consecutive months, while Asian semiconductor exports tell the same story from the production side. Inflation-adjusted orders remain below their 2021 average, suggesting that the cycle is still relatively young.

The increase in investment is only the starting point. Its economic significance will depend on whether it produces measurable gains in output and productivity, how those gains are divided between labour and capital, and whether the expansion runs into physical or financial constraints. The framework below traces that transmission before the subsequent section considers what the available evidence suggests about how far it has progressed.

Since the introduction of ChatGPT in November of 2023, investments in artificial intelligence (AI) have surged. Economists await indisputable evidence of a positive return on those investments manifested in faster trend growth of productivity. Meanwhile, a popular – albeit “partial” - narrative is that without the AI buildout the economy would have been considerably softer over the past three years than observed.

We can agree that, in an accounting sense, the contribution to recent growth of GDP made by AI investments - defined here as fixed investment in computers & peripheral equipment, in software and in R&D, as well as construction of data centers - has become quite significant. My first chart shows the recent decomposition of 4-quarter growth in real GDP (not compounded). Since early 2024, the contribution to growth of AI investments has (roughly) quadrupled and now accounts for roughly 40% of total GDP growth. Yet during that span GDP growth itself has not moved noticeably higher.

Brent crude traded near $70 a barrel when the Strait of Hormuz closed at the end of February. It reached $138 in April and has since retraced part of that move while remaining well above its pre-disruption level. The macroeconomic response has so far been modest. Consensus forecasts for world output in 2026 have been revised down, but only very modestly.

The natural interpretation is that the structural sensitivity of output to energy prices has fallen, through lower energy intensity, more flexible product and labour markets, and better-anchored inflation expectations. There is something in each of these. But the evidence points to a more contingent explanation. The adjustment has been accommodated principally on the inventory margin rather than through quantity rationing or demand destruction, and the capacity to continue accommodating it that way is close to exhausted.

Implied global stock changes, derived from the EIA's world supply-demand balance, show a cumulative six-month drawdown roughly half as large again as the previous maximum in 2007. The initial condition matters here. Stocks were being accumulated rapidly through late 2025, so the disruption arrived with storage unusually full; the buffer was the incidental product of a preceding surplus rather than the result of deliberate precautionary accumulation. Commercial cover across the OECD has since fallen to around twenty-one days, and non-OECD buyers have begun to draw down the precautionary holdings built in the spring.

State labor markets were somewhat improved in August. Four states reported a statistically significant increase in jobs—California gained 39,400 and New Mexico Eight states saw statistically significant drops in their unemployment rate, though none was larger than .2 percentage points. The highest unemployment rates were in DC (5.7%), Connecticut (5.1%), Oregon (5.1%), California (5.1%), and Michigan (5.0%). Nebraska, New Hampshire, Hawaii, Vermont, North Dakota, and South Dakota had unemployment rates under 3.0%, while South Dakota’s 2.0% was the lowest in the nation.

Puerto Rico’s unemployment rate increased to 6.1%, reflecting a .7% increase in the island’s labor force (July saw a .8% rise), while the island’s job count was up 1,000.

10-year Treasury yields averaged 4.2% in the 12 months ending February 2026. They now hover near 5%, a significant increase. Besides higher oil prices, several explanations are plausible. Real rates have risen associated with the sustained healthy economic growth and higher expected rates of return on investment, particularly reflecting the booming AI buildout that is involving more debt financing. Inflationary expectations have remained generally well-anchored, but market-based measures have risen a few tenths of a percent. And the persistence of budget deficits and projections of ever-rising government debt-to-GDP may be exerting pressure on yields.

This note focuses on oil prices and yields. As a temporary negative supply shock to the economy, the higher oil prices are supposed to hurt economic activity which may be associated with lower real rates and are not supposed to impact intermediate and longer-run inflationary expectations. But Chart 1 shows convincingly that short-run movements in WTI oil spot prices (measured in 21-day changes in dollar-per barrel) and 10-year Treasuries (21-day percentage point changes) have been very closely correlated since end of February 2026 when the bombing in Iran commenced. Chart 2 shows that the levels of the Treasury yields and oil prices.

WTI rose dramatically from $65/barrel end-of February to the high-$90s end-of-March, bounced around and rose to $108 in May and plummeted to $70 in early July during the temporary ceasefire and then ratcheted up during the summer to its current above $102. Treasury bond yields rose and fell and rose again with these fluctuations.

At the recent Jackson Hole annual Kansas City barbecue, Fed Chairman Warsh mentioned that if the Fed had paid more attention to the surge in the M2 money supply in 2021, it might have raised the federal funds rate sooner, which, in turn, might have moderated the sharp increase in consumer price inflation that ensued. Wow! The ghost of Milton Friedman, may his memory be a blessing, must have been talking to Warsh while he was at Stanford’s Hoover Institution. The Fed could do worse than paying more attention to the behavior of the M2 money supply. It could do better, though, by paying more attention to behavior of the sum of the monetary base (reserves of depository institutions held at the Fed plus currency) plus securities and loans on the books of depository institutions (broad thin-air credit). But where Fed Chairman does not quite get it is his continued emphasis on using the federal funds interest rate as the Fed’s operating instrument. If the ghosts of Milton Friedman and Robert Laurent (also, may his memory be a blessing) would have had more séance time with Warsh, they might have been able to convince him to abandon using the federal funds interest rate as the Fed’s policy instrument and instead directly set the level of depository institution reserves such that a constant rate of growth in a monetary quantity, be it M2 or thin-air credit, could be achieved. This would require some institutional changes such as the elimination of interest payments on reserves and the re-introduction of reserve requirements. (See, my August 25, 2026 commentary “A Modest Monetary Policy Proposal for Fed Chairman Warsh to Consider” for a discussion of these necessary institutional changes.)

Let’s compare the relationship between percentage changes in the M2 money supply and thin-air credit with goods/services price inflation. Plotted in Chart 1 are the percent changes in the annual averages of the M2 money supply (the blue bars) vs. the percent changes in the annual averages of the Gross Domestic Purchases chain-price index (the red line) starting in 1960. The highest positive correlation between the two series occurs when percent changes in M2 are advanced (lead) by two years. The correlation coefficient is 0.48. Although a correlation coefficient of 0.48 is a long way from the maximum possible value of 1.00, it’s not bad for government work.

There is little to get excited about in the ASEAN-4 cyclical picture. Westbourne Research is underweight Indonesia across asset classes and has become distinctly more pessimistic about its cyclical and structural growth prospects. It is also underweight Malaysian equities and remains ambivalent about the Philippine and Thai stock markets. On sovereign bonds and currencies, positioning is either neutral or underweight.

Figure 1 shows the business-cycle assessment for the big four. As expected, economic activity slowed modestly in Indonesia and the Philippines in the second quarter. More concerning is Indonesia, where the corporate profit cycle has deteriorated sharply, moving from upswing to downturn, while both investment and credit remain in downswing. The positives are recovering broad-money growth and a falling real cost of capital, despite Bank Indonesia raising policy rates by 100bp since January 2026. In the Philippines, the only material change is the inflation signal, which has turned positive.

The latest round of inflation releases will help determine what central banks do next. A larger question lies behind the monthly numbers. Has inflation once again become more responsive to economic pressure? For much of the decade after the global financial crisis, falling unemployment generated surprisingly little inflation. That apparent flattening of the Phillips curve, the relationship between economic slack and inflation, encouraged policymakers to believe that economies could run hot at relatively little cost. Evidence since 2020 suggests that assumption is now less secure.

This argument complements, rather than overturns, the analysis in The age of abundance is over. That article argued that today’s inflation overshoot largely reflects scarcer labour, energy and productive capacity, none of which higher interest rates can directly replace. The evidence considered here makes a separate point. Even if supply constraints explain much of the current level of inflation, additional demand may now pass into prices more readily than it did in the 2010s. The source of inflation and its sensitivity to demand are related questions, but they are not the same question.

The evidence from the United States The US provides the clearest example. Comparing a real-time measure of economic tightness with inflation one year later reveals a marked break after 2020. Before then, a one-standard-deviation rise in the tightness measure was associated with an increase of about 0.4 percentage points in subsequent inflation. Since 2020, the estimated response has been roughly four times as large, at about 1.7 percentage points (Figure 1). Put simply, inflation now appears to react more strongly when demand presses against the economy’s capacity.