The Bank for International Settlements (BIS), the Basel-based umbrella body for the world’s central banks, warned Tuesday that the artificial intelligence boom is making it harder for policymakers to read their own economies. Debt-fueled spending on chips and data centers is lifting growth, trade and stock prices well before any broad productivity payoff shows up, and that timing gap is scrambling the signals central banks use to set interest rates.
The confusion is not landing evenly. A handful of chip-exporting economies, led by Taiwan and South Korea, are already posting some of their best growth in decades, while most of the rest of the world absorbs the same blurred inflation picture without that offsetting windfall.
Two Signals, One Blurry Picture
AI is not behaving like an ordinary investment cycle, according to the BIS. It is pushing on both sides of the economy’s ledger at once, which is exactly what makes it hard to price.
By simultaneously affecting demand and supply, AI blurs cyclical signals.
The institution’s bulletin, published Tuesday, said that blur complicates how policymakers judge underlying economic conditions and calibrate monetary policy. Three effects are unfolding at once, on different timelines:
- Demand today: spending on chips, servers and data centers is already showing up in GDP, trade flows and corporate investment figures.
- Supply tomorrow: if AI eventually raises worker output and business capacity, it would expand the economy’s speed limit and ease price pressure, but only once the technology is deployed at scale.
- Wealth effects now: AI-linked optimism has driven rapid equity market gains, and that paper wealth is already feeding consumer spending.
Robust spending on data centers, chips and digital infrastructure can resemble an overheating economy, the BIS said, even when part of that spending reflects a genuine, longer-term increase in productive capacity. Read the other way, real productivity gains could mask demand pressure building underneath, making inflation trends harder to interpret in either direction. The size, timing and distribution of any eventual payoff remain highly uncertain, which is the crux of the problem for anyone setting a policy rate today.
Paying for the Boom With Borrowed Money
Part of why the signal is so hard to read is how this particular boom is being financed. Unlike most corporate investment waves, a growing share of AI infrastructure spending is not coming out of retained earnings. It is coming out of the bond market.
The BIS’s own annual economic report, published about a month before Tuesday’s bulletin, had already sized the exposure. The report found the five largest hyperscalers were on pace to spend more than $1 trillion combined on AI infrastructure across 2025 and 2026, a sum outpacing their combined earnings and free cash flow and pushing some of them to issue debt to cover the gap.
That gap has only widened since. Bond analysts tracked by Forbes this month put global AI-related debt issuance on pace to reach $570 billion in 2026, with coverage ratios on hyperscaler bonds sliding, a sign investors may soon demand steeper yields to keep buying. A separate BIS review of the sector’s financing structure flagged a related concern: a rising share of data center investment is routed through on- and off-balance-sheet borrowing that keeps leverage out of easy view, with private-credit estimates for the sector running as high as $800 billion.
Estimates of the total buildout vary widely. Morgan Stanley and Moody’s Ratings put data center capital spending at more than $3 trillion over the next several years, while JPMorgan’s estimate, which folds in the power plants needed to run it all, tops $5 trillion. Morgan Stanley separately expects hyperscalers alone to issue $250 billion to $300 billion in bonds this year.
The warnings have been stacking up for weeks:
- Late June 2026: The BIS’s annual report flags hyperscaler AI spending outpacing free cash flow.
- July 17, 2026: Forbes reports bond investors pushing back as AI-related debt issuance heads toward $570 billion.
- July 28, 2026: The BIS bulletin warns the debt-financed boom is blurring the inflation signals central banks rely on.
Each step added detail to the same worry: a lot of this buildout is leveraged, and leverage turns a forecasting error into a balance-sheet problem.
Four Economies, One Lopsided Boom
The BIS bulletin flagged something else worth separating out: AI’s economic effects are landing unevenly across countries, and that unevenness is its own policy headache. Economies that supply semiconductors, computing infrastructure or AI services are already growing faster, while others lag, producing different inflation and growth paths in different jurisdictions at the same time.
Nowhere is that split sharper than in East and Southeast Asia. Taiwan, South Korea, Malaysia and Thailand together are the world’s four largest net exporters of AI hardware, and their numbers are pulling away from the rest of the global economy.
| Economy | 2026 Signal | AI Hardware Role |
|---|---|---|
| Taiwan | Growth near 10%, fastest quarterly pace since 1987; current account surplus could top 20% of GDP | Produces about 90% of the world’s most advanced AI chips |
| South Korea | IMF raised its 2026 growth forecast to 2.6%, up from 1.9% in April; current account surplus could exceed 10% of GDP | Major exporter of memory chips and AI hardware |
| Malaysia and Thailand | Grouped with Korea and Taiwan in a first-quarter growth surprise averaging 4.4 percentage points above forecast | Round out the world’s four largest net AI hardware exporters |
| Rest of the world | First-quarter 2026 growth surprise averaged 0.3 percentage points below forecast | Largely outside the AI hardware supply chain |
That gap between the four exporters and everyone else, nearly five full percentage points in a single quarter, is the clearest evidence yet of what the BIS means when it warns that AI is producing different growth and inflation trajectories in different places at once.
The Rest of the World Gets the Static
That divergence is exactly the problem the BIS is describing. A central bank in a chip-exporting economy is looking at a genuine investment boom with real currency and wage effects. A central bank almost anywhere else is looking at the same AI-driven equity rally and global trade shift, minus the growth offset, which makes its own inflation reading murkier rather than clearer.
Even inside the winning economies, the gains are not landing on everyone. Al Jazeera reported in May that many Taiwanese households are not feeling the boom in their own paychecks, even as the island’s exporters post record numbers, pointing to a widening gap between the chipmakers driving the headline growth and workers outside that supply chain.
Haven’t Central Banks Faced This Fog Before?
Yes, and the last time it happened, the bet paid off. In the mid-1990s, Alan Greenspan’s Federal Reserve held its policy rate steady despite an unusually tight labor market, betting that unmeasured productivity gains from computers and networking equipment were quietly expanding the economy’s capacity. Inflation stayed low for years afterward.
Robert Solow, the economist who first flagged the mismatch, had put it plainly a decade earlier: you could see “the computer age everywhere but in the productivity statistics.” At the Fed’s September 1996 meeting, Greenspan convinced the rate-setting committee to hold its benchmark at 5.25 percent even as wages rose and unemployment fell, wagering that productivity was accelerating faster than official data showed. A Federal Reserve Bank of St. Louis review of that decade, published this month, credits the bet with helping deliver years of low, steady inflation alongside strong growth.
The difference this time is leverage. The 1990s technology wave rode mostly on corporate cash and stock financing. Today’s data center buildout leans on borrowed money, so a wrong rate call would not just mistime the cycle. It could squeeze the companies servicing that debt directly.
What a Wrong Rate Call Breaks First
The BIS stopped short of telling any central bank what to do. Its bulletin said policymakers need to disentangle temporary investment surges from lasting productivity improvements, or risk what it called “policy miscalibration.”
Its own numbers point to the costlier mistake: mistaking debt-financed capex for an overheating economy, then raising rates to cool it, would land directly on a bond market already showing strain, with hyperscaler coverage ratios sliding and investors demanding higher yields to keep buying. That is the mistake with the faster fuse, since it hits leveraged balance sheets before any productivity payoff has a chance to arrive.
The next test comes quickly. Hyperscalers report quarterly results in the coming weeks, and those numbers will show whether the gap between AI capital spending and free cash flow is narrowing or still widening, the same gap the BIS says central banks cannot yet read with confidence.
Frequently Asked Questions
What Is the Bank for International Settlements?
The BIS is an international financial institution based in Basel, Switzerland, often called the central bank for central banks. It is owned by dozens of member central banks worldwide and publishes regular research on financial stability and monetary policy, including the AI bulletin at the center of this warning.
Why Does AI Investment Raise Both Demand and Supply at Once?
Buying chips, building data centers and hiring engineers all show up immediately as spending, which lifts demand right away. The productivity gains AI is supposed to deliver, letting businesses produce more with the same workers, only materialize later, once the technology is actually deployed and adopted at scale, which is why the two effects arrive on separate timelines.
How Much AI-Related Debt Has Been Issued So Far?
Global AI-related debt issuance is on pace to reach $570 billion in 2026, according to bond market tracking reported this month, with hyperscalers expected to account for $250 billion to $300 billion of new bond sales this year alone, per Morgan Stanley estimates.
Could the AI Boom Cause a Recession If It Unwinds?
The BIS has not predicted a recession, but its own annual report acknowledged the current investment pace may not be sustainable, even as it credited AI with real long-term productivity potential. A sharp pullback in capital spending, especially one financed heavily by debt, would ripple through construction, chipmaking and credit markets at the same time.
Do Rising AI Stock Prices Affect Inflation?
Yes, through what economists call a wealth effect. When equity portfolios rise in value, households and businesses tend to spend more even without a change in income, and the BIS specifically named the AI-driven equity rally as a channel that can add to near-term demand and inflationary pressure, separate from direct spending on chips and data centers.
