The Federal Reserve held interest rates steady this week, but AI is making those decisions increasingly difficult. Massive investment in data centers and power infrastructure is boosting demand today, while AI’s promised productivity gains could ease inflation in the future. This leaves policymakers caught between two opposing forces.
With money pouring into data centers, advanced chips, power plants and transmission networks, accelerated funding is lifting investment, employment and demand today, which increases economic pressure on resources. The same technology may let workers and companies produce far more tomorrow, expanding economic capacity and easing price pressure, which can in turn reduce economic pressure. Those forces point in opposite directions for interest rates. One argues for tighter policy. The other supports lower borrowing costs.
The Bank for International Settlements, often called the central bank for central banks, has now put that contradiction near the center of the monetary policy debate. In a July 2026 bulletin, the BIS said the AI boom is shaping short term economic activity through a huge infrastructure buildout, yet the size and timing of the promised productivity dividend remain hard to measure.
That uncertainty turns AI into something larger than a technology investment cycle, and making it more of a monetary policy problem. Central banks must decide whether the current torrent of spending marks the start of a durable productivity expansion, an inflationary capital expenditure surge or a speculative bubble that could end with idle server farms and damaged balance sheets. They may not know the answer until after yesterday’s rate decisions have worked through the economy.
The AI Economy Runs On Copper, Concrete And Credit
The popular image of AI centers on software, algorithms and chatbots. But that’s just the digital, software side. Its physical footprint looks more like a real estate and industrial investment. Training and operating large AI models requires semiconductors, servers, cooling systems, warehouses, substations, electricity generation and thousands of miles of new grid capacity. Each component draws on scarce labor, commodities and manufacturing capacity.
The BIS estimates that the five largest hyperscale technology companies are set to spend more than $1 trillion on AI related capital projects during 2025 and 2026. Those commitments are growing faster than earnings and free cash flow at some firms, pushing parts of the sector toward debt financing.
Federal Reserve researchers have described a similar surge. One Fed analysis said spending on U.S. data centers alone was expected to exceed $500 billion in 2025, with announced projects pointing to years of construction and equipment purchases. However, much of the hardware comes from abroad, which means the gross investment figures can overstate the direct contribution to domestic economic growth. The spending still creates demand for land, engineers, electricians, energy equipment and financing.
This matters for rate setters. A conventional tightening cycle raises borrowing costs, slows construction and persuades executives to shelve projects that only offer marginal gains. AI investment may prove to be less responsive to rate changes. The companies funding the largest projects often hold immense cash reserves and view computing capacity as a strategic necessity. They may continue building through high interest rates rather than risk losing ground to a rival.
Demand Arrives Before Productivity
The problem of whether supply is outstripping demand is also part of the problem. Investment spending impacts the economy first, while productivity gains tend to arrive later, after companies install the technology, redesign workflows, train staff and discover uses that offer the best returns. History shows that general purpose technologies can take years to spread beyond the firms that created them.
AI may raise demand long before it raises supply, even as the hyperscale AI platform companies struggle to deal with day-to-day usage. Federal Reserve Governor Lisa Cook said in March 2026 that soaring investment in data centers and chips was already visible, despite borrowing costs that remained high compared with much of the previous two decades.
Her May 2026 speech on AI, productivity and inflation sharpened the concern. Companies had announced more than $1.5 trillion in data center plans, she said, yet only a small share of those projects had been completed. Prices had already risen for chips, advanced equipment and software. Wages were climbing in specialized construction trades. Strong investment can lift the economy’s neutral interest rate, the theoretical rate that neither stimulates nor restrains growth. A higher neutral rate would mean policy is not as tight as it appears.
Fed Vice Chair Philip Jefferson positions the inflation risk more directly. He said excitement about AI was contributing to a data center construction boom and that the immediate rise in demand could lift inflation before any increase in productive capacity appeared.
That lag leaves central bankers staring at conflicting evidence. Surging business investment may signal confidence in future growth. But it may also signal that too much money is chasing scarce chips, electricity and construction labor.
This poses a real challenge for regulators and policy makers. Raising rates too much could choke off investment that would expand supply and lower costs. Holding rates too low could feed construction inflation, financial leverage and inflated technology valuations.
A BIS working paper found that AI driven productivity growth could raise output, consumption and investment in both the near and distant future. Yet its effect on inflation was uncertain. Stronger supply can reduce price pressure, but rising income, investment and consumption can push demand higher at the same time.
Productivity statistics also offer only partial clues as to how AI is improving or impacting business profits. Companies can record stronger output per worker after companies cut low productivity jobs, run existing equipment harder or shift activity toward more efficient sectors. In many ways, those cuts and changes are not really directly related to their level of AI usage or whether those returns can be sustained.
A law firm using AI to draft contracts faster may not cut prices. It may take more cases or raise profit margins. A software company may use AI to release features with fewer engineers. A hospital may reduce paperwork without increasing the number of patients treated. Each case changes economic performance, but the gains may surface in different data series and on different schedules. This poses yet more challenges for central banks which cannot base policy solely on demonstrations of what AI can do. They need proof that the technology is lifting economy wide supply.
Looking To The Past To Predict The Future
The late 1990s dot-com boom and bust cycle offers a useful analogy. The internet changed commerce and communication, yet many of the companies financed during the dot com boom failed. The technology was transformative, but the era of “irrational exuberance” as former Federal Reserve chairman Alan Greenspan said in a televised speech, led to an investment cycle that overshot, contributing to a bust and decline.
AI can follow the same pattern without repeating it exactly. Central bankers must account for both futures: one in which AI lifts trend growth, and another in which excess capacity leaves lenders and investors nursing losses. AI has trained machines to detect patterns humans miss, but it is now scrambling the patterns central banks have spent decades learning to read.
