The Greenspan Lesson for the AI Race

Winning the AI race against China is going to take a gargantuan amount of patience. It will require U.S. corporations to borrow trillions of dollars to finance the massive physical capital investment required to outpace China. It will require data centers, power generation, electricity transmission, autonomous vehicle factories, and dozens of communication and computation satellites launched into low Earth orbit every single day.

Deft monetary policy plays a crucial role in U.S. economic security. In particular, the Federal Reserve should avoid stifling the productivity revolution needed for the AI race; official statistics register all that investment as an increase in input costs, rather than as a down payment on massive efficiency gains.

The country has faced a similar challenge before. In the early 1990s, official data had not yet caught up to the ongoing productivity boom. Standard models suggested that the economy was approaching its limits. A more cautious central bank could have concluded that the safe move was to choke off growth before inflation returned.

Alan Greenspan reached a different judgment.

Greenspan’s death this week will naturally revive the argument over his legacy: “the maestro” of the 1990s boom, or the central banker whose faith in markets helped set the stage for the financial crisis. That debate is deserved. But it should not obscure the lesson from his best moment in office.

Greenspan understood that macroeconomic data is, at best, a backward-looking instrument. It is useful— often indispensable—but it is also incomplete when the structure of production is changing faster than the statistical system can observe. Good central bankers need trendlines. They also need eyes.

That is the Greenspan lesson for the AI era, and the danger today is larger than a mistimed rate decision. U.S. macroeconomic institutions could misread a new productivity paradigm. As a result, they could raise interest, slow AI investment, and weaken the country’s long-term technological position.

The Economy Greenspan Saw

At the start of the 1990s, U.S. technological leadership was real, though less secure than memory suggests. Japan was the quintessential high-tech society, rapidly building competitive advantage in consumer electronics and industrial production. Europe had its own champions in telecommunications, Ericsson and Nokia, that seemed positioned to dominate the emerging mobile phone sector. The United States was powerful, dynamic, and innovative, but it was not yet the undisputed technology behemoth that it would become.

The domestic mood was grim. By 1992, the United States was emerging from Operation Desert Storm and the 1990–91 recession. The country was consumed by anxiety over fiscal deficits, trade deficits, corporate restructuring, and job losses. Only 10 percent of Americans described the economy as good or excellent, compared with 90 percent who called it fair or poor. Those were the worst numbers ever recorded outside of a recession.

Economists had a clean enough model for what the Fed was supposed to do: Lower rates when unemployment is high and inflation is falling, and raise rates when unemployment is low and inflation is rising. The Taylor rule gave that intuition mathematical form: Feed in measures of inflation and unemployment, and the formula spits out the prevailing interest rate.

The trouble is that inflation and unemployment do not always move in opposite directions, and official statistics do not always capture the economy’s productive capacity in real time. Monetary policy is full of words such as “pressure,” “slack,” “potential,” and “neutral” because the underlying objects are hard to see.

In 1992, the Taylor rule suggested that rates were about right. Greenspan looked out the window and saw something else. The economy was stuck in what became known as the jobless recovery. Businesses were timid. Households thought the economy was in the dumps. The official model said the economy was nearing full employment, but the lived economy looked far from full.

Even more important, Greenspan took seriously Bob Solow’s famous quip that computers were visible everywhere except in the productivity statistics. The point was simple: Computers had become an increasingly important part of the economy since the mid-1980s, yet productivity statistics were tepid at best and negative at worst. At the same time, standard inflation statistics showed prices rising 3–4 percent per year, even as electronics prices were falling steadily and product quality was improving rapidly. The statistics and the view from the window did not add up. Greenspan figured the fault lay at least partly with the statistics.

The Greenspan Fed continued to cut rates through 1992 and held them low through 1993. Unemployment began to fall in March 1992 and continued to decline. By the end of 1993, unemployment was approaching levels that economists considered sustainable. The profession largely believed that roughly 6 percent unemployment represented the economy’s natural resting point.

Greenspan began raising rates in 1994. Unemployment kept falling, crossing below 6 percent by September. The standard story suggested that inflation was coming unless the Fed moved sharply. Greenspan moved, yet he did so slowly and steadily. He resisted the urge to crush the expansion simply because unemployment fell below the level the models had treated as natural.

Then the economy began to do what Greenspan suspected it could. Productivity growth, which had barely averaged 1.5 percent per year from 1990 to 1995, rose above 2 percent in 1996 and kept climbing through the end of the decade. By 1999, productivity growth reached as high as 4.2 percent. Unemployment eventually fell to 3.9 percent, far below what economists had thought sustainable. The economy had more room to run because the supply side was improving faster than the official framework had assumed, thanks to the productivity enhancement that the official data did not fully capture.

The Boom That Became Strategic Infrastructure

While the hunger to invest in anything remotely related to the internet led to more than a few unproductive ventures—look, for example, to Pets.com—the ecosystem that the investment produced has proven enormously productive. 

First, note that despite both the dot-com crash and the recession following 9/11, labor productivity continued to grow above 2 percent through 2003. The average yearly increase from 1996 to 2003 was 3.15 percent. All in all, based on these numbers, the U.S. economy was 14 percent larger by 2003 than it would have been without the productivity surge. 

Second, the Magnificent Seven stocks—Alphabet (Google), Apple, Meta, Nvidia, Amazon, Microsoft, and Tesla—account for a third of the S&P 500’s market capitalization. Those seven companies alone exceed the combined market capitalization of the European Union and Japan.

The winners of the 1990s boom became the core of America’s AI advantage today. Microsoft and Amazon built the cloud platforms on which the AI economy runs. Google built the research culture that produced the transformer and remains one of the world’s leading AI labs. Nvidia supplied the chips that turned deep learning from a research program into an industrial race. Meta pushed open-source models into the market. Apple controls the devices through which much of this technology will eventually reach ordinary users.

Even the new AI labs grew up around that core. OpenAI scaled through Microsoft’s cloud, capital, and distribution. Anthropic built around the same world of frontier chips, cloud infrastructure, venture financing, and Bay Area talent. The dot-com boom created the operating system for U.S. AI dominance.

This dominance can be traced back to the investment boom enabled by Greenspan’s skepticism of official statistics and standards during a turning point in the economy. The United States now face a similar position. Rather than unemployment falling to record levels, we have stubbornly high inflation that seems stuck above the Fed’s 2 percent target. And the Fed seems biased toward raising rates. 

Looking Out the Window in Today’s Economy

Yet, just outside the window is an economy where AI is everywhere but in the productivity statistics. The entire software development industry, home to some of the world’s most dynamic companies, has undergone a radical change overnight. Programming is now done almost exclusively using AI. There is every reason to believe this rapid productivity growth will spread to nearly every sector of the economy.

Moreover, the cost of AI capabilities is dropping at a rate too rapid to measure. The best analysis suggests that, depending on the benchmark, the price of a given level of AI performance has fallen by ninefold to nine-hundredfold per year. It is implausible that our current statistics accurately account for these rapid changes. 

The pace of growth, however, will be influenced by borrowing costs. The AI economy will require trillions of dollars in data center investment. Until now, Silicon Valley has been able to mostly self-finance the AI buildout. That era will likely come to an end sometime this year. The cost of that financing will depend largely on where the Federal Reserve sets interest rates.

Greenspan’s legacy will be argued over for a long time, and it should be. But the question for the AI era is whether the Fed can recover its willingness, at a decisive moment, to look beyond backward-looking statistics and ask whether the economy’s productive frontier has moved.

Karl Smith is a senior fellow in the Economic Security and Technology Department at the Center for Strategic and International Studies in Washington, D.C.