The Fed's AI Predicament: Learning from Greenspan is a "Dead End," Not Lowering Interest Rates is a "Dead End"

The current fervent narrative surrounding AI is pushing the Federal Reserve into an unsolvable dilemma: following Greenspan is a "dead end," while not lowering interest rates is a "desperate path."

On November 28th, according to news from the Chase Trading Desk, the globally renowned independent research institution TS Lombard stated in its latest research report that it is still uncertain whether AI will bring about a deflationary productivity boom like in the 1990s or whether it will push up the equilibrium interest rate (r*) due to massive capital expenditures. These two possibilities will lead to completely opposite monetary policy paths, which is the core dilemma currently faced by the Federal Reserve.

Following Greenspan's lead to lower interest rates is a "dead end": If the Federal Reserve lowers interest rates merely because of the expectation that AI can enhance productivity, emulating Greenspan in 1996, it would be extremely dangerous. This is because today's inflation environment is far less favorable than in the 1990s (when the core PCE inflation rate was trend-wise below 2%), and this approach completely ignores the historical lesson that Greenspan turned hawkish in 2000 for the same reasons.

If interest rates are not lowered, it is a "desperate path": The Federal Reserve may inadvertently push the market into a dead end. The real risk is that if inflation resurges in 2026 and becomes a primary concern, the Federal Reserve will be forced to adopt tightening policies. By then, even if they do not intend to burst the bubble, raising interest rates itself could become the last straw that breaks the camel's back.

What will the Federal Reserve do in the face of the asset bubble triggered by AI? The research report points out that Greenspan's famous strategy is "clean, don’t lean," meaning not to actively burst the bubble but to clean up the mess after it bursts. It is foreseeable that his successors, especially a chairperson who may be appointed by Trump and holds a pro-technology stance, will continue to adhere to this principle.

Greenspan's "Dual Legacy": A Complex Policy Template

The report states that currently, all potential candidates for the Federal Reserve chair are trying to package themselves as Greenspan's successors, claiming that the AI revolution is an excellent reason for lowering interest rates. They cite Greenspan's famous decision in the mid-1990s:

Despite the unemployment rate falling below the so-called "natural rate of unemployment" (NAIRU), Greenspan firmly believed that official data underestimated productivity growth and used this to persuade his hawkish colleagues to delay interest rate hikes.

The report points out that in the fall of 1996, Greenspan commissioned a study by Federal Reserve staff that "proved" productivity was severely underestimated, especially in the service sector. In hindsight, Greenspan's judgment at that time was correct—the latest data revisions show that productivity growth in the 1990s indeed far exceeded the statistics at that time. The annual contribution of productivity growth during the internet bubble period was about 1.5 percentage points.

But that is only half the story. By 2000, Greenspan's attitude underwent a 180-degree turn. He explicitly stated at the FOMC meeting in May 2000 that the sustained productivity boom had pushed up the equilibrium interest rate (r*), and the Federal Reserve needed to raise interest rates to prevent monetary policy from becoming overly accommodative He believes that this strong demand driven by supply-side factors must be balanced by higher real long-term interest rates. Ultimately, the Federal Reserve raised interest rates by 50 basis points at that meeting, accelerating the tightening pace. This move, along with signals from insiders beginning to sell stocks, collectively led to the eventual burst of the internet bubble.

Therefore, simply shouting "learn from Greenspan to cut interest rates" is a selective interpretation of history. The experience of this "master" precisely reveals the dilemma faced by the Federal Reserve in the face of technological revolution: should it embrace the deflationary effects it brings, or be wary of its potential to raise the equilibrium interest rate?

Is AI a remedy for deflation or a driver of inflation?

The report states that the impact of AI on the economy is key to determining Federal Reserve policy, but its direction remains hotly debated.

• On one hand, AI could become a powerful deflationary force. If productivity accelerates while wage growth remains stable, unit labor costs will decline. Companies can pass on cost savings to consumers by lowering prices. If new technologies intensify market competition, businesses will be forced to do so.
• On the other hand, the capital expenditure boom triggered by AI could push up the equilibrium interest rate. New technologies increase the expected return on capital, encouraging companies to make large-scale investments. With savings unchanged, more investment demand means the equilibrium interest rate (r*) will rise. If the central bank does not raise interest rates at this time, monetary policy will inadvertently become too loose.

It is worth noting that the reality of the 1990s was: wage growth outpaced productivity growth, and the share of corporate profits in GDP continued to narrow after peaking in the mid-1990s, even as the stock market continued to soar. This indicates that it was workers, not companies, who reaped the benefits of productivity improvements.

Three key questions determine the path of Federal Reserve policy

The report points out that TS Lombard believes the following three key questions will determine the path of Federal Reserve policy:

First, is the large-scale capital expenditure in the tech industry inflationary?

Although companies like Nvidia create value, a large amount of equipment is imported from overseas, and a larger trade deficit will obscure some inflationary effects. Data centers are not labor-intensive, making it difficult to overheat the labor market. The most obvious inflation risk lies in the energy sector—data centers require astonishing amounts of electricity due to cooling systems.

The share of electricity used by data centers in the U.S. is expected to rise from about 2% in 2005 to 12% by 2030. However, overall, the impact of AI capital expenditure on pushing up the equilibrium interest rate is limited, serving merely as another reason why the U.S. economy can withstand higher interest rates than in the 2010s.

Second, can AI bring about a significant increase in productivity like in the 1990s?

Researchers have found through experiments that AI can improve the efficiency of specific tasks like programming by 40%, but only about 30% of economic tasks can deploy this technology. This means an overall economic productivity increase of about 12%, which is dispersed over the adoption period The key question is: only a small portion of employment is based on cognitive or knowledge-intensive activities (construction, manufacturing, and professional services still have a strong physical component). Estimates of AI's annual productivity contribution vary widely: McKinsey predicts up to 4%, while MIT scholar Daron Acemoglu predicts only 0.5%. It is very difficult to replicate the productivity gains of the 1990s (an average annual increase of 1.5 percentage points).

Third, who benefits from the productivity gains?

Historical experience shows that workers, rather than companies, will reap the main benefits. In the mid-1990s, despite Greenspan and Yellen believing that workers were "traumatized" by digitization, wages actually grew rapidly, and there was no mass unemployment among workers.

The "flattening effect" of AI indicates that low-skilled workers benefit the most—this is different from the polarization of the labor market in the 1980s and 1990s. At that time, workers either "upskilled" into complementary industries like finance or entered low-skilled labor-intensive jobs. AI is more likely to help the middle class and drive faster wage growth.

The Federal Reserve will not actively burst bubbles, but may do so unintentionally

The report states that Greenspan's "clean up afterward, do not prevent beforehand" strategy has become a tradition of the Federal Reserve. He believes that central banks cannot identify bubbles in real-time, and attempting to curb bubbles through interest rate hikes will only cause unnecessary collateral damage to the broader economy (an additional 100 basis points hike will not stop investors chasing huge returns, but will harm sectors of the economy that are not in a bubble).

It is certain that the new Federal Reserve chair appointed by Trump will not actively burst asset bubbles. However, if inflation becomes a primary concern again in 2026, the situation will become delicate. The market is focused on a K-shaped economy, but if the bottom of the K recovers, there may be problems at the top.

TS Lombard believes that central banks have a habit of bursting bubbles, even if usually unintentionally. The current inflation dynamics are far less favorable than in the 1990s—during the 1990s, core PCE inflation remained below 2%, providing room for Greenspan-style accommodative policies.

This means that attempts to "replicate Greenspan" face higher risks, potentially igniting a tech bubble while fighting inflation, which is the biggest risk investors need to be wary of