The Japanification of AI

The efforts of the US Treasury and, arguably, the Federal Reserve to keep interest rates low by citing subdued core inflation, even as asset prices boom on the back of AI infrastructure spending, recall the policy choices that accompanied Japan’s asset bubble.

History, they say, rhymes. For the philosophically minded, the rhythm may lie in the constancy of human impulses. For market participants, it is more familiar: greed and fear repeatedly drive economic and financial cycles.

Against this backdrop, it is instructive to compare the present US-centric AI infrastructure boom with Japan’s real estate boom of the late 1980s and consider what the parallels may imply for the endgame.

Japan entered the 1980s as an economic marvel, having recovered rapidly from the Second World War. A very weak yen, at about ¥250 to the dollar, had helped turn it into an export powerhouse. Its economy was roughly 80% the size of the US economy (on a PPP basis), and Japan aspired to build an Asian trading bloc around its economic strength.

The 1985 Plaza Accord changed that equation. The yen appreciated sharply to around ¥150 to the dollar. Fearing a recession in its export-driven economy, the Bank of Japan cut interest rates aggressively to 2.5% from 7.25% at the beginning of the decade.

Abundant liquidity, low interest rates and an economy with the wind in its sails produced a speculative boom. Stocks surged, followed by real estate values. Banks, taking advantage of deregulation and greater flexibility, increasingly lent against assets, principally equities and property.

Cross-shareholdings among non-financial companies, and between companies and banks, obscured ownership, economic exposure and leverage. The system appeared prosperous, stable and self-reinforcing.

Once lending became collateralised by asset values rather than underlying business cash flows, rising prices themselves created room for larger loans and repeated refinancing against the same assets. Non-financial companies were incentivised to establish entities whose sole purpose was to speculate in equities and property.

An often-cited illustration of the excess is that, at one point, the grounds of Tokyo’s Imperial Palace were said to be worth more than all the real estate in California.

Stable goods and services inflation gave the Bank of Japan a rationale for maintaining loose monetary policy even as asset-price inflation accelerated. The Louvre Accord of 1987 reinforced this bias by seeking to stabilise the yen after it was judged to have appreciated too far.

AI Echoes
The AI-driven US boom now presents an unsettlingly familiar picture. The rally in equities has been breathtaking, with the US accounting for around 65% of global stock market capitalisation and AI-adjacent stocks driving much of the recent increase.

Investment in every layer of the AI ecosystem, including land and construction for data centres, chips, models and hyperscalers, has reached unprecedented levels. Yet estimates from long-established market observers suggest that the industry’s eventual cash returns may prove inadequate relative to the scale of investment.

At the same time, President Donald Trump has publicly pressed for lower interest rates, while the Federal Reserve under Kevin Warsh has kept rates low despite strong and public dissent from other members of its policymaking committee. Once again, the focus has remained on subdued core goods and services inflation, with less attention being paid to asset-price inflation.

The US Treasury has also intervened against market-driven increases in long-term interest rates. Several indices classify financial conditions as very easy, while fund-manager surveys indicate that cash holdings are close to historic lows, reflecting a high degree of investor optimism.

Leverage is also substantial. The recent demise of the 4x-leveraged hedge fund Situational Awareness in the AI trade, together with rumoured losses among proprietary trading firms such as Jane Street, illustrates the scale of the risk-taking.

Cross-holdings and the associated opacity have returned in a different form, through circular financing arrangements that allow investments to be round-tripped as revenue. Private credit has also produced an explosion in lending to opaque counterparties, with significant sums financing the AI infrastructure build-out.

The parallels are difficult to miss: low interest rates under political pressure, a narrow focus on core inflation, very easy financial conditions, credit-fuelled asset prices and opaque cross-holdings. Greed, at least, appears to rhyme.

Japan’s Reckoning
Japan’s government and central bank eventually recognised the dangers posed by asset-price inflation in the early 1990s. The discount rate was raised from 2.5% to 6.0%, money supply was reduced significantly, and substantial real estate and consumption taxes were introduced. Credit controls also imposed severe restrictions on real estate finance.

The lost decades that followed need little retelling. Asset values fell by as much as 50% from their peak in the immediate aftermath, followed by further declines of 30–35% from the peak over the decades that followed. The economic pain accompanying the collapse was profound.

Will fear rhyme as well? Could the AI bubble burst, triggering an asset rout and leaving investors with sub-par returns, particularly when the US has limited fiscal room to manage the fallout?

Only time will tell. No one has 20/20 foresight. Central bankers still lack a settled framework for identifying bubbles, let alone deciding whether they should be deflated pre-emptively.

What is persuasive, however, is that the probability of fear rhyming appears uncomfortably high.