The Yen Beneath the AI Boom

When we talk about the cost of artificial intelligence, we usually talk about GPUs, data centres and electricity.

We talk less about the money beneath them.

For decades, Japan supplied the world with something almost as valuable as compute: exceptionally cheap capital. Investors could borrow yen at very low interest rates, convert the money into dollars and invest it in higher-yielding bonds, equities and other assets. As long as the yen remained weak and markets stayed calm, the strategy produced relatively steady returns.

This is the yen carry trade. And as Japan gradually raises interest rates, it is coming under renewed scrutiny.

The question is not whether a quarter-point increase from the Bank of Japan will suddenly stop the AI boom. It will not.

The more important question is what happens if higher Japanese rates cause the yen to strengthen rapidly, forcing leveraged investors to sell assets at precisely the moment that the AI industry is becoming more dependent on external financing.

Japan's quiet change

The Bank of Japan ended its negative-interest-rate policy in 2024 and has since raised its policy rate to 1.0%, its highest level since 1995. The Bank has indicated that further increases remain likely if inflation and economic activity develop as expected.

Economists surveyed by Reuters largely expect a gradual path: 86% forecast another quarter-point increase to 1.25% by the end of 2026, while 70% expect the rate to reach at least 1.5% by the end of June 2027.

Across the Pacific, the US Federal Reserve held its target range at 3.5 to 3.75% on 29 July, although three policymakers voted for an immediate increase. That still leaves a 2.5 to 2.75 percentage-point gap between US and Japanese policy rates.

In other words, the carry trade has become less attractive, but it has not disappeared. The yen was still trading at more than 163 to the dollar shortly before the Bank of Japan's July meeting, close to its weakest level in four decades.

How a small return becomes a large risk

The basic trade is simple:

1. Borrow yen at a low interest rate.

2. Convert the yen into dollars.

3. Buy a higher-yielding asset.

4. Later sell the asset, buy back yen and repay the original loan.

Imagine borrowing 1.6 billion yen at 1%, converting it at 160 yen per dollar and investing the resulting $10 million at 3.75%.

If the exchange rate remains unchanged, the approximate return from the interest-rate difference is 2.75%.

But if the dollar falls from 160 yen to 144 yen, the currency movement overwhelms the interest earned. The position loses approximately 7.6%, before fees or any movement in the underlying asset.

This is why the direct cost of a quarter-point Japanese rate increase is not the principal danger. The danger is that the increase, or the expectation of further increases, causes a sharp appreciation of the yen.

The latest US futures data illustrate the vulnerability. As of 21 July, leveraged funds held 81,033 long yen contracts and 177,218 short contracts: a net short position of 96,185 contracts, equivalent to approximately 1.2 trillion yen in notional exposure. Futures are only one visible part of a much larger market involving forwards, swaps, options and bank lending, so they cannot tell us the true size of the carry trade.

However, they do show that a meaningful number of investors remain positioned for a weak yen.

Research from the Bank for International Settlements found that, during periods of heavy carry-trade activity, an unexpected 25-basis-point monetary tightening has historically been associated with an almost 10% move in the yen. That is not a forecast. It demonstrates how leverage can transform a modest policy surprise into a much larger currency movement.

As the yen strengthens, investors have to buy yen to close their positions. That pushes the currency higher still. Falling asset prices then create margin calls, which force additional selling.

The warning from 2024

We saw a version of this process in August 2024.

Technology and AI shares had already begun correcting when a perceived hawkish decision from the Bank of Japan collided with weaker US employment data. The yen appreciated, leveraged positions were closed and volatility accelerated.

On 5 August, Japan's TOPIX index fell 12%, the S&P 500 lost 3% and the VIX briefly moved above 60. The BIS estimated that yen-funded carry positions entering the episode may have totalled roughly 40 trillion yen, or around $250 billion at the time, while warning that the figure was probably understated.

Yet the disruption was short-lived. The S&P 500 recovered its losses within the same week, and credit markets were affected far less severely than equities.

That distinction matters. The carry trade did not create the original concerns about technology valuations or the US economy. It amplified them.

The lesson is not that a stronger yen automatically causes a technology crash. It is that when markets are already vulnerable, an unwind can turn an ordinary correction into a disorderly one.

AI investment is becoming more exposed to debt

The AI industry of 2026 is financially different from the industry of 2024.

The largest technology companies historically financed infrastructure from their enormous operating cash flows. But the scale of the current build-out, covering data centres, accelerators, cooling systems, grid connections and new power generation, is testing even those balance sheets.

The BIS estimates that private-credit loans to AI-related companies already exceed $200 billion and could reach $300 to $600 billion by 2030. It also found that these loans carry an average spread of approximately 6.2 percentage points above their reference rate.

This creates a direct connection between global liquidity and physical AI infrastructure.

One current example is the $14 billion data-centre venture between Meta and BlackRock in El Paso. BlackRock-managed funds will own 80% of the project, while $12.5 billion of the investment is expected to be financed with debt. By early July, AI-related bond issuance had reportedly reached $270 billion in 2026, almost twice the amount raised during the whole of 2025.

The visible AI boom may be built from concrete, steel, fibre and silicon. Increasingly, however, it is also built from credit.

What would be affected first?

A yen carry-trade unwind would not affect every part of the AI market equally.

1. Public AI and semiconductor shares

Listed technology companies would feel the effect first. They are liquid, widely owned and frequently used as collateral. When leveraged investors need to reduce risk quickly, highly valued market leaders are among the easiest assets to sell.

That can compress valuations even when the companies' underlying revenues have not changed.

2. AI start-ups and model developers

Private companies would experience the effects more slowly, through lower valuations and more difficult fundraising. Investors would demand clearer revenue, shorter paths to profitability and less dependence on repeated capital rounds.

Projects relying on continuous injections of speculative capital would be particularly vulnerable.

3. Debt-financed data-centre developments

Greenfield data centres, GPU-financing vehicles and projects dependent on refinancing would face higher return requirements and wider credit spreads. Some would proceed in phases; others could be delayed until tenants, power and financing were secured.

The greatest pressure would fall on speculative capacity rather than infrastructure supported by long-term customer commitments.

4. Hyperscaler investment

Amazon, Microsoft, Meta and other hyperscalers are more resilient, but not immune.

Amazon expects to invest around $200 billion in capital expenditure during 2026. It generated $148.5 billion of operating cash flow over the twelve months to March, but free cash flow had fallen to $1.2 billion, principally because of AI investment.

Microsoft generated $46.7 billion of quarterly operating cash flow against $31.9 billion of capital expenditure, while Meta held $81.2 billion in cash and marketable securities but expects 2026 capital expenditure of $125 to $145 billion.

These companies are unlikely to cancel strategically important AI programmes because of one Japanese rate increase. But a sustained rise in financing costs could change how projects are funded, where they are built and how quickly new capacity is added.

Joint ventures, leasing arrangements, private credit and off-balance-sheet financing are therefore likely to become even more important.

Capital scarcity would change where AI is built

If capital becomes more selective, AI investment will not necessarily disappear. It will move towards infrastructure that can demonstrate a credible economic return.

Projects with available power, committed tenants, existing connectivity and a realistic route to utilisation should be more resilient than speculative campuses based on distant demand assumptions.

It could also strengthen the case for a more distributed infrastructure model. Instead of concentrating every workload inside a small number of enormous training campuses, investors may increasingly favour phased capacity and inference infrastructure located closer to customers, data and sources of demand.

The next phase of AI infrastructure may therefore be judged less by the size of the announcement and more by the quality of the underlying economics.

The AI-Japan feedback loop

There is a final irony.

The AI boom is itself helping create the conditions for higher Japanese interest rates. The Bank of Japan has said that strong global AI demand is supporting Japanese corporate profits and business investment, while developments in AI-related demand are now among the factors it is monitoring when assessing the economy.

AI demand strengthens Japan's semiconductor and industrial sectors. That supports profits and investment. A stronger economy gives the Bank of Japan more room to normalise monetary policy. Higher Japanese rates can then unsettle the leveraged financial positions supporting global AI valuations and infrastructure.

The yen will not switch off AI

It would be an exaggeration to say that Japan can burst the AI boom simply by raising interest rates. The rate differential with the United States remains substantial, hyperscaler demand remains strong and many infrastructure projects are supported by long-term contracts.

But it would be equally mistaken to dismiss the yen as a distant currency with little relevance to technology investment.

The yen is not the power cable feeding the AI boom. It is one of the cheap funding lines running beneath it.

If that line tightens gradually, markets can adjust.

If it snaps, the lights will not go out. But we may quickly discover which projects were built to generate returns, and which were built because money was cheap.