The most crowded narrative of the past two years has been AI: computing power, chips, data centers, cloud services, and every asset being revalued around artificial intelligence. But now a more fundamental issue is emerging: AI does not just need computing power, it needs money, and it needs an enormous, long-term, low-error-tolerance amount of capital.

At the same time, the U.S. government, European governments, and the Japanese government all need money too. Soaring fiscal deficits, rolling debt refinancing, and surging interest expenses are pushing the global bond market into a new phase: capital is no longer cheap, and no longer abundant.

This means the capital market logic built over the past decade and more on low rates, ample liquidity, and unlimited financing capacity is being rewritten.

The real risk is not merely that one AI company faces higher borrowing costs, nor merely that the U.S. deficit has breached $2 trillion. Behind it, two enormous financing demands are crowding into the same market at the same time: on one side, the AI super-capital expenditure cycle; on the other, the government fiscal deficit cycle.

There is another key variable behind this storm: Japan. The world had grown used to Japan's cheap money, and now it must pay a higher price to compete for that capital. As domestic Japanese rates rise, Japanese investors are becoming more selective, with French bonds becoming the first casualty, followed by U.S. Treasuries.

When the two collide, the result is likely to be a systemic rise in the global cost of capital, a forced demand rationing in the bond market, followed by shocks to capital markets and ultimately transmission into the real economy.

AI financing enters an era of enormous scale and enormous cost

Over the past two years, when the market discussed AI, the focus was mostly on chips, models, applications, and valuations. But as AI competition enters deeper waters, the variable that determines victory or defeat is capital expenditure capacity. Training and deploying large models requires high-end chips, servers, networking equipment, cooling systems, land, power connections, and massive data centers. Each item means heavy asset investment, and the spending happens before revenue is realized.

This has rapidly turned AI from a technological revolution into a global-scale infrastructure financing revolution.

Data shows that the five major AI hyperscale data center operators have already issued about $220 billion in debt this year, more than double last year's scale. Wall Street expects AI-related capital expenditure to rise from about $700 billion this year to around $1 trillion next year. Over the past two years, companies have borrowed about $625 billion for data centers and AI investment, and Wall Street estimates the entire buildout cycle will cost nearly $5 trillion by the end of 2030.

Financing pressure is already visible in the cash flows of tech giants: Alphabet posted its first negative free cash flow on record in the second quarter, burning $5.9 billion, and raised its 2026 capital expenditure forecast to as high as $205 billion; Amazon's 2026 capital expenditure plan rose to about $220 billion, with free cash flow turning negative by $7.6 billion over the past 12 months.

More worrying is that the AI financing curve is shifting upward, and lower-rated AI-related borrowers are already bearing significantly higher capital prices. For smaller, lower-rated new cloud companies such as CoreWeave, Lambda, and Crusoe, the high-cost leveraged loan market is the only way out.

A typical case is Volta Infrastructure Holdings. The company launched a $5 billion leveraged loan for a Norwegian data center project, and market quotes showed its blended financing yield once approached 11%.

For a company in an expansion phase that still needs continuous investment in data centers and computing equipment, this is a very high profitability threshold. It means the project must not only be built and operated at full capacity, but also generate sufficiently high and stable cash flow for a long period to cover interest, depreciation, and subsequent refinancing pressure. The AI debt market has already moved from a phase where money was willing to pay into a phase where money demands high compensation.

Goldman Sachs statistics show that about $88 billion of lower-rated AI-related borrowing has entered the market this year. AI financing is gradually sinking down the credit spectrum: the highest-quality companies issue investment-grade bonds; slightly weaker companies rely on leveraged loans; even weaker borrowers must accept double-digit financing costs.

This is precisely the signal that the credit cycle is beginning to stratify.

AI may not be a bubble, but the financing conditions for AI are no longer cheap.

The U.S. government is also desperately seeking money

If it were only financing expansion in the AI industry, the market might still be able to digest it. The real trouble is that the world's largest borrower, the U.S. government, is also absorbing capital at an alarming pace.

Data from the U.S. Congressional Budget Office shows that for fiscal 2026 ending September 30, the U.S. federal budget deficit rose to $1.993 trillion, up 12% year over year and the highest level since 2021. Federal spending reached $7.4 trillion, while federal revenue was $5.4 trillion. The deficit is expected to exceed 6% of GDP.

Historically, a U.S. deficit ratio above 6% usually appears during wars, financial crises, or recessions. But the current U.S. economy is not in recession and has been expanding for years. This means U.S. fiscal policy has already entered a crisis-style deficit state outside of a crisis period.

What is more troublesome is that the deficit itself is creating a self-reinforcing effect. U.S. net interest expense has exceeded $1.1 trillion, accounting for more than one-fifth of all tax revenue and surpassing defense spending and Medicare spending. High rates push up refinancing costs, interest expenses devour tax revenue, the deficit then widens, and the government needs to issue even more debt, creating a dangerous loop.

It is worth noting that the impact of the recent rise in U.S. Treasury yields on the budget has not yet fully appeared. Higher rates will only enter the fiscal accounts gradually as new debt is issued and old debt is refinanced, meaning fiscal pressure will still be released with a lag.

Japan's era of cheap money is ending

Beyond the two torrents of financing demand, the global bond market also faces a long-underestimated demand-side contraction variable: Japanese investors are becoming more selective.

Data shows that as of July, Japanese investors held about 23 trillion yen, or $145 billion, of French government bonds, the largest overweight position in the euro area. But as Japan's 10-year government bond yield rose above 3% last month, a 30-year high, the yield advantage of French 10-year government bonds over Japanese government bonds after currency hedging has narrowed to only about 40 basis points, greatly reducing the appeal of holding French debt.

With the yield advantage disappearing and France's fiscal position continuing to deteriorate, Japanese money is beginning to loosen. Japan's holdings of French debt have fallen 2.5% from the end of last year. A fund under the global fixed income team led by Shinji Kunibe at Sumitomo Mitsui DS Asset Management has already liquidated French bonds due to fiscal concerns. Masayuki Nakajima of Mizuho said: Even if valuations look cheaper, Japanese investors' motivation to rebuild positions is declining.

U.S. Treasuries are similarly unable to escape. In the first half of this year, local Japanese investors net sold 4.45 trillion yen, or about $28 billion, of U.S. long-term government bonds, the first half-year net sale since 2022, while massively increasing holdings of domestic government bonds, with retail subscriptions hitting a record high since 2014.

Reidy pointed out that Japan does not need to sell off assets for U.S. borrowing costs to rise; simply buying less is enough. The world had grown used to Japan's cheap money, and now it must pay a higher price to compete for that capital.

The cost of AI and governments competing for the same pool of capital: rising rates

JPMorgan CEO Jamie Dimon gave the most direct judgment: Rates are rising, demand for capital is strong, and government financing is also heavy.

AI needs money, and governments need money. Companies need to build data centers, and governments need to cover deficits. The private sector needs financing to buy GPUs, and the public sector needs financing to pay benefits, interest, and defense costs. Everyone is looking for capital in the bond market, the loan market, and the private credit market.

When demand for capital surges at the same time, unless the supply of savings also rises sharply, the only possible result is higher prices.

And the price of capital is the interest rate.

Dimon does not deny the long-term value of AI. He even believes AI will eventually create enormous productivity returns, much like the internet. But he stresses that the issue is not only whether AI has value. The key point is the funding cost, construction cycle, regulatory resistance, legal litigation, power bottlenecks, and changes in technology routes for building AI infrastructure.

In other words, AI may be right in the long run, but short-term financing costs may also truly be high.

This is exactly what the current market most easily overlooks: an industrial trend that is correct in the long term can also create severe capital market pressure in the short and medium term.

The internet eventually changed the world, but the capital bubble around 2000 still burst. Railways eventually changed America, but 19th-century railway debt also triggered multiple financial crises. Infrastructure revolutions and capital market turmoil are not contradictory, and often occur at the same time.

In addition, Dimon also warned that there are structural upside risks to U.S. inflation. Facing a combination of factors including rearmament, immigration policy, and supply chain restructuring, he said bluntly that there is a risk that inflation remains sticky and rates continue to rise.

A great capital tightening may already have begun

Global capital tightening is transforming from macro pressure into market damage.

As competition for capital between sovereign financing and AI financing needs intensifies, European bank stocks fell sharply. On October 9, the European banking sector index SX7E fell about 4% at one point during the session and broke below its 100-day moving average; French assets are also being repriced, and the spread of Italian and Spanish 10-year government bonds over French government bonds has fallen sharply and dropped below zero.

Goldman Sachs analyst Privorotsky believes a larger problem is brewing beneath the surface of market moves.

In the post-financial-crisis era, the world economy operated on abundant savings and structurally low interest rates. Governments' fiscal systems were also built on the assumption that cheap capital would exist for a long time. Now that assumption is collapsing.

Sovereign financing demand remains enormous, and artificial intelligence is bringing unprecedented capital demand. According to reports, SpaceX is seeking $40 billion in financing, and Broadcom is also exploring a financing plan of more than $50 billion. These large financing needs will compete for the same pool of funds with government bonds and other corporate financing.

If AI investment can generate internal rates of return of 20% to 30%, then financing at rates close to double digits is still reasonable. But the question is what happens to other sectors that also need capital. This may explain why real yields are hard to fall. If the market expects a large amount of high-quality private-sector bonds to be issued soon, the appeal of investors buying sovereign bonds will decline.

Privorotsky proposes two possible outcomes: first, AI gradually becomes self-funding and financing pressure eases; second, capital is rationed in other areas and the more fragile parts of the economy begin to break.

CCC-rated credit spreads may already be sending an early warning signal. If the weakest credit tier continues to come under pressure, it means capital tightening is no longer just a problem in the rates market and is being transmitted to companies and assets with weaker financing capacity.

Who gets squeezed out first?

In a latest exclusive interview on The Julia La Roche Show, Dalio described this structure as a debt-induced heart attack: when the cost of funds rises enough to screen out marginal borrowers, the market will enter a passive and brutal demand rationing.

The pressure on the supply side is quantifiable: the U.S. federal government spends about $7 trillion a year and collects about $5 trillion in taxes, creating a deficit of about $2 trillion that must be continuously filled by issuing debt, while a considerable portion of fiscal space has already been eroded by high interest expenses. At the same time, companies making huge capital expenditures for AI data centers are also borrowing heavily. The two forces combined put sustained pressure on the supply side of bonds.

The demand side, meanwhile, is unstable. Foreign investors hold about one-third of U.S. Treasuries, but Japan and China, the two major holders, are both reducing allocations. If traditional buyers retreat at the margin, the bond market will need higher yields to attract new buyers, or force some borrowing demand out of the market.

Dalio believes that financing expansion by governments and AI giants pushes up the risk-free rate and credit spreads; lower- and medium-quality borrowers are then crowded out; investment in real estate, consumption, manufacturing, and small and medium-sized enterprises weakens; credit risk is exposed, and pressure ultimately transmits to the real economy. Mortgage borrowers and ordinary homeowners may be squeezed out first, and only then will pressure hit broader capital markets, finally dealing a heavy blow to the real economy.

Capital has a cost, and growth has a price

AI's long-term value may not be in doubt. As Dimon said, AI may create enormous value and even change productivity and social structure like the internet.

But AI has value does not mean AI investment at any price is reasonable; and governments can issue debt does not mean debt can expand indefinitely without cost.

When AI needs money to build data centers, the U.S. government needs money to cover a $2 trillion deficit, Europe needs money for fiscal repair, and Japan's rate normalization is drawing domestic capital away, competition in global capital markets will inevitably intensify.

Rising rates do not necessarily mean the economy will collapse immediately, but they do mean capital will no longer be universally accessible, financing will no longer be cheap, and asset prices can no longer easily be built on expectations of permanent easing.

The most profound change in the bond market may not be yields rising to some specific level, but that the market is beginning to recognize a fact again: money is not infinite.

When capital begins to be rationed by force, the first to be hit are marginal assets and weak-credit borrowers; then capital markets; and ultimately pressure will transmit to employment, consumption, real estate, and real investment.

The global battle for capital has already begun, and the bond storm may only just be starting.