Two facts sit side by side on the same financial page. US AI infrastructure investment is still accelerating. The thirty-year US Treasury yield touched five point seven percent on October 7, a level not seen since 2002.
Together they describe a layered capital-absorption system — risk capital into AI equities, safe capital into Treasuries, middle-of-the-road capital into investment-grade US corporate debt — that has been quietly making the United States the single most attractive destination for global capital. The system is now bumping against its own weight.
It does not have to fail all at once. It only has to fail partially, in the right places, in the right order.
The Four Layers of the Siphon
Conventional commentary describes the US financial position as dollar hegemony. The actual structure is more interesting than that word suggests. The dollar sits at the top — most global trade is invoiced and settled in it, most cross-border financing is denominated in it. US Treasuries sit underneath as the largest sovereign bond market in the world and the benchmark against which almost every other fixed-income instrument is priced. The ten-year and thirty-year yields, currently near five point three and five point seven percent, are the risk-free rate anchors that determine how every other asset is valued.
US equities sit underneath that. The largest concentration of technology platforms, semiconductor companies, cloud providers, and internet businesses anywhere on earth trades on US exchanges. The AI build-out has reinforced that concentration rather than diffused it. The financial plumbing — clearing systems, custody, banking, exchanges, legal frameworks, the venture capital industry — sits underneath everything. The plumbing is what turns a collection of markets into a single integrated capital-allocation machine.
The four layers together give the United States one specific advantage that no other economy has. Whatever an investor's risk tolerance, there is a US asset for it. That is the real US card. Not the dollar. The dollar ladder.
The Left Hand: Risk Capital
AI is the story US equity markets are currently telling most loudly. The largest technology companies continue to spend hundreds of billions of dollars on data centres, GPU clusters, power infrastructure, and model training. What markets are pricing is not the AI revenue booked so far. They are pricing the productivity, profit pools, and new industries that AI may produce over the next decade. The positive feedback loop is real. AI expectations rise. Tech valuations rise. Companies raise more capital. Companies spend more on AI infrastructure. The chain repeats.
The risk is also real. Some AI companies are growing revenue fast. Their infrastructure and compute bills are growing faster. Losses at the leading AI labs are widening, not narrowing. The question is whether the value arrives on a schedule that matches the capital already deployed. That gap is where bubbles are born.
The Right Hand: Safe Capital
Treasuries attract the most conservative capital in the world — pension funds, insurance companies, sovereign wealth funds, corporate treasuries. For investors who need to preserve principal and stay liquid, there is no substitute.
What is interesting about the current configuration is the implication of high yields. In a globally dollar-saturated system, they can pull capital in rather than push it out. If short-dated US Treasuries pay above five percent, the relative attractiveness of foreign-currency risk premia falls. US safe assets do not need to force capital in. They only need to remain competitive enough that capital chooses not to leave.
The trap is the same. High yields attract capital. High yields are also the cost of US government borrowing. Annual interest expense has reached roughly one trillion dollars. With long rates near their highest level in two decades, every newly issued bond locks in higher future costs.
The Shared Pool
Surface appearances suggest AI stocks and Treasury bonds are competing assets. In fact, they share a capital pool. US AI companies need hundreds of billions of dollars of fresh capital to fund the build-out. The US Treasury needs trillions to roll over existing debt and finance the deficit. Long rates rise. Risk asset valuations compress. If long rates stay elevated long enough, the AI funding math starts to look very different from how it looks today. The two largest capital demands in the world right now are drawing from the same savings base, and the marginal investor who chooses between them is the same marginal investor.
The Correlation That Has to Stay
For decades, one piece of the system has been quietly load-bearing. When global investors panic, they sell risk assets, buy dollars, and buy Treasuries. The negative correlation between US equities and US bonds during stress events is one of the central reasons the US financial system has been able to absorb shocks without spiralling.
The structural risk is that this correlation breaks down. Stocks drop. Bonds drop. The dollar drops. The combination is rare, but it is the most dangerous one for the US financial system, because it implies that the usual internal hedge no longer functions.
Early warning signs are visible. The New York Fed's ten-year term-premium measure has risen to its highest level in twelve years.
The Fed's Limited Lever
The Federal Reserve controls the short end of the curve. It does not control the long end. The September meeting kept the policy rate at three point seven five to four percent. The Fed is keeping rates high not because the economy is collapsing, but because the AI-driven investment boom has been stronger than expected.
The long end is being set by a different combination. Inflation expectations, real growth, fiscal supply, term premium, and global demand for dollar safety. Cutting too quickly risks reigniting inflation. Holding too long turns the fiscal interest bill into the dominant political-economy question. Engineering a yield decline through unconventional tools risks undermining confidence in the Fed's commitment to price stability.
The Three Conditions
Whether the structural strain produces a crisis depends on three things happening more or less together. A system-wide downgrade of AI profit expectations. The loss of stable marginal demand for long-dated Treasuries. The breakdown of the internal hedge, where stocks and bonds fall together and the dollar falls with them.
What is genuinely novel is that the US financial system is being asked to fund the largest peacetime capital programme in history, while simultaneously absorbing the world's safe-capital flows, while simultaneously keeping its internal hedges functional.
What the Rest of the World Has to Do
The most important implication for non-US economies is not that the United States is about to collapse. The right question is whether the capital-absorption machine can keep running at full capacity for another decade.
If the answer is yes, global capital continues to flow into US assets at scale. If the answer is no, the structure that currently makes US assets uniquely attractive begins to lose its edge. The strategic task for every other economy is not to wait for the United States to fail. It is to make its own assets good enough that capital allocation becomes a real choice.
The US card is also its largest constraint. The larger the AI-driven capital siphon becomes, the more the United States depends on it continuing to work.