Cursus Labs

What the yen intervention, AI financing and the Treasury long end reveal about the next liquidity regime.

The clearest sign of a liquidity problem is not always a shortage of money. Sometimes it is a shortage of institutions willing to hold what that money must finance.

The coordinated intervention in the yen, discussion of a larger FIMA backstop, the expansion of long-end Treasury buybacks and Nvidia’s plan to mobilise more than $500 billion of third-party capital for AI infrastructure have mostly been treated as separate stories.

They are separate stories. A central-bank repo facility is not an infrastructure fund, and a Treasury liquidity operation is not quantitative easing. But they may be responses to different versions of the same constraint: the marginal balance sheet is demanding a higher price to warehouse duration, illiquidity and residual-value risk.

That is what we mean by saying the world is short balance sheets. We do not mean it lacks dollars, savings or financial assets. Dollar cash demand remains strong, the dollar still dominates global finance, and stablecoins are extending its reach. The potential shortage is narrower: balance sheets willing and able to absorb long-term risk at yields that governments, companies and asset markets can all tolerate.

This distinction changes both the macro diagnosis and the sequence for Bitcoin. The next liquidity phase may not begin with a rate cut or a new round of QE. It may begin with facilities, buybacks, issuance changes and private funding vehicles that keep important financing channels open without broadly easing financial conditions.

The structural direction may be right while the path is wrong. Bitcoin does not necessarily benefit from the first stage of that process. Elevated real yields and competition for capital can remain a headwind even as policymakers support market functioning.

The yen was a stress test, not the story

Japan remains the largest foreign holder of US Treasuries, with approximately $1.12 trillion at the end of June. A prolonged defence of the yen could require the mobilisation of dollar reserves, potentially including Treasury holdings.

The Federal Reserve’s FIMA Repo Facility gives approved foreign monetary authorities a temporary source of dollars against Treasuries. Its purpose is to reduce the need for those institutions to sell securities into the open market when dollar funding is scarce.

But FIMA is an option, not evidence that the Fed has already monetised Japanese holdings. In the Federal Reserve’s latest H.4.1 release, covering 2 September, foreign-official repurchase agreements outstanding remained at zero. Its significance is architectural: it identifies the route through which a large official holder could turn Treasury collateral into dollars without adding another forced sale to the market.

Nor should Treasury-market protection be presented as the official reason for US participation in the yen intervention. Currency stability, trade and foreign-policy considerations all matter. The more useful signal is simpler: the portfolio choices of a major creditor have become more consequential just as the United States requires persistent access to long-term capital.

The mismatch is duration, not dollars

The generic de-dollarisation debate obscures a more interesting split.

Stablecoins are creating a large new constituency for dollar-denominated assets. Tether reported approximately $141 billion of direct and indirect exposure to US Treasury bills and overnight repo at the end of March. The BIS finds that growth in stablecoin demand can compress the short end of sovereign yield curves.

But a payment instrument promising rapid redemption naturally prefers cash-like reserves. Its home is bills and repo, not 30-year bonds. Stablecoin demand can therefore strengthen the dollar network while doing little to solve the long-duration funding problem.

This is the maturity split highlighted by Gavekal’s work on the changing functions of reserve currencies. Dollar adoption and long-duration Treasury demand do not have to move together.

The market can always clear that mismatch at some price. If a 30-year yield rises far enough, a buyer will appear. The Cursus thesis begins one step later: what happens when that market-clearing price becomes difficult to reconcile with fiscal arithmetic, strategic investment and valuations across the private economy?

That is the sense in which balance sheets can become scarce. The scarcity is not absolute. It is a scarcity of risk-bearing capacity at a politically and economically acceptable price.

The sovereign meets a new capital sink

The largest claim on long-term capital remains the state.

The Congressional Budget Office projects a federal deficit of $1.9 trillion in 2026, rising to $3.1 trillion by 2036. Net interest outlays are projected to increase from approximately $1.0 trillion to $2.1 trillion over the same period. The precise path will change, but the direction is clear: Treasury financing needs are large, persistent and increasingly sensitive to the cost of capital.

AI is a second and very different claim on capital. The BIS estimates that the five largest hyperscalers will spend more than $1 trillion on AI-related capital expenditure across 2025 and 2026. The next stage of that buildout is broadening beyond hyperscaler cash flows into corporate bonds, private credit, leases, infrastructure funds and special-purpose vehicles.

Nvidia’s $500 billion figure is an ambition to mobilise third-party capital over time, not a funded commitment. The Dallas Fed estimates that AI-related financing could produce as much as $360 billion of ten-year-equivalent duration supply in 2026, roughly one-eighth of the duration generated by Treasury issuance.

One-eighth is large enough to matter, but far too small to make AI the principal cause of long-end repricing. AI debt is also not a direct substitute for a 30-year Treasury. The overlap is the broader pool of institutional risk budgets, financing capacity and long-dated capital.

AI is therefore better understood as an amplifier arriving at an inconvenient point in the sovereign funding cycle. The state must refinance the past while the private sector finances a strategic buildout of the future.

Why plumbing comes before printing

The next policy response does not need to look like QE.

FIMA can convert official Treasury collateral into temporary dollar funding without an open-market sale. Treasury buybacks can improve the liquidity of older, less-traded bonds. Private AI vehicles can move capex, lease commitments and residual-value risk away from hyperscaler balance sheets and into funds, lenders and securitised structures.

These instruments are neither coordinated nor economically equivalent. Their common feature is functional: they change who holds liquidity risk, duration risk or residual-value risk when the original balance sheet is unwilling or unable to carry more.

The risk does not disappear. A Treasury remains duration after it becomes repo collateral. A GPU remains exposed to utilisation, pricing and obsolescence after it enters an infrastructure fund. A buyback can make an off-the-run bond easier to trade without improving the fiscal trajectory or removing the duration created by new issuance.

This is why “money printing” is too crude a description of the early phase. The initial response may be collateral engineering: mobilising existing assets, improving market functioning and distributing private risk before the public balance sheet absorbs it outright.

There are reasons to monitor intermediation capacity. Dealers do not have infinitely elastic balance sheets, and the SEC’s transition to broader central clearing of Treasury cash and repo transactions will require new operational, margin and clearing arrangements through the end of 2026 and into 2027. But no single auction proves a balance-sheet shortage. The harder evidence would be a repeated cluster of auction concessions, weak coverage, rising dealer inventories, deteriorating repo conditions and elevated rate volatility.

The put may sit under state capacity

The expansion of Treasury buybacks prompted immediate talk of a “Bessent Put” under the long bond. The evidence does not yet support that conclusion.

On 19 August, Treasury announced that from 9 September it would at least double the maximum size of liquidity-support buybacks in the 10-to-20 and 20-to-30-year sectors, from $2 billion to $4 billion per operation. The change takes effect tomorrow and will remain in place through the current refunding quarter. The amounts remain small relative to the stock and flow of Treasury debt. Treasury described the purpose as greater liquidity support. It did not announce a yield ceiling.

The distinction between three possible regimes matters:

1. Market-functioning support: limited buybacks, repo facilities and operational changes designed to prevent disorder without targeting an asset price.
2. Price suppression: issuance shifts, increasingly large duration purchases or an observable reaction threshold aimed at containing long yields.
3. Financial repression: sustained negative or artificially compressed real yields, captive regulatory demand, explicit yield-curve control or a durable transfer of duration onto the public balance sheet.

The current evidence fits the first category. The second is a hypothesis being tested. The third has not arrived.

The Cursus variant perception is that any future put may sit less under Treasuries, equities, AI stocks or Bitcoin than under the financing capacity of the state and infrastructure deemed strategically essential. A correction in a speculative asset need not trigger a response. A deterioration in Treasury-market functioning, weak auctions, repo stress or a funding problem that threatens strategic investment might.

That is a proposed reaction function, not an established one. It becomes credible only if the instruments scale systematically when those stresses appear.

AI may be a headwind for Bitcoin before it becomes a tailwind

Crypto narratives tend to collapse the sequence: fiscal dominance produces liquidity, liquidity produces debasement, and debasement benefits Bitcoin.

The structural direction may be right while the path is wrong.

In the first regime, public borrowing and AI investment increase demand for capital, energy and physical inputs. If inflation remains sticky, nominal and real yields can stay high. That tightens financial conditions and raises the opportunity cost of holding an asset with no contractual cash flow. For Bitcoin, this is a headwind.

In the second regime, funding stress produces targeted support for market functioning. That support may prevent a forced Treasury sale or improve collateral liquidity without creating a broad private-sector monetary impulse. Bitcoin can still be sold during the deleveraging that makes the facility necessary. FIMA availability is not the same thing as risk-asset liquidity.

Only the third regime is unambiguously more constructive: sustained yield suppression, declining real rates, broader monetary accommodation or regulatory measures that turn private balance sheets into captive buyers of government debt.

Even then, the Bitcoin conclusion is conditional. Lower real yields and broader liquidity must coincide with continued access to crypto markets. Financial repression can arrive with capital controls or tighter regulation as well as with easier money.

This produces a less exciting but more useful link between AI and Bitcoin. AI does not need autonomous agents to transact in digitally native money for the buildout to matter. It matters first because it adds to the demand for capital. If that demand contributes to a policy transition from market plumbing to sustained repression, Bitcoin may eventually become a convex beneficiary of the monetary response.

It is not necessarily a beneficiary of the initial pressure.

Gold remains the cleaner reserve expression of this regime. It has official-sector demand and does not require broad risk appetite. Bitcoin is the higher-beta expression: more vulnerable during the high-real-yield and deleveraging phases, but potentially more responsive if targeted support develops into a genuine liquidity cycle.

What would prove the thesis wrong?

The framework should be discarded, or at least materially weakened, if private markets absorb Treasury and AI-related risk at higher but stable yields without persistent auction stress, repo pressure or rising rate volatility.

It would also weaken if AI capex remains primarily cash-flow-funded; if Nvidia’s financing platforms fail to convert targets into commitments; if stablecoin and institutional demand broaden naturally along the curve; or if Japanese investors return as domestic and hedging economics change.

The policy thesis would be wrong if FIMA remains unused, buybacks remain small and technical, auctions continue to clear cleanly and no observable market threshold produces further intervention. A stable high-yield equilibrium is possible: expensive capital, but no disorder and no repression.

The Bitcoin thesis would be wrong if the asset continues to trade mainly as leveraged technology exposure even after real yields fall and broad liquidity improves. Digital scarcity is not automatically monetary adoption.

The variables to watch are therefore concrete: long nominal and real yields, term premium, auction concessions and bidder composition, dealer inventories, repo conditions, rate volatility, FIMA usage, buyback size and tenor, bill share of issuance, AI debt and lease growth, stablecoin reserve composition and Bitcoin’s response to changes in real yields.

The yen intervention gave the market a visible signal. The more consequential question is whether it remains an isolated currency operation or becomes an early example of a broader regime in which public and private institutions repeatedly redesign balance sheets to stop the long end from becoming the binding constraint.

The world is not short dollars. The open question is how far the price of long-term risk can rise before plumbing designed to preserve market functioning begins to suppress that price itself.

---

Sources and further reading

• Gavekal Research, A Shifting Global Currency Landscape, 6 August 2026.
• Jordi Visser, The Yen Signal: An AI Agent Macro Nexus Point, 10 August 2026.
• Lyn Alden, The $500 Billion AI Infrastructure Boost, 16 August 2026.
• TDR, Is AI Entering Its DAT Season?, 19 August 2026.
• Raoul Pal, The Bessent Put, 20 August 2026.
• Federal Reserve: FIMA Repo Facility
• Federal Reserve: H.4.1 release
• US Treasury: Major Foreign Holders of Treasury Securities
• US Treasury: Increased Long-End Liquidity Support Buybacks
• Congressional Budget Office: The Budget and Economic Outlook, 2026 to 2036
• Bank for International Settlements: Annual Economic Report 2026
• Dallas Fed: How AI debt financing impacts duration supply and interest rates
• SEC: Treasury Clearing Implementation
• Nvidia: AI Compute Infrastructure Financing Platforms

Cursus Labs publishes research frameworks, not investment recommendations. The analysis above is intended to separate structural direction from tactical sequence and should not be read as personalised investment advice.