Crosscurrents | The Marginal Buyer, Pacific Chokepoints, AI’s Consent Problem Inbox

Sanjeev Krishnan | September 29, 2026

When I Die, I Want To Come Back as the Marginal Buyer of the 10-Year U.S. Treasury

James Carville famously said, ​“When I die, I want to come back as the bond market.” He said it was mainly because the bond market indirectly reframed the public policy preferences of Bill Clinton’s 1992 presidential campaign. Instead of middle-class tax cuts and public investment, the bond market regulated federal budget behavior to focus on deficit cutting, and the 10-year fell from 7.5% to 5.2% by 1993, driven by tax increases and spending cuts. Bond market policy preferences ran counter to what many in the White House wanted to do. 

In 1992, the marginal buyer of the 10-year Treasury was mostly domestic, private, and price-sensitive, including banks, bond funds, and retail investors, with Japan emerging as a growing buyer. That changed over the next decade. After the 1997 – 98 Asian financial crisis and China’s entry into the WTO, the marginal buyer became official institutions, mainly Asian central banks. They were buying for policy reasons, not price. After 2008, the Federal Reserve itself became the largest price-insensitive buyer of all. Through successive rounds of quantitative easing, it absorbed trillions in Treasuries. For different reasons, each of these buyers kept the price of money low to serve its own objectives, including cheaper exports, reserve safety, and financial stability.

That 30-year arrangement produced a lot of free riders. Homeowners got cheap mortgages, and private equity got cheap leverage. Congresses and presidents of both parties spent on wars, tax cuts, entitlements, and nearly anything else without the discipline of a price-sensitive marginal buyer.

This isn’t new. The marginal buyer of U.S. debt has always defined America’s tug-of-war between captive, price-insensitive capital and free-market, price-sensitive capital. Some examples:

  • Post-Revolutionary War: Amsterdam bankers, domestic speculators, and the First Bank of the United States financed the new republic.
     
  • The Civil War: Northern households bought Union bonds, and so did German and Dutch investors. National banks were required by law to hold Treasuries to back their notes, which made them captive buyers by statute.
     
  • World War II: Patriotic households bought war bonds, but the real story is that from 1942 to 1951, the Federal Reserve pegged long-term yields at about 2.5%. It was the purest captive-buyer régime in American history.
     
  • The 1980s to the mid-1990s: This was the most expensive period. Free-market buyers demanded real returns after the inflation of the 1970s.

The pattern is consistent. Captive buyers, whether patriotic, regulatory, or policy-driven, kept the cost of capital low. Free-market vigilantes kept it high.

We are at a fork in the road, and unlike Yogi Berra, we can’t just take it. We are deep in debt after a long spending spree. We subsidize expensive weapons systems and wars that don’t end quickly. We have an aging population that is getting more expensive to finance. We have a healthcare system built around disease care, fed in part by our food system. And payroll taxes on labor alone account for roughly a third of federal revenue at a time when we are worried about labor substitution by robotics and AI.

Meanwhile, the old captive buyers are stepping back. Foreign official holdings have been shrinking as a share of the market for a decade. China has cut its Treasury holdings substantially from their peak. The Fed has spent the past several years running down its balance sheet, not up.

So who is the marginal buyer of the 10-year today? Nobody quite knows. It might be a leveraged hedge fund that holds Treasuries as repo collateral in a basis trade. That is capital that can vanish in a week. It might be stablecoin issuers, though they buy bills at the short end, not the 10-year. Or it might be Mr. and Ms. Market, who will want to be paid. History suggests these regimes last decades, not years. The headlines already show the consequences, with mortgage payments going up and portfolio returns under pressure.

That leaves two paths. The first is to rebuild a captive buyer through bank regulation that forces Treasury holdings, imposes yield caps, or puts the Fed back in the market. That is financial repression, quietly taxing savers to keep the government’s borrowing cheap. The second is to earn the trust of price-sensitive buyers through fiscal choices credible enough that free capital will lend to us at a fair price.

Either way, the free ride is over. We need to finance the new reality.

The Economics of the Pacific Ocean — Coming to an Instagram Scrolling Session Near You…Soon

Like many people in financial markets, I have become an expert in the Strait of Hormuz since February. Thanks to several cognitive offloading tools like Claude, Gemini, and Instagram, I know a lot about this chokepoint and can lecture on why it was so obviously a global threat to inflation, food supply, and the future of natural gas prices. (Satire aside, natural gas regional price volatility is a very important metric for understanding and underwriting many sectors and asset classes.)

Iran’s closure of the Strait disrupted 20% of global oil supplies and significant LNG volumes, which the IEA has called the largest supply disruption in the history of the global oil market. Global oil prices rose 50% between February and May 2026. By late April, European TTF gas futures were 35% higher than before the closure, and Asia’s JKM benchmark was up 51%. And because gas becomes fertilizer, the Kiel Institute estimates the cascade lifts food prices by roughly 2.7%, with the bottleneck running from energy through fertilizers to food.

We think of chokepoints as narrow constraints such as Hormuz, Malacca, Bab el-Mandeb, Suez, Panama, and the Taiwan Strait. Physically, they are modest. At its narrowest, Hormuz is only about 33 kilometers across. The Taiwan Strait is wider, yet at roughly 180 kilometers across, it handles ~87,300 annual ship transits, the highest volume of any global chokepoint according to IMF PortWatch.

But chokepoints aren’t really about narrowness. I believe a chokepoint exists when a larger system’s throughput concentrates, there is no substitute, and failure is non-linear. Economic chokepoints, whether as narrow as Hormuz or far less so, tend to share four properties. The first is concentration, in which a large share of a system’s throughput passes through a single point. The second is no substitute when rerouting is slow or expensive. The third is non-linear failure, when a small disruption causes disproportionate loss. The fourth is a broad transmission mechanism, through which the shock travels wide and far, from tanker insurance to fertilizer to your grocery bill.

A handful of small biological and climatic nodes qualify as economic chokepoints by this definition. Peru’s Humboldt Current is a strip of cold, upwelling water whose anchovy catch produces about 20% of global fishmeal supplies, the feed behind the world’s farmed salmon and shrimp. Panama’s Gatun Lake is a rain-fed reservoir that supplies the canal’s locks and drinking water to Panama City, home to nearly half the country’s population. Southeast Asia’s palm belt is the source of about 90% of global palm oil supply. Each is small, concentrated, hard to substitute, and prone to sudden failure. But none qualifies more than the Pacific Ocean, which is the common driver behind all three.

The Pacific is being tested like nothing in modern history. A historic El Niño is underway. The weekly Niño 3.4 index climbed to +3.0°C in mid-September, and all 22 forecast models agree it will strengthen further, staying very strong through winter 2026 – 27. NOAA puts a greater than 90% chance of a very strong event during the Northern Hemisphere fall and winter, and UCLA climate scientist Daniel Swain sees roughly a 70% likelihood that 2026 will be the strongest event on record since at least 1950, likely peaking between October and December. 

That has been in the headlines for a while. What has gotten less attention is how the Pacific is becoming one of the world’s scarcest and most contested economic spaces, potentially affecting us far more than Hormuz.

Start with fishing. The Pacific is the source of about 70% of commercially landed tuna and 65 – 70% of global tuna value. Peru set its main anchovy quota at 1.9 million tons, down about 36% from 2025 and the smallest cap the fishery has seen in 10 years. IMF benchmark fishmeal prices went from $1,626 per ton in October 2025 to $2,423 by June 2026. 

Then logistics. The Panama Canal, the Pacific’s back door to the Atlantic, is rationing depth. The canal authority has cut the maximum Neopanamax draft 5 times this year, down to 14.48 meters (47.5 feet) from September 3 until further notice. Scarcity is being priced in real time, with SK Gas bidding $5.3 million for an auctioned transit slot on Sept. 1, the highest reported price ever for the waterway. On the open ocean, Shanghai-Los Angeles spot rates rose to $7,838 per 40ft container last week, more than double the $3,485 from Shanghai to Rotterdam. 

Then energy. With Qatari LNG trapped behind the Strait of Hormuz, the marginal gas cargo is increasingly a Pacific decision. Europe is heading into winter with gas storage at the lowest level for this time of year since records began in 2011, alongside intense competition with Asian buyers for LNG cargoes. Goldman Sachs estimates that if Middle East exports normalize only gradually, December 2026 TTF would likely need to move above €100/​MWh. The Pacific’s climate also drives demand directly, as El Niño-driven heat and hydropower losses across Asia are among the reasons gas and thermal coal are rallying together. 

Finally, geopolitics. The Pacific’s own chokepoints make Hormuz look small. According to CSIS, the Strait of Malacca and the Strait of Taiwan each dwarf the Strait of Hormuz in maritime traffic, handling trillions of dollars in goods every year. Japan and South Korea alone ran about $801 billion in goods trade through the Taiwan Strait in 2022, and Taiwan’s chipmakers account for over 60% of global contract chipmaking capacity and over 90% of the most advanced chips. Hormuz concentrates energy. The Western Pacific concentrates energy, food, and silicon in everything else.

The Pacific Ocean is the world’s largest economic space. It covers about a third of the Earth’s surface and is bigger than all the land on Earth combined. While Hormuz’s impact is visible at the pump, the Pacific is a hidden tax that could show up everywhere, from European heating bills and your grocery bill to who controls the world’s supply chains, from semiconductors to car batteries to a can of tuna.

Become an expert, watch a Reel.

2 Supply Chains and 1 Force Majeure: The Supply Chain of Trust and the Supply Chain of Useful Energy to Useful Intelligence 

I was in New York last week for Climate Week, which I would retitle ​“10,000 Years of Useful Energy to Useful Materials,” but I digress. A few storylines converged throughout the week that I want to unpack together. Load growth of data centers, Oracle’s force majeure in New Mexico, midterm polling on AI, and continued financial market enthusiasm for capex financing all collided. 

Let me start with first principles: love, money, and power. OK, money and power. OK, money.

Money

The hyperscalers are the greatest compounders of EPS in the history of capital markets, with 15 years of 20%+ growth and close to a trillion dollars returned to shareholders. The Magnificent 7 went from a market cap of $2.2 trillion (12.4% of the S&P 500) in 2015 to $6.9 trillion in 2022 to over $22 trillion today, more than a third of the index. Their market cap grew nearly 800% over a decade, compared with 150% for the index. They generate nearly 70% of the S&P 500’s economic profit on a fraction of its revenue. Standard Oil and U.S. Steel come to mind as peers, but they are not. 

In a historic embrace of AI, these hyperscalers have gone from capex-light to capex-heavy. Hyperscaler capex was about $241 billion in 2024, roughly 0.8% of GDP, and is on track to reach an estimated 2.2% of GDP in 2026. The 2025 figure alone exceeded the telecom peak of the dot-com boom and passed the Manhattan Project, Apollo, and the broadband buildout as an annual share of the economy. Only the railroads compare, at roughly 6% of GDP in the 1880s, when U.S. rail mileage nearly doubled in a decade.

The bet is simple. Take useful energy (electrons) and turn it into useful intelligence (tokens). The token economy is digital. The generation is physical, and it lives in data center construction and operation. The best EPS compounders in capital markets are now spending like utilities, without a utility’s social license to do so.

Power

There are two supply chains core to underwriting the scale-up. One is physical and fairly well understood by the market. The other is trust and consent, which is not well understood but may carry greater urgency and risk. 

The physical one is constrained by capacity. The gas turbine market is essentially made up of three companies. GE Vernova ended the second quarter with 116 gigawatts of gas turbine backlog, up from 100 a quarter earlier, and is taking reservations for 2031 delivery. Siemens Energy sits at 69 gigawatts, Mitsubishi at 35, up from 23 a year ago. Lead times for a large-frame turbine that used to run 1 – 3 years now average 5 and stretch to 7, and prices for 2030 – 31 delivery are up 75%. That has tripled the cost of a new gas plant from about $800 per kilowatt in 2021 to 2,600−2,800 today. 

High-voltage transformers, which used to be a 24- to 30-month wait before 2020, are now a 5‑year wait. Sightline tracked 12 gigawatts of U.S. data center capacity announced for 2026 across 140 projects; only 5 gigawatts are actually under construction, and a quarter of the rest have not disclosed a power strategy at all. Electrical equipment accounts for under 10% of a data center’s cost but close to 100% of its bottleneck.

But turbines and transformers are solvable problems. They are a backlog, and backlogs clear as manufacturers add lines, prices ration demand, and delivery dates slip and are re-underwritten. The market can see the queue, price it, and hedge against it. Nothing about a 116-gigawatt backlog is unknowable.

The other supply chain is trust and consent, and it has none of those properties. It is not manufactured by three companies but granted by 3,000 counties, with no order book, lead time, or price list. Paying more does not clear the constraint and can make it harder. Yet this is the supply chain that the entire financing structure assumed was infinite.

To be clear, I don’t think social license will stop the hyperscalers from delivering compute. They have never lost a consent fight that mattered, through 20 years of antitrust, the Double Irish, Section 230, and Stargate announced from the White House. Workloads are portable in a way railroads never were. The tokens will be made, but the less-modeled question is how consent friction shifts the risk of this buildout into the financial system through channels nobody has mapped.

The canary is last week’s news from New Mexico. On Thursday, Oracle sent a force majeure notice to the Blue Owl Capital unit developing Project Jupiter, the 2.45-gigawatt Stargate campus in southern New Mexico. Oracle is not walking away. It is trying to delay the start of rent payments, potentially by three years, if the campus misses its 2028 date due to power. It remains on the hook for the project’s debt and, by one account, cannot terminate the lease under any circumstances. Blue Owl has about $3 billion of its own capital in, earning 9% during construction and roughly 11% levered once the building is done. The delays involve a gas pipeline that ran late, litigation over air and water permits, protests, and a project that became a midterm election issue. 

Jupiter is not a New Mexico story. Texas has halted new data center permits pending an audit. Ireland, Singapore, and the Netherlands have capped or frozen new capacity. Every physical buildout eventually hits the same wall because it needs things only a place can grant. Think interconnection, water rights, a gas lateral, a variance, or a tax abatement that survives the next election. Compute can be shipped from a fab in Taiwan, but consent cannot. Consent is not a commodity, and delays and denials are how a democracy is supposed to work.

The Domino

As democracy inches toward absorbing the social license implications of this historical capex cycle, what happens in a zoning hearing does not stay in a zoning hearing. 18 months ago, the buildout was, for all its size, contained. Circular financing had a closed loop in which Nvidia invested in OpenAI, OpenAI bought Nvidia chips through Oracle, and the whole thing sat on hyperscaler equity, hyperscaler credit ratings, and hyperscaler guarantees. If it went wrong, the losses landed on the shareholders of the most profitable companies on earth, who could afford them.

Then, financial markets did what they always do at this point in a cycle. They scaled the structure and distributed the risk. Consider what now sits between a county hearing and a pension fund.

  • Off-balance-sheet SPVs. Meta’s Hyperion campus in Louisiana is owned by a vehicle called Beignet Investor, 80% Blue Owl, 20% Meta, funded with about $27 billion of loans from PIMCO, BlackRock, and Apollo. Meta leases it back. The $27 billion is not on Meta’s balance sheet. Days later, Meta issued $30 billion of corporate bonds anyway.
     
  • Residual value guarantees. Meta gave the SPV’s investors a guarantee of up to $28 billion if the data center’s value falls below a threshold, disclosed in a footnote, with no liability recorded. NVIDIA is now providing up to 25% residual-value guarantees on GPUs through a $500 billion financing platform launched in August with Apollo, BlackRock, Blackstone, Brookfield, Goldman, and KKR. Broadcom is reported to carry up to $29 billion of residual value support on a roughly $36 billion Apollo and Blackstone facility that buys Google TPUs to lease to Anthropic. The chipmakers are manufacturing the residual value that a liquid market would otherwise provide.
     
  • GPU-backed and contract-backed lending. CoreWeave’s fourth delayed-draw term loan, $8.5 billion in a bankruptcy-remote SPV secured by GPUs and customer contracts, was rated A3 and priced at SOFR plus 225. Private credit’s outstanding loans to AI-related companies went from near zero to over $200 billion in a few years; Morgan Stanley projects another $800 billion. 
     
  • Unsecured and high-yield paper. Former bitcoin miners are raising billions in high-yield bonds to build data centers, backed by a hyperscaler’s conditional backstop. The five hyperscalers went from 3% of investment-grade debt outstanding to 15% of new issuance.

Each of these instruments includes a line for hardware obsolescence, tenant credit, and construction delay. The lawyers have started adding a force majeure clause. Every one of them is a contract whose value depends on another contract, so the failure of one propagates. A residual value guarantee is only as good as the guarantor’s cash flow, which is only as good as the lease, which is only as good as the power, which is only as good as the permit.

A permit delay becomes a force majeure notice, which becomes deferred rent, which becomes a covenant question in an SPV, which becomes a mark in a private credit fund that promised insurers and pensions 11% on AI infrastructure yield. A stalled campus means a stalled lease, which can put a residual value guarantee at risk of being called. Once called, that guarantee becomes a contingent liability that has just become real for a company whose free cash flow is already under pressure. 

AI capex is heading toward 3% of GDP and has carried much of the country’s growth since 2024, so a pause driven by consent friction rather than demand is a recession trigger. Meanwhile, utilities are building generation on a 30-year rate base, based on load forecasts that assume the campus lands. If it lands late or elsewhere, ratepayers carry the stranded cost, which feeds the backlash, which delays the next permit.

In 1873, railroad defaults were on county bonds, and the damage stayed within the county. This time, the county’s ​“no” is marked simultaneously in an investment-grade index, a pension’s private credit sleeve, and a utility’s rate case. The hyperscalers will eventually get their turbines and their tokens. What remains unclear is who holds the paper while the supply chain nobody underwrote decides how long ​“eventually” is.

Reserving a 2031 turbine slot is much easier than a 2031 zoning board approval. Financial markets beware. 

—

The views and opinions expressed are those of the S2G team members identified and do not necessarily reflect the views of S2G. This content is for informational purposes only, should not be taken as legal, business, tax or investment advice, or be used to evaluate any investment or security, and is not directed at any investor or potential investor in any investment vehicle sponsored by S2G. Investing involves risk, including the risk of loss. Specific companies mentioned in this podcast are for educational purposes and should not be construed as an endorsement of any kind. Please note that S2G may maintain investments in some of the companies discussed on this podcast. Any past performance discussed is not indicative of future results. For more important information, please see s2gin​vest​ments​.com/​d​i​s​c​l​o​sures.