The Debt Is the Point

There is a number that the AI trade has been quietly praying nobody notices. That number is 4.45%. That is roughly where the 10-year Treasury yield is sitting right now, and for about three years it was largely irrelevant to how people thought about Nvidia, Meta, Alphabet, or any company whose stock price is functionally a discounted cash flow model stretched forty years into the future. Rates were coming down. The Fed had told you so. You could afford to believe.

Kevin Warsh changed that last Wednesday, and the market spent Monday still absorbing what happened.

To be clear about what Warsh actually did: he held rates at 3.50%–3.75%, which everyone expected. What he did not do is what mattered. He stripped the easing bias out of the statement, abstained from the dot plot entirely — deliberately, as a philosophical statement about forward guidance — and allowed nine of his eighteen colleagues to pencil in at least one rate hike before year-end. The 2-year yield jumped 11 basis points on the day of the decision, and by Monday it had pushed to a 2026 high of 4.23%. The median dot for December now sits at 3.8%, which is above the current midpoint of 3.625%. From "we'll probably cut" to "we might hike" in one quarterly cycle. Warsh didn't even need to say a word of his own.

He also announced a communications task force to review, among other things, the dot plot itself. Gregory Daco at EY-Parthenon read that as a signal June might be the last dot plot ever published. Think about what that means for a market that spent the last three years anchoring every positioning decision to the Fed's own projections. The most powerful forward guidance tool in the world's most watched institution is potentially being dismantled by a man who publicly said he doesn't believe in forward guidance. Markets don't know how to price a Fed that refuses to tell them what it's thinking. That opacity is the point.

Into this environment — rates elevated, the Fed potentially hiking, the 10-year at 4.45% — SpaceX has decided to sell at least $20 billion in investment-grade bonds. It is burning roughly $4.3 billion per quarter. Its AI division lost $6.4 billion in 2025 on $3.2 billion in revenue. S&P projects negative free cash flow through 2029. Goldman Sachs, Evercore, and Oppenheimer collectively estimate SpaceX's capital expenditures could breach $1 trillion by 2031, with net debt potentially reaching $400 billion within five years.

Oppenheimer is not a bearish shop. That number is not a warning. It's a projection they expect management to pursue.

SPCX has now given back nearly 24% from its post-IPO peak over three sessions. On Monday alone it was down another 16%. The bond announcement panicked equity holders who had not quite grasped that the IPO was, at least in part, a prelude to becoming one of the largest corporate debt issuers in America. And the equity holders are right to be unsettled — not because the bonds are necessarily mispriced, but because there is a live conflict between what the debt market is being asked to fund and what the Federal Reserve is telling everyone about the cost of doing so.

This is the tension nobody on Wall Street is stating cleanly: the AI capex cycle has become a massive bet against the Fed.

Look at the construction. The four biggest hyperscalers — Amazon, Alphabet, Microsoft, and Meta — are on pace to combine for roughly $750 billion of AI infrastructure spend this year, an 80%-plus jump versus 2025. Goldman Sachs notes that big tech's capital spending as a share of cash flow is at its highest level since the dot-com era. For the profitable ones — Nvidia generating $200 billion in free cash flow annually, Alphabet with its balance sheet — the debt is manageable, the carry is fine, and rate sensitivity is real but not existential.

But SpaceX is not that. SpaceX lost nearly $5 billion in 2025 and posted a $4.28 billion net loss in Q1 2026 alone on $4.69 billion in revenue. The losses more than quadrupled compared to the same period a year earlier. Much of the bleeding traces back to its artificial intelligence division, which lost $6.4 billion in 2025 while generating just $3.2 billion in revenue.Strip out Starlink — which is the real collateral the rating agencies are lending against — and you have an AI operation that is an enormous cash furnace being handed a $20 billion hose. Meanwhile the people who set the cost of that $20 billion just signaled they are thinking about making it more expensive.

Alphabet fell 10% on Monday. Amazon, Palantir, and Meta were each down around 4%. The rotation was real: Micron and Sandisk each gained 5%, the Russell 2000 hit a record close, the Dow added 200 points. Chips and cyclicals up, hyperscalers and software down. Investors are not abandoning AI. They are abandoning the specific trade where a company spends $150 billion a year and earns less than it costs. The distinction matters.

There is a version of this that ends well. The hawkish turn came even as oil prices fell following a US-Iran interim peace agreement, suggesting the inflation concern the Fed flagged is broader than the energy supply shocks its statement If inflation stays elevated while growth holds, the Fed hikes once, the 10-year moves to 4.75%, and the companies that can service their debt — which is not all of them — survive the reset. SpaceX's Starlink cash flows are real. Twelve million subscribers generating $11.4 billion in annual revenue is a business. The rating agencies are not delusional.

The version that does not end well involves a scenario where Warsh is actually serious, the dot plot disappears, the Fed communicates less rather than more, and rates stay elevated in an environment of genuine opacity. Debt markets hate uncertainty more than they hate high rates. High rates with clear guidance is manageable. A Fed chair who has described forward guidance as a distortion, who abstains from his own projections, who is now forming a task force to potentially eliminate the only window into committee thinking — that Fed is a different animal. And it is that animal that SpaceX is borrowing $20 billion against.

The irony is structural. The AI boom was partly justified by the assumption that capital would stay cheap. The Fed cuts in late 2024, the narrative solidified, and every model that required enormous upfront investment and deferred profitability got a tailwind. Warsh arrived May 22nd, held one meeting, and the median dot flipped from cuts to hikes. Whether he hikes or not is almost secondary. The message was the message.

PCE lands Thursday. Core is expected somewhere around 3.3%. Watch what Warsh does with that number. He won't tell you in advance.