The most important number from the Federal Reserve's Sept. 16 meeting wasn't the one the committee set. It was 5%.
The Fed raised its benchmark rate a quarter point to 3.75%-4.00%, the first hike in three years, with all 12 voting members in favor. The market had expected it. Prediction markets had all but assured it.
And yet the Dow closed down 631 points, off 1.21%, with the S&P 500 down 0.45% and the Nasdaq off 0.01% — all three having bounced off their intraday lows in the final 40 minutes.
The culprit was two words. In his prepared remarks, Chair Kevin Warsh said the action would support a "timelier return" to the committee's 2% goal. Wall Street read the phrase as a warning. At Jackson Hole on Aug. 28, Warsh had already drawn the line: "We must be confident that underlying inflation is moving to our objective, clearly and at sufficient speed. Otherwise, we have work to do."
"Timelier return" told the market that inflation isn't falling fast enough — it has run above the 2% target for more than five years — and that the Fed intends to do something about it.
The dot plot reinforced it. The median projection calls for another 50 basis points before year-end; one report has the majority ruling out cuts until 2027, another until 2028 at the earliest. Core inflation returning to 2% is now pushed out to 2029.
Then there's the rate the Fed doesn't set, and the one that actually prices everything: the 10-year Treasury yield, which just hit 5%. The 30-year touched 5.3% last month.
Warsh was asked what was behind the rise. He gave three reasons: economic growth, geopolitics — the Iran war showing up in the crack spread between crude and refined products like diesel — and, third, the AI buildout's borrowing.
"The so-called hyperscalers are out in the market raising funding," Warsh said. "And so the competition for capital is real. And I think it partly explains the increase in yields."
He called the 10-year "the most important asset anywhere in the world" and "the risk-free asset upon which every price of virtually every asset in the world is related to."
The scale is not subtle. The five major hyperscalers issued $121 billion in U.S. corporate bonds in 2025, against an average of $28 billion a year between 2020 and 2024, per BofA Securities. Morgan Stanley estimates AI-related global debt hit nearly $236 billion by the end of May — four times the pace of a year earlier — and forecasts roughly $570 billion for full-year 2026.
Hyperscaler capital spending now runs close to 100% of operating cash flow, with some dipping negative. These companies can no longer fund the buildout from profits alone.
That is the contradiction worth sitting with. The AI trade is now expensive to finance partly because the AI trade is so big that its own borrowing is pushing up the discount rate applied to its own future cash flows. The loop is self-tightening.
The AI trade is now expensive to finance partly because the AI trade is so big that its own borrowing is pushing up the discount rate.
Not everyone buys it. PIMCO argues the AI effect on Treasuries is overstated, attributing most of the move to the Iran conflict and a repricing of Warsh's willingness to hike. MSCI notes hyperscaler spreads have widened to normal investment-grade levels, having once traded at something close to government quality.
Warsh conspicuously did not list the federal deficit, the explanation most bond investors themselves give. He left that alone, consistent with his stated view that Fed independence means "we stay in our lane."
He did announce an internal Fed task force on AI, due to report by year-end, to study "the implications for our future policy conjuncture." He offered no details. Asked about AI safety risks, he said those are decisions for other parts of government.
So the market has to price a Fed that says it may or may not be done. Goldman Sachs vice chairman Robert Kaplan said the market has interpreted Warsh as seeing more hikes than the dot plot shows, "but that may be an overreading." Mackenzie's Dustin Reid expects "more rather than less" hikes in 2027. Goldman's Kay Haigh says the Fed "has signaled it does not at this stage envisage an aggressive tightening cycle."
Former Fed vice chair Roger Ferguson framed the real question: is this "just a little bit of a tightening to make sure they're definitely neutral, or just slightly restrictive."
Warsh himself said he'd be "hard-pressed to describe broad financial conditions as restrictive." In the press conference he was blunter: "The plain fact is that inflation is too high, and has been for too long. This summer's inflation readings do not tell me that underlying trends have meaningfully improved."
For AI names, the practical sorting has already started. Morgan Stanley's August report told investors to prioritize AI businesses with existing profits and positive cash flows, arguing the boom is "entering a more mature phase" — the first stage was excitement and possibility, the current stage is "what AI costs, what it earns and how quickly it can be monetized."
Cash-generative firms rely less on capital markets. They can self-fund capex. As borrowing costs rise, that becomes a competitive weapon in what is still an arms race.
The fundraising data shows how concentrated the easy money had become. Global startup investment hit records near $510 billion in the first half of 2026, per Crunchbase, with OpenAI and Anthropic alone capturing more than $217 billion — over 40% of the total. Deal counts outside those mega-rounds fell sharply.
SpaceX's June IPO raised more than $85 billion, and the company completed an upsized $25 billion bond offering the same month. OpenAI and Anthropic are reportedly pursuing their own listings. Private credit lending to AI-related firms approached $200 billion in 2025 and keeps growing, at premium pricing that reflects the new regime.
Housing is the cleanest read on what a 3.75%-4.00% funds rate does downstream. Mortgage rates flirted with 7% recently, and purchase applications last week ran 19% below the same week a year ago, per the Mortgage Bankers Association. Nobody refinances into rising rates. That hits originators and home improvement retailers, since buyers and sellers spend heavily to renovate around a sale.
Private equity has its own version of the problem. The Wall Street Journal flagged that the hike is expected to wreak havoc on funds already struggling to sell companies and return cash. The metric that matters when distributions slow is DPI, not IRR — IRR can hold up on carrying marks while nothing has actually been returned. DPI only moves when cash lands.
That distinction is the whole story of this cycle in miniature. Marks can be defended. Cash cannot.
Which is why one strategist's worry deserves more attention than it got: the consensus that the Fed hikes and the long end rallies may not survive contact with the arithmetic — higher interest costs, lower real GDP growth, less ability to inflate the debt away, bigger deficits, more issuance, higher yields. Jim Cramer says history offers a playbook for navigating a hiking cycle. History also says playbooks written during easing rarely survive the first real test.
The AI trade doesn't die at 4%. It reprices. The companies that can pay for their own data centers get cheaper capital relative to everyone else, and the ones that can't discover what their cost of equity was always going to be.