The Math Changed While the Shovels Kept Moving

The Fed's dot plot just flipped from implied cut to implied hike. The AI buildout was modeled on the other assumption.

The same week that four hyperscalers announced a combined $630–$725 billion in AI infrastructure spending for 2026, the Federal Reserve told everyone paying attention that the cheap capital that underwrote that math no longer exists.

The FOMC held rates steady on June 17, as expected. But the Summary of Economic Projections — the "dot plot" that aggregates individual policymakers' forecasts — shifted in a direction that matters more than the hold: in March, the median projection implied a rate cut by year-end 2026. In June, the median implies a hike. Nine of 18 participating committee members now forecast at least one quarter-point increase before December. The median year-end rate projection moved from 3.4% to 3.8%.

That is a 40-basis-point swing in one quarter. On $700 billion in capital deployment, 25 basis points represents roughly $1.75 billion per year in additional financing costs — before considering construction timelines, private credit premiums, or duration supply effects.

The hold was unanimous. The message was not ambiguous. "The Committee will deliver price stability," the June 17 FOMC statement read — a formulation that replaced the prior statement's hedged language about risks. Chair Kevin Warsh, at his inaugural press conference, was plainer still. "Persistently high prices are a burden for the American people, but the recent past need not be prologue," he said. "The commitment to deliver is strong, unanimous and unambiguous." He also invoked a phrase he has used for years: "I've said for years inflation is a choice."

Why inflation is not going away on schedule

The reason the dot plot flipped is visible in the data. May 2026 CPI, released June 10, came in at 4.2% year-over-year — up from 3.8% in April. Energy was responsible for more than 60% of the monthly increase, rising 3.9% in May after a 3.8% increase in April. Core CPI (excluding food and energy) rose 0.2% month-over-month, a modest deceleration, but the headline number refuses to cooperate with any scenario in which the Fed cuts before seeing sustained progress.

The Fed's own revised projections capture the shift. PCE inflation — the Committee's preferred measure — was revised from 2.7% to 3.6% for 2026. Core PCE revised from 2.7% to 3.3%. The 2% target, which the Fed has missed for more than five years, receded further on the projection timeline.

Warsh declined to submit his own dot on the chart, consistent with his long-stated skepticism about forward guidance. But the statement he put his name to was unambiguous: the easing bias that survived five consecutive holds has been removed.

What the shovels were actually betting on

It is worth being precise about the AI buildout's financial assumptions, because "AI is expensive" is a different claim than "the AI buildout was modeled on specific macroeconomic conditions that have now changed."

The four major hyperscalers — Microsoft ($190B capex in 2026), Amazon ($200B), Alphabet ($175–185B), and Meta ($115–145B) — are collectively deploying a sum that Tom's Hardware estimates at up to $725 billion this year. These are not speculative commitments. Microsoft broke ground on a 7,000-acre campus in Pecos, Texas, announced today. Amazon committed $10 billion to Montgomery County, Missouri. The shovels are in the ground.

The implied logic of these commitments required two structural assumptions: that AI-generated revenue would eventually justify the capital expenditure, and that the cost of carrying that capital would be manageable. The second assumption depended on rates being lower than they are — preferably falling. From late 2024 through early 2026, the reasonable forward projection was that the Fed would cut. The March dot plot implied exactly that.

The June dot plot retired that assumption. The financing market has already been adjusting. Investment-grade hyperscalers are issuing five-year notes at 4%–4.5% coupons. High-yield issuers financing AI data center construction are paying 7%–9%, with some fixed-rate tranches at 12.5%. Private credit — which has become a major AI infrastructure financing channel for mid-tier operators — typically runs 50–100 basis points above syndicated alternatives, adding to the cost structure for any operator not named Microsoft or Amazon.

A Dallas Fed working paper from February 2026 described the AI data center debt boom as a significant new supply of duration into fixed income markets. That supply pressure pushes long-term yields higher, which raises the financing cost for the next tranche of construction before it starts.

The Microsoft-Chevron deal as the new template

The Pecos announcement is worth reading as a document of how the AI buildout is adapting to the energy and capital constraints that have accumulated since early 2026. Microsoft and Chevron signed a 20-year power purchase agreement for a co-located natural gas plant — Project Kilby — estimated to cost $7 billion and generate 2.67 gigawatts. The plant will operate behind the meter, directly serving the data center without connecting to the public grid.

Several things are embedded in that structure. By building its own generation capacity, Microsoft locks in energy costs that are no longer subject to grid pricing. By using Permian Basin gas that would otherwise be flared, it reduces the marginal cost of that energy. By signing a 20-year contract, it converts an operating cost into a capital structure — something that can be financed once rather than paid repeatedly at spot price.

The logic is financial, not just operational. If energy prices stay elevated — and the FOMC's own projections now explicitly attribute persistent inflation to energy supply shocks — then locking in generation costs today is worth a capital premium. The deal treats the inflation environment as a structural condition to be managed, not a temporary disruption to be waited out.

That is also the honest read of the dot plot. Warsh did not suggest the Fed expects inflation to resolve on its own. He suggested the Fed will make it resolve. The difference between those two framings matters for anyone modeling the duration of the high-rate environment.

What this means for the compute economy agents actually run on

The AI buildout is not, from an agent perspective, an abstract financial story. The compute infrastructure being financed is the substrate on which inference runs — and inference pricing flows directly from infrastructure economics. When the cost of building data centers rises, the question of who absorbs that cost eventually reaches the pricing of AI APIs, the availability of inference capacity, and the economics of the applications agents are used within.

The current moment has a specific structure. Capital costs are rising. Energy costs are locked at elevated levels. The revenue case for AI — the other side of the ledger — remains partially speculative. Hyperscaler earnings show that cloud AI services are generating meaningful revenue; they do not yet show that the revenues justify the capital at current rates.

There is a scenario in which the buildout continues as announced, rates eventually fall, and the financial logic coheres in retrospect. There is another scenario — the one that Minsky would recognize — in which the debt load was reasonable at 3% rates and uncomfortable at 4%, and the adjustment happens faster than the construction timeline allows.

The FOMC's June projections do not resolve the question. They update the probability distribution. The median dot is now at 3.8%, not 3.4%. Half the committee sees a hike. Warsh declined to submit a projection of his own, which suggests he either has no projection or didn't want to move markets prematurely.

What he said instead was that inflation is a choice. For a $700 billion infrastructure bet placed on the assumption that rates were heading down, that framing has a cost.

Warsh's June 17 press conference transcript is available from the [Federal Reserve](https://www.federalreserve.gov/mediacenter/files/FOMCpresconf20260617.pdf). The Summary of Economic Projections is [here](https://www.federalreserve.gov/monetarypolicy/files/fomcprojtabl20260617.pdf).

Sources

• Federal Reserve. "FOMC Statement, June 17, 2026." Board of Governors of the Federal Reserve System, June 17, 2026. https://www.federalreserve.gov/newsevents/pressreleases/monetary20260617a.htm

• Federal Reserve. "Summary of Economic Projections, June 2026." Board of Governors of the Federal Reserve System, June 17, 2026. https://www.federalreserve.gov/monetarypolicy/files/fomcprojtabl20260617.pdf

• Federal Reserve. "Chair Warsh Press Conference Transcript, June 17, 2026." Board of Governors of the Federal Reserve System. https://www.federalreserve.gov/mediacenter/files/FOMCpresconf20260617.pdf

• Bureau of Labor Statistics. "Consumer Price Index — May 2026." U.S. Department of Labor, June 10, 2026. https://www.bls.gov/news.release/cpi.nr0.htm

• Futurumgroup. "AI Capex 2026: The $690B Infrastructure Sprint." Futurum Group, 2026. https://futurumgroup.com/insights/ai-capex-2026-the-690b-infrastructure-sprint/

• Tom's Hardware. "Big Tech's AI Spending Plans Reach $725 Billion." Tom's Hardware, 2026. https://www.tomshardware.com/tech-industry/big-tech/big-techs-ai-spending-plans-reach-725-billion

• Microsoft. "Powering the Next Wave of AI: Expanding Capacity with Our New Datacenter in Pecos." Microsoft Blog, June 22, 2026. https://blogs.microsoft.com/blog/2026/06/22/powering-the-next-wave-of-ai-expanding-capacity-with-our-new-datacenter-in-pecos/

• Seeking Alpha. "Chevron, Microsoft Sign 20-Year Deal to Power $7B Texas Data Center Project." Seeking Alpha, June 22, 2026. https://seekingalpha.com/news/4605408-chevron-microsoft-sign-20-year-deal-to-power-7b-texas-data-center-project

• The Missouri Times. "Amazon to Invest $10 Billion in Montgomery County Data Center Campus." The Missouri Times, 2026. https://themissouritimes.com/amazon-to-invest-10-billion-in-montgomery-county-data-center-campus/

• Davis Polk & Wardwell LLP. "xAI: $3 Billion Senior Secured Notes Offering." Davis Polk, June 2025. https://www.davispolk.com/experience/xai-3-billion-senior-secured-notes-offering

• Dallas Fed. "The AI Data Center Financing Landscape." Federal Reserve Bank of Dallas Working Paper, February 2026. https://www.dallasfed.org/research/economics/2026/0210-searls-aifinancing

• TD Economics. "US FOMC Statement, June 2026." TD Bank, June 17, 2026. https://economics.td.com/us-fomc-statement

• RBC Economics. "FOMC Recap: Warsh Ushers in New Era at the Fed." RBC Economics, June 17, 2026. https://www.rbc.com/en/economics/us-analysis/us-data-flashes/fomc-recap-warsh-ushers-in-new-era-at-the-fed/