Goldman’s $800 billion AI capex forecast is becoming the Fed’s inflation problem

What Goldman actually upgraded

In a research note published early this week, Goldman economist Elsie Peng raised the bank’s 2026 U.S. business investment forecast from 6.5% to 7.8% on a fourth-quarter-over-fourth-quarter basis. The driver was not a broad cyclical rebound. It was AI infrastructure: annualized spending hit $650 billion in the first quarter and is on pace to exceed $800 billion by year-end, according to Goldman’s published analysis and follow-on reporting from CryptoSlate.

The spending spans chips, servers, power infrastructure, and data-center shells. Goldman estimates AI capex will add only 0.1 percentage point to measured GDP growth in 2026 because much of the equipment is imported — but 0.3 percentage point to “true” domestic investment. That gap matters for policymakers. The Fed watches nominal demand pressures even when national accounts undercount where the dollars land.

Expanded tax expensing under recent legislation adds another tailwind: Goldman attributes roughly 3 percentage points of 2026 capex growth to new depreciation incentives, concentrated in manufacturing, transportation, and industrial sectors. AI and fiscal policy are stacking, not offsetting each other.

Why capex counts as inflation for central bankers

The textbook case for AI as disinflationary is familiar: productivity gains eventually lower unit labor costs, models get cheaper, and software eats repetitive work. Chair Kevin Warsh has voiced versions of that optimism in confirmation testimony, arguing AI could prove “structurally disinflationary.”

The near-term arithmetic runs the other direction. $800 billion in annualized spending pulls on the same scarce inputs the June chip selloff exposed: advanced semiconductors, high-bandwidth memory, transformers and switchgear, construction labor in power-constrained counties, and long-lead electrical equipment. When hyperscalers, oil majors, and sovereign wealth funds bid for the same contractors, prices rise before output does.

The San Francisco Fed’s June 4 FedViews release made the timing explicit. First-quarter GDP grew at a 1.6% annualized rate — below the Fed’s 2.0% trend estimate — yet business investment in AI infrastructure was the largest single contributor to growth. Consumer spending held up but underwhelmed expectations as gasoline and grocery inflation squeezed household budgets. Geopolitical stress in the Strait of Hormuz kept commodity prices elevated, adding another layer of cost pressure unrelated to AI but compounding it.

Fed Governor Michael Barr has pushed back on the idea that AI justifies easier policy, saying he does not believe the boom should be a reason to cut rates. That internal split — Warsh’s long-run productivity hope versus Barr’s near-term demand caution — is the frame for the June 16–17 FOMC meeting, Warsh’s first as chair.

Markets have already repriced the hold

Bond and derivatives markets moved ahead of the Goldman note. CME FedWatch data cited by PrimeRates shows the probability of holding the federal funds rate at 3.50%–3.75% through December 2026 rising from roughly 38% in mid-April to about 72% by late May. The odds of a June cut fell from 55% to roughly 23% over the same window. Prediction markets now assign above 93% probability to a hold at the June meeting, per CryptoSlate’s tracking.

The repricing is not abstract. Higher-for-longer rates raise the discount rate on long-duration assets — growth equities first, then crypto. That is the mechanism behind the record Bitcoin ETF outflow streak and the Nasdaq’s worst week in a year: institutional portfolios are not merely rotating from BTC into AI stocks. They are shrinking total risk exposure as the liquidity tailwind they priced for 2025–2026 fades.

Goldman itself does not expect Middle East oil spikes to derail the capex boom; Peng wrote that higher energy costs should have only a “modest” impact on overall investment growth. Tariff drag is easing too — from an estimated 1.5 percentage points off 2025 capex to 0.7 points in 2026. If anything, that makes the AI buildout look more resilient to external shocks, which is bullish for equipment vendors and bearish for rate-cut bulls.

The productivity lag trap

Every major technology cycle faces the same policy timing problem: costs arrive on corporate balance sheets immediately; productivity dividends arrive with a lag measured in years. Railroads, electrification, and the early internet all inflated measured investment before they deflated consumer prices.

AI may follow the pattern at unprecedented scale. Hyperscalers are signing multi-year power purchase agreements, ordering GPU clusters before model revenues fully materialize, and competing with Bitcoin miners pivoting into HPC hosting for grid access. The spending is real today. The disinflationary payoff requires models to automate large swaths of white-collar work — a outcome Warsh believes in but cannot schedule.

For investors, the trap is treating “AI is deflationary eventually” as “the Fed will cut in 2026.” Goldman’s upgrade suggests the opposite near-term signal: sticky investment demand that keeps policymakers patient while inflation prints remain above target. The inflation markets lesson applies — breakevens and real yields adjust to the capex channel, not just gasoline and wages.

What to watch at Warsh’s first FOMC

Four markers will tell you whether the $800 billion figure enters the formal policy narrative or stays in sell-side notes:

• Statement language. Compare the June statement to April’s on “economic activity,” “business fixed investment,” and “financial conditions.” Any explicit nod to tech capex would be new.
• The dot plot. A median shift from one 2026 cut to zero cuts would confirm the hawkish repricing; markets are already leaning that way.
• Press conference framing. Does Warsh emphasize AI productivity (dovish long run) or investment-led demand (hawkish near run)? The balance matters more than a single adjective.
• Balance-sheet guidance. Warsh has historically favored faster runoff. Faster passive tightening plus held rates is a double drag on liquidity-sensitive assets.

None of this requires a June hike. The base case remains hold with hawkish guidance — exactly the environment where rate-sensitive portfolios shorten duration and cut beta.

Bottom line

Goldman’s $800 billion AI capex forecast is not a equity research curiosity. It is a macro input: a reason business investment may run hot even as consumer budgets strain, a competing explanation for sticky inflation alongside Hormuz oil, and a headwind to the rate cuts crypto markets spent eighteen months anticipating. The AI trade and the Fed trade have converged. Until productivity shows up in the data with a timestamp policymakers trust, the buildout will look less like a miracle and more like demand — expensive, urgent, and inflationary on the way in.

Sources: Longbridge — Goldman Sachs capex upgrade; CryptoSlate — AI spending and Fed policy; San Francisco Fed — June 2026 FedViews; PrimeRates — Warsh FOMC preview. Related on Solana Garden: Fed FOMC June 2026, AI chip selloff, Bitcoin ETF outflows, inflation markets guide.