Warsh Tied the AI Boom to a 'Shock' in Bangkok. The Bond Market Can't Easily Unwind What the Fed Just Named

Federal Reserve Chair Kevin Warsh used his Oct. 16 'fireside chat' with IMF Managing Director Kristalina Georgieva at the IMF-World Bank Annual Meetings in Bangkok to make explicit what the September FOMC minutes had only implied six weeks earlier. AI is not just transforming the economy — it is now a named force inside the Fed's reaction function. 'If I look back over the last few months, it was a synchronous tightening in monetary policy in most places around the world,' Warsh told Georgieva. 'And there are no free riders.' He called AI 'the most consequential change' to the U.S. and global economy in his lifetime and warned the path would be 'very bumpy' — language that the WSJ and CNBC read as an explicit acknowledgement of short-term inflation pressure.

The Fed had already moved. The September 16, 2026 FOMC raised the funds rate 25 basis points to 3.75–4.00% — its first hike since2023 — and the minutes released October 7 said participants viewed 'surging AI-related investments' as contributing to inflation pressures. The staff review attributed part of core inflation to 'an increase in technology-related consumer goods prices associated with the AI buildout.' What Bangkok added was international scope: 'rich countries face the same inflation fight,' Warsh said, and central banks had moved in lockstep. The implication for fixed income is that the long end is no longer just absorbing short-rate expectations — it is absorbing a structural demand for capital that the Fed cannot fully drain.

Read carefully, the September FOMC minutes name the mechanism. Under the heading of market developments, staff flagged that term premiums and yields were partly attributable to 'competition for capital … to finance the development of artificial intelligence (AI) infrastructure.' The minutes also discussed 'AI-related hyperscaler debt conditions.' That language — written by Fed staff, approved by FOMC participants — is the first time the central bank has formally identified the AI capex cycle as a term-premium driver rather than a one-off CPI component.

The capital intensity Warsh now has to lean against did not appear in2026 — it was assembled over the prior two years and is now compounding. Alphabet (Class C shares trading as GOOG), Microsoft, Meta Platforms and Amazon.com together projected roughly $700–725 billion in 2026 capital expenditures per a Forbes tally on Oct. 7. Futurum estimated Alphabet at $175–185 billion, Meta at $115–135 billion, Microsoft at $120 billion-plus, and Oracle at $50 billion. Goldman's AI capex model runs $765B in 2026, climbing to $1.6 trillion by 2031 — a cumulative $7.6 trillion from 2026 to 2031.

These are not forecasts — they are procurement pipelines already under contract with NVIDIA, Taiwan Semiconductor Manufacturing (TSM), SK Hynix, and the U.S. merchant power complex. The 10-year Treasury at 5.09% reflects the fact that $200+ billion of incremental hyperscaler debt and project finance will compete with Treasuries for the same marginal dollar over the next 12 months.

The supply chain feeding AIGAI capex has consolidated around four listed chokepoints: AI silicon, HBM memory, foundry capacity, and dispatchable power. Warsh's 'synchronous tightening' implies the inflation pressure — and the revenue captured by these chokepoints — extends for as long as the AI capex cycle holds its current slope.

Above the silicon sits the power bottleneck — and that is where Warsh's framework breaks down. The IEA's latest Energy and AI report projects data-center electricity consumption will roughly double to 945 TWh by 2030, with U.S. data-center demand growing 130% over the same window. Constellation Energy, Vistra, GE Vernova, and Vertiv — the cooling and power-conversion chokepoint — sit on top of a generation capacity stack that cannot be expanded on hyperscaler timelines.

The next two FOMC meetings — Oct. 28–29 and Dec. 16–17 — are the immediate catalysts. Warsh explicitly declined to 'prejudge' either meeting in Bangkok, but the September dot plot shows the median FFR projection at 4.125% for both 2026 and 2027, implying one more hike is in the consensus. That keeps front-end yields supported and the curve flatter, but the long end is governed by a different variable: AI-driven hyperscaler debt issuance competing with Treasury supply.

Warsh's Bangkok framing implies a bet, not a forecast. He told the WSJ the AI build-out is 'ultimately a positive supply shock, which increases the economy's capacity to produce more goods and services' — but acknowledged the path will be 'bumpy' and the productivity dividend may not arrive for years. The disinflationary payoff, in other words, is back-end loaded. Until it arrives, the Fed is fighting a sector it funds through low real rates, and the bond market is pricing capital scarcity that the Fed cannot reverse by moving the funds rate alone.

The supply chain sits in the middle of this bet. NVIDIA's $89.0 billion Data Center line is the prime beneficiary while hyperscalers keep ordering; Taiwan Semiconductor is foundry-monopoly rent collector with limited downside if Vera Rubin ramps as guided; Constellation Energy and GE Vernova sell into a power demand curve the IEA says more than doubles by 2030. The bond-market hedge problem is that none of these balance sheets can be hedged against a productivity shock that doesn't arrive — which is precisely the risk he put on the table in Bangkok.