The AI Boom Has Become a Scissors Crisis

Federal Reserve Chairman Kevin Warsh has described artificial intelligence in terms no one on his staff disputes — as a historic productivity wave. Where they break is on what it is doing to prices right now.

a significant disinflationary force — Kevin Warsh

The June minutes his own staff produced identify AI infrastructure spending as a primary driver of rising prices, with PCE inflation climbing to 4.1% and the FOMC divided on whether rate hikes are now necessary ^[1].

Many participants noted that ongoing strong demand for AI infrastructure would likely sustain upward pressure on prices for technology products and electricity. — Federal Reserve System

The disagreement is not academic — it determines whether the Fed raises rates into an economy that its own Monetary Policy Report describes as "heavily dependent on AI-related capital spending, while housing and consumer spending remain soft" ^[2]. The contradiction is real because AI is doing two things at once, and they cut in opposite directions. On the input side, the build-out is inflationary. RAM prices have spiked 290% as data centers lock up chip supply. Computer software and accessories prices surged 14.5% year-over-year in May — the largest jump since 2000 ^[3]. Fed Governor Lisa Cook and New York Fed President John Williams have cited $1.5 trillion in announced data center plans as already forcing Apple and Microsoft to raise prices on laptops, iPads, and Xbox consoles ^[4]. The AI boom is making consumer electronics more expensive, not less. On the output side, the product of all this spending — AI tokens — is being commoditized at a speed that makes the infrastructure investment look reckless. In May, DeepSeek permanently cut prices on its V4-Pro model by 75%, to 0.025 yuan per million tokens. The model costs 12 to 19 times less than OpenAI's GPT-5.5 and Anthropic's Claude Opus 4.7 for equivalent tasks, and the floor is set by the cost of Huawei Ascend 950 chips, not by a promotional discount that expires ^[5].

It is an efficiency gain being passed through. — Sanchit Vir Gogia

Chinese models now process roughly four times more global tokens each week than US alternatives — 12.96 trillion versus 3.03 trillion — and all six of the most-used models globally are now Chinese ^[6]. US enterprises have noticed. Pinterest and Airbnb are adopting Chinese open-source models to cut costs, with Pinterest reporting its in-house models are 30% more accurate than leading proprietary alternatives ^[7]. The price of intelligence is collapsing, and the US firms that spent hundreds of billions to produce it are watching their pricing power evaporate. The investment bank Jefferies has reached a blunt conclusion.

The reason credit risk has become more of an issue is that it can no longer be assumed... that large language models will ever be profitable given the related ongoing collapse in token pricing. — Jefferies Group

The two blades meet in a single income statement. Baidu's AI revenue grew 49% and overtook legacy search as the company's primary business — and net profit fell 55% ^[8]. Alibaba's cloud revenue grew 34%, with AI product revenues growing at triple-digit rates for nine consecutive quarters — and GAAP net income fell 53%, while free cash flow turned negative at -21.8 billion yuan ^[9]. Revenue is migrating to AI, and margins are collapsing under it. The four largest US hyperscalers are projected to spend $695 billion in 2026 and $870 billion in 2027 on infrastructure whose output they cannot price ^[10]. This is where the scissors close on the central bank. The rate hikes that could contain AI-driven inflation would also collapse the growth engine the economy has come to depend on. Royal London Asset Management described the mechanism in plain terms ^[11].

You need a pin that pricks the bubble and it will probably come through tighter money. — Royal London Asset Management

The market has already begun pricing the outcome. Hyperscaler stocks that are deploying AI capital have declined — Microsoft down 23%, Meta down 13% — while the picks-and-shovels suppliers that sell into the build-out have surged: Vertiv up 62% ^[12]. Broadcom's custom AI chip sales rose 65% to $20 billion. At the application layer, the collapse is starker: BigBear.ai saw revenue decline and C3.ai's net losses widened to $289 million ^[13]. Capital is fleeing the deployers and the apps and concentrating in the suppliers — exactly the pattern you would expect if the market has concluded the middle of the stack cannot earn back what it is spending.