Anthropic’s Amodei Calls for a Slower AI Race, but the Chip Spending Question Remains Open

Anthropic chief executive Dario Amodei used a weekend essay to urge the companies building the most powerful AI systems to deliberately slow the pace at which their models advance. The appeal for a more measured frontier has landed at a sensitive moment for the sector, with stock futures drifting lower Sunday evening as traders parsed the safety message — Nasdaq futures showed the steepest decline of the major index contracts.

The timing is notable for the semiconductor industry, which has spent the past two years riding enormous data center buildouts tied to frontier model development. NVIDIA (NASDAQ:NVDA), the dominant supplier of AI accelerators, entered the weekend under pressure: shares slid roughly 5% last week to around $218 before the essay was published. In Monday trading, the stock was changing hands at $223.67, down 0.73% from its prior close of $225.31, leaving its market capitalization near $5.56 trillion.

NVIDIA’s own outlook underscores how much compute is still being purchased. The company’s guidance calls for $108.0 billion in revenue for the current quarter — a figure that reflects demand from hyperscalers and AI labs racing to train and deploy increasingly capable systems.

The central question raised by Amodei’s essay is whether a voluntary slowdown at the frontier would actually reduce demand for the hardware underneath. A careful reading of the piece suggests it does not call for reduced chip purchasing; it addresses the speed at which model capabilities improve, not the infrastructure spending required to reach or operate them. Even a slower frontier trajectory could still entail vast compute commitments, particularly if labs redirect resources toward safety testing, evaluation, and deployment rather than scaling.

For NVIDIA, that distinction matters. The company has positioned itself not merely as a chip vendor but as a full-stack AI infrastructure provider, with its data center segment spanning accelerators, networking, and software. Any meaningful pullback in frontier training runs would eventually show up in orders — but nothing in the essay asks for that yet.

Anthropic, which is privately held and builds its own frontier models, has long been vocal about AI risk. Amodei’s latest intervention frames pacing as a collective-action problem: unilateral restraint is difficult if competitors keep accelerating, which is why he is appealing to the industry as a whole.

What to watch

• NVIDIA’s upcoming quarterly earnings report and whether results land in line with the $108.0 billion revenue guidance.
• Whether other frontier AI labs publicly respond to or endorse Amodei’s call for a slower pace.
• Hyperscaler capital-expenditure updates in coming earnings seasons, a key indicator of accelerator demand.
• Broader AI policy developments that could translate voluntary pacing appeals into formal commitments.

Source: original release