Anthropic just bought $3.4B of power generators
Today's biggest story is about power and hardware, not models. Anthropic is reportedly spending $3.4B on generators for its data centers, which shows how much physical capacity the big labs now need. If you rent compute, watch the cost and supply of it closely, because the suppliers you depend on are also under scrutiny.
- Anthropic's $3.4B power buy
- Nvidia's board probes Jensen Huang's ties
- Is the AI bubble overblown?
- Quick hits: launches and a policy story
- AI Index: down 22.1% in 30 days
Top stories
All news →What happened: The Financial Times reported that Anthropic agreed to pay $3.4B for gas-fired generators to power its data centers.
- The package covers 1.7 gigawatts of large generators, based on one source's description.
- OnePWR had ordered the machines from Caterpillar about two years ago but needed money to finish the purchase. Anthropic signed in late September with OnePWR Solutions in San Antonio, Texas.
- Anthropic did not buy OnePWR and none of its staff moved over. The FT called it the biggest known single buy of power kit by an AI company, and it is not known where the generators will be used.
Why it matters: Power and physical equipment are becoming a limit on how much AI can be built, not just chips. Founders who rent compute should expect capacity costs and supply to be shaped by buyers at this scale.
What happened: Nvidia's board will run an independent review after reports that Huang was the target of a foreign intelligence operation.
- The review runs through a board committee working with outside counsel. Nvidia says it has no evidence its secrets were compromised and that Huang is cooperating.
- The episode involves Alice Huang, a Megaspeed executive that U.S. intelligence has linked to China's Ministry of State Security.
- Congressional committees are preparing for a briefing from Huang on November 18, and some Republicans have demanded that he testify.
Why it matters: Nvidia supplies much of the compute that AI startups rent, so any change in leadership, licensing or China sales could affect your pricing and supply. Governance and export-control scrutiny can now reach the suppliers the whole AI stack depends on.
What happened: Independent engineer Matteo Raso argues that an AI bubble is not settled, and that weak finances at some AI companies would not mean the bubble bursts.
- He concedes some AI companies may have poor finances, but says their failure would not amount to a bubble bursting.
- If AI is a bubble, he expects it to resemble the internet bubble, when the core technology kept its usefulness.
- Anthropic's public S-1 filing, made October 26, shows $16.3B of first-half 2026 revenue and an $80.7B net loss.
Why it matters: Fundraising conversations depend on what investors believe, and an S-1 (the filing a company makes before selling shares publicly) gives you hard numbers to read. Raso's point is that a technology can outlast weak balance sheets, which matters when you pick where to build.
Quick hits
- The AP reports the US AI safety center's staff were asked to report informally to the White House. Unite.AI (citing Associated Press) →
- Sage2 offers APIs that pull dates, amounts and durations from live conversations, so apps can decide their next step. Product Hunt →
- DocsAlot's bot answers customer questions in Slack and Discord from a company's existing knowledge base. Product Hunt →
- Thinking Machines Lab released Inkling 2 Small, a 594B-parameter model with 17B active, built for coding and agents. Product Hunt →
- AXL combines sales, marketing and AI lead scoring in one tool, trained on 1M deals from its users. Read →


