Skip to content
Türkçe

Anthropic · Chips and hardware · Nvidia

Every big lab has its own chip. Anthropic is last.

Published: Updated: 12 sourcesTürkçe

This story was updated after publication (Oct. 9, 2026). Details

On August 5, Anthropic confirmed publicly for the first time that it’s building an in-house team to design custom chips for Claude.

The listing pays $320,000 to $485,000. Its language is unusual: candidates must have “shipped silicon” and be “comfortable making consequential calls without a large organization behind them.”

That describes a small, very senior core, not a conventional chip org.

The real signal isn’t the listings, it’s a name: Clive Chan.

Chan was the second hardware engineer ever to join OpenAI’s custom chip team. He spent 2.4 years on “Jalapeño,” the inference chip OpenAI co-designed with Broadcom. Before that, Tesla’s Dojo supercomputer program.

He moved to Anthropic in June and posted on X: “It’s time to build.”

The cost arithmetic behind it is steep. By Morgan Stanley’s estimate, a single Nvidia Vera Rubin GPU runs about $55,000, and a 72-GPU rack about $7.8 million before infrastructure. Anthropic’s annualized revenue passed $47 billion in May.

Amodei’s own warning: “What if the country of geniuses comes, but in mid-2028 instead of mid-2027? You go bankrupt.”

There are reasons for caution, though. Getting a chip from design to working silicon runs 18–24 months in this category, and production looks unlikely before 2028. Google took nearly a decade to turn TPUs into a real cost advantage.

Sources

  1. Business Insider, “It’s official: Anthropic is building an in-house chip team for Claude”, (businessinsider.com)
  2. Anthropic (Greenhouse), “Job Application for Silicon Engineer at Anthropic” (job-boards.greenhouse.io)
  3. Clive Chan (@itsclivetime), post on X announcing his move from OpenAI to Anthropic, (x.com)
  4. Clive Chan (@itsclivetime), post on X announcing his move from Tesla Dojo to OpenAI, (twitter.com)
  5. OpenAI, “OpenAI and Broadcom unveil LLM-optimized inference chip”, (openai.com)
  6. Tom's Hardware (Morgan Stanley Research), “Nvidia’s memory costs soar 485%, latest AI systems now cost $7.8 million to build — memory now comprises 25% of the total cost, Rubin GPUs a mere $50,000 apiece”, (tomshardware.com)
  7. Anthropic, “Anthropic raises $65B in Series H funding at $965B post-money valuation”, (anthropic.com)
  8. Dwarkesh Podcast, “Dario Amodei — “We are near the end of the exponential””, (dwarkesh.com)
  9. Google Cloud, “Google supercharges machine learning tasks with TPU custom chip”, (cloud.google.com)
  10. Amazon Web Services, “Announcing Amazon EC2 Trn3 UltraServers for faster, lower-cost generative AI training”, (aws.amazon.com)
  11. Microsoft, “Maia 200: The AI accelerator built for inference”, (blogs.microsoft.com)
  12. Meta, “Expanding Meta’s Custom Silicon to Power Our AI Workloads”, (about.fb.com)

Corrections and updates

  1. The version first posted on Instagram said a 72-GPU rack clears $8 million before infrastructure. By Morgan Stanley’s estimate, the rack costs about $7.8 million; the text has been corrected.
  2. On August 17, the day this story was published, Bloomberg reported that Anthropic’s annualized revenue had topped $65 billion as of the end of July.

About this story

This story was posted on Instagram by @jarrus.tech on Aug. 17, 2026.

Spotted an error in this story? [email protected] · Instagram

Short link: thejarrus.com/en/anthropic-chip

This story in Turkish: Anthropic çip yarışına katılan son büyük lab oldu

On the same topic