Anthropic begins early work on custom AI chips, eyes Samsung as partner
The startup is weighing Samsung’s advanced manufacturing and packaging as AI developers broaden their hardware options, while Anthropic keeps Amazon, Google and Nvidia chips at the center of its compute strategy.
Anthropic is in talks with Samsung Electronics about producing a custom AI chip, adding to a wider industry push by major AI developers to reduce dependence on Nvidia’s dominant hardware. The discussions are focused on Samsung’s 2-nanometer manufacturing process and packaging capabilities, and Anthropic has hired Clive Chan, who was part of OpenAI’s early custom-chip team. The company’s chip plans remain early, however, and Anthropic has indicated that Amazon Web Services’ Trainium, Google tensor processing units and Nvidia graphics processors will remain central to its compute strategy. The report also says Anthropic is considering hardware from Microsoft and U.K. startup Fractile, underscoring a multi-supplier approach as AI companies seek more control over cost, supply and performance.
- custom AI chip: A processor designed for a company’s own artificial intelligence workloads rather than bought off the shelf.
- packaging: A chipmaking process that connects processor components and helps determine performance, power use and integration.
- 2-nanometer process: An advanced semiconductor manufacturing technology used to produce smaller, more powerful and more efficient chips.