Anthropic hires team to design custom AI chips
▼ Summary
– Anthropic is building a team to design custom chips for AI usage, aiming to co-design hardware and models for faster, more efficient performance.
– The company has existing AI hardware deals with AWS, Google, Nvidia, and AMD, but scaling to meet demand requires moving beyond reliance on external partners.
– Anthropic was reportedly scouting Samsung as a potential partner for chip production, per a prior report from The Information.
– Anthropic is not alone in this move, as OpenAI unveiled its Broadcom-built Jalapeño chip in June, while Google DeepMind uses TPUs and Meta develops its own MTIA accelerators.
– The company is hiring engineers with chip design experience for its “custom silicon team,” according to a job listing.
Anthropic is quietly assembling a dedicated engineering team to design its own custom AI chips, a move that signals the company is looking to reduce its reliance on external hardware providers as demand for its Claude models skyrockets. The Claude maker confirmed it is exploring ways to co-design hardware alongside its software models, a strategy aimed at boosting performance and cutting operational costs.
The initiative, first reported by Business Insider, comes on the heels of a separate report from The Information last month indicating that Anthropic had approached Samsung about a potential partnership for chip manufacturing. If the plans materialize, Anthropic would join a growing list of AI labs that have concluded that off-the-shelf processors are no longer sufficient for their long-term scaling goals.
Anthropic has already locked in major infrastructure agreements with AWS, Google, Nvidia, and AMD to secure access to high-end AI computing capacity. Yet the company appears to believe that relying on third-party suppliers alone will not be enough to keep pace with the explosive growth in AI adoption. Building its own silicon would give Anthropic greater control over the intersection of hardware and software, potentially unlocking gains in speed and energy efficiency that generic chips cannot deliver.
The company is not breaking new ground here. OpenAI unveiled its own custom inference-focused processor, the Broadcom-built Jalapeño chip, back in June. Google DeepMind has leaned on Alphabet’s TPU family for years to train and run its models, while Meta has been quietly developing its own MTIA accelerators to handle AI workloads internally.
Anthropic is now actively recruiting engineers with chip design expertise for its newly formed “custom silicon team,” according to a current job posting. The company did not respond to a request for comment in time for publication.
The push toward in-house chip development reflects a broader industry trend: as AI models grow more complex and compute demands surge, the companies building them are increasingly looking to control their own hardware destiny. For Anthropic, the stakes are high. Claude’s rising popularity has made it one of the most sought-after AI assistants on the market, and any edge in inference speed or cost per query could translate into a meaningful competitive advantage.
(Source: TechCrunch)
