Anthropic is moving deeper into the hardware stack. The company has confirmed that it is hiring a “custom silicon team” to design chips for running its AI models, placing it among the frontier AI companies trying to gain more control over the compute systems behind their products.
The move does not mean Anthropic plans to rely only on its own chips. A company spokesperson said Anthropic will continue with a “multi-chip approach,” using hardware from other companies alongside designs developed in-house.
What Anthropic has confirmed
The clearest signal came through hiring. Business Insider noticed a job listing for a senior engineer with experience shipping semiconductor designs. Anthropic’s job board also shows listings for a Silicon Engineer and a Technical Program Manager, Silicon.
After those listings drew attention, a spokesperson for Anthropic confirmed the company’s plans to Business Insider and TechCrunch. The confirmation turns a circulating rumor into a clearer statement of intent: Anthropic wants more internal silicon expertise as it scales its AI systems.
The company has already worked with partners on certain hardware. What changes now is the decision to bring more of that chip knowledge inside Anthropic itself. That matters because the company says its teams will co-design new hardware and models side by side.
Why AI companies want their own chips
The source gives two main reasons AI providers are pursuing custom silicon. The first is dependence. Much of the AI industry relies heavily on Nvidia hardware to run its models, and that creates a strategic vulnerability for companies competing in a market where compute infrastructure is scarce and highly contested.
When demand keeps exceeding current capacity, access to hardware becomes more than a procurement issue. It shapes how quickly companies can train, deploy, and improve their models. For a company like Anthropic, building silicon expertise is a way to reduce reliance on a single dominant supplier while keeping multiple hardware options open.
The second reason is performance. Chips designed around specific models, and models designed with specific chips in mind, could work better together than a more generic setup. That is the promise of tighter coordination between hardware and AI model development.
In practical terms, this is why the phrase “co-design” matters. Anthropic is not just looking at chips as equipment it buys after a model is finished. It is describing a process where hardware and model teams influence each other earlier.
Anthropic is following a wider industry pattern
Anthropic is not the only AI company taking this path. OpenAI recently announced a custom chip called Jalapeño, designed for large language model inference in data centers. OpenAI partnered with Broadcom to develop that chip.
Other major AI players have also moved in this direction. Google has been running its models on its own hardware for awhile. Meta has designed and deployed its own chips. Mistral is reportedly looking into doing the same.
The pattern is clear: leading AI companies do not want compute strategy to be entirely external. As models grow and infrastructure becomes more competitive, chip design is becoming part of the broader AI race.
- Anthropic is hiring for a “custom silicon team.”
- OpenAI has announced Jalapeño with Broadcom.
- Google has been using its own hardware for its models.
- Meta has designed and deployed its own chips.
- Mistral is reportedly exploring a similar direction.
What this could mean for Anthropic
For Anthropic, custom silicon could eventually become a competitive advantage. If the company can align its models and hardware more closely, it may be able to improve how its systems run. That could matter as frontier model providers compete not only on model quality, but also on the infrastructure that makes those models practical to operate.
The source also points to another pressure: software developers and other users are beginning to explore cheaper, smaller, or open-weight models that can run on their own hardware or on edge devices. In that environment, frontier AI companies may need additional advantages to justify their larger systems.
Custom chips do not automatically solve that challenge. But they may help companies like Anthropic improve the economics, performance, or flexibility of running large models. The key idea is control: more control over hardware choices, more control over model-hardware design, and less exposure to bottlenecks in the broader chip market.
Benefits will take time
Anthropic’s plans are still early. The company is hiring key team members, which means any user-facing benefits are not imminent. Designing chips, coordinating them with model development, and bringing that work into real infrastructure is not presented in the source as a near-term change.
For now, the important development is strategic. Anthropic has confirmed that custom silicon is part of its roadmap, while also saying it will keep using a mix of hardware. That combination suggests the company wants optionality: its own designs where they make sense, and partner or third-party hardware where that remains useful.
The AI race is often described through models, benchmarks, and product releases. This announcement is a reminder that the contest also runs through the hardware beneath those systems. Anthropic’s custom silicon team is not just a hiring plan; it is a sign that chip strategy is becoming central to how frontier AI companies scale.