# Anthropic Reportedly Wants to Make Its Own AI Chips for Claude > Source: > Published: 2026-08-05 16:19:24+00:00 Following in the footsteps of Google, Meta, Amazon, and OpenAI, Claude developer Anthropic is reportedly building a new team to design its own custom AI chips, Business Insider [reports](https://www.businessinsider.com/anthropic-in-house-silicon-chip-team-claude-2026-8). For most companies designing their own chips, this solves two key problems: a demand for chips that has completely outstripped supply and a need to run AI tasks as efficiently as possible. As [Anthropic's services have grown more popular](/ai/165639/the-us-government-banned-anthropics-fable-5-ai-i-tried-it-before-it-disappeared), demand has increased dramatically. So, like other AI companies before it, Anthropic is getting into the hardware game and putting together a team for it. It told Business Insider that it would "co-design" the hardware and models, with previous [rumors](https://www.theinformation.com/articles/anthropic-talks-samsung-manufacture-custom-ai-chip) pointing to Samsung Electronics as a potential manufacturing partner. Anthropic says it will take a "multi-chip" approach to AI inference and will continue to use hardware from cloud providers such as Amazon's AWS and Google Cloud, leveraging Nvidia and AMD chips. It seems clear that any chips it makes will be only for inference. If the cutting-edge Chinese developers challenging Anthropic and OpenAI use smuggled and otherwise-blocked Nvidia chips for training, you can bet companies like Anthropic will use them too, given their far easier access to the best of the best. The engineer position that Anthropic is looking to fill says any potential candidates must demonstrate "direct personal contribution" to the final states of shipping semiconductor designs. Although it suggests some semblance of autonomy, this job is about getting a chip design over the line on schedule. Time is the ever-present pressure of the rapidly evolving AI industry. Internal, bespoke chip designs appear to be the way the industry is going, at least for running the models. In China's upstart labs, the ruling party mandates heavy use of domestic chips, and the growth of models like DeepSeek and Kimi K3 suggests that optimizing for domestic hardware may be the best solution. [Google continues to develop its TPUs](https://www.pcmag.com/encyclopedia/term/tensor-processing-unit) to power its own chips, and Meta has a new AI chip design that is expected to enter production in September. French firm Mistral is also said to be [considering developing its own silicon chips](https://www.cnbc.com/2026/05/28/mistral-arthur-mensch-design-chips-ai-data-centers.html). Although designing chips is expensive up-front, the potential efficiency and economic advantages of having an in-house design that your models are optimized for may well outweigh the costs. Especially if you have the kind of money these AI developers have to throw around.