Key Points:
- Anthropic is looking into designing its own AI chips to beat industry hardware shortages.
- The company hit a $30 billion revenue run rate in 2026, tripling its 2025 revenue.
- Anthropic recently committed $50 billion to build up U.S. computing infrastructure.
- Designing a new AI chip costs businesses around $500 million to cover engineers and testing.
Artificial intelligence startup Anthropic is looking into designing its own computer chips. Three people familiar with the matter shared this information, highlighting how the company wants to secure enough hardware to build and run its advanced AI systems. Right now, the tech industry faces a major shortage of the specific chips that power complex artificial intelligence models. As demand grows, Anthropic and its competitors are seeking new ways to keep their servers running without waiting in line for hardware.
The chip design plans remain in the very early stages. Two people close to the situation said Anthropic might still decide to simply buy existing chips instead of building custom ones. One source added that the company has not yet committed to any specific design. Anthropic has not created a dedicated team for this hardware project. When asked about these details, a spokesperson for the San Francisco startup declined to comment.
This hardware push coincides with Anthropic’s explosive financial growth. Demand for the company’s AI chatbot, Claude, skyrocketed throughout 2026. Earlier this week, Anthropic announced its run-rate revenue crossed the $30 billion mark. This represents a massive jump from the $9 billion the company reported at the end of 2025. Because more people and businesses use Claude every day, the company needs significantly more computing power to keep the service fast and reliable.
Currently, Anthropic relies on hardware from other massive tech companies. The startup uses a variety of processors to train and operate its software. These include tensor processing units from Google and custom processors from Amazon. Anthropic runs its models on these third-party platforms to ensure Claude has the processing power it needs to answer user questions and analyze large amounts of data.
To secure its current hardware supply, Anthropic signed a major long-term deal with Google and Broadcom earlier this week. Broadcom actively helps companies design and build custom tensor processing units. Through this partnership, Anthropic ensures it has enough computing resources for the near future. The company also stated that this deal supports its broader promise to spend $50 billion to strengthen computing infrastructure across the United States.
Anthropic is not alone in wanting to create custom hardware. Many large technology companies are actively designing their own artificial intelligence chips. Industry rivals like OpenAI and Meta are leading similar projects. These companies realize that relying entirely on outside chipmakers limits their growth and leaves them vulnerable to supply chain issues. By creating their own hardware, these tech giants hope to control costs and tailor chips to their specific software needs.
However, designing a custom artificial intelligence chip requires massive financial resources and technical expertise. Industry insiders estimate that a company must spend about $500 million just to design one advanced chip. Businesses have to hire highly skilled engineers and pay for extensive testing. They must ensure the manufacturing process produces zero defects, as even tiny mistakes can ruin a batch of expensive silicon.
For Anthropic, taking on this $500 million challenge could make sense given its recent $30 billion revenue milestone. The startup certainly has the money to fund a hardware division. Over the next few months, company leaders will have to decide whether building custom chips offers enough benefits to justify the massive price tag and the long development time. If they move forward, Anthropic will transform from a pure software business into a heavy-hitting hardware player in the artificial intelligence market.
