Anthropic is thinking about designing its own AI chips as more and more people are using Claude. People with knowledge of these discussions say the company has looked at possibilities, but hasn’t actually started designing anything or put together a team of chip experts. When asked for a comment, someone who speaks for the company wouldn’t say anything.
Rising demand and strategic motivation
Claude has become really popular in 2026, and the amount of money Anthropic is making is increasing quickly. Because of this growth, the company needs to find a way to get a lot more computer processing power and to control how much it costs to get chips from other companies.
Making their own chips would allow Anthropic to get the best possible performance from their large language models, and reduce the cost of each time Claude answers a question. Also, if they owned the design for the chips, they would have more control over how many they get and how their computer systems will work in the future.
Current chip supply and partnerships
Right now, Claude runs on a combination of hardware from outside sources, including TPUs and processors from the biggest cloud companies. These relationships allow Anthropic to get a lot of computing power, but they also mean the company could be affected by problems with manufacturing and the plans of the companies making the chips.
Previously, Anthropic made a long-term deal with a company that designs TPUs and another one that actually makes the chips. Anthropic has also promised money to improve computer infrastructure in the United States, which shows they’re planning to both buy and build.
Industry trend toward custom AI silicon
Other big AI companies are creating their own chips to fit the specific needs of their AI models and to avoid shortages. These custom chips can be more efficient when the AI is learning and when it’s providing answers, and they can make one AI service perform better than another.
However, designing a modern AI chip requires a lot of money. Experts in the industry estimate it would cost around $500 million just to start, including the cost of the engineers, the equipment for making the chips, and making sure the chips actually work when they are made.
Technical and financial challenges of chip design
You need teams of specialists in hardware, the basic software of the hardware, and testing, and they all have to work closely with the factories that make the chips and the companies that make the circuit boards. Even little mistakes in the design can cause very expensive re-makes and delays.
Beyond the engineering side, Anthropic would have to get the chips, get them to where they need to be, and it takes a long time to actually have the chips made. The company would need to compare how much it would cost to start this process with how much money they think they would save and how quickly Claude is growing.
Strategic options and potential outcomes
Anthropic could do lots of different things: keep buying from the companies that already make chips, design chips with a partner company, or create a complete in-house team to do all the design work. Each of these paths has different advantages and disadvantages in terms of how quickly they can get something done, how much control they have, and how much money they need.
A safer option would be to strengthen relationships with existing companies and guarantee a supply of chips for the long term, while also putting money into customized “boosters”. A more daring move would be to hire a full team of chip designers to try to get better performance and potentially make money by letting others use their designs.
Implications for the AI hardware ecosystem
If Anthropic does start to make its own chips, it will be part of a trend of AI companies controlling all parts of their business, including the hardware. Having more companies designing chips could lead to more competition and new ideas, and push for chips that are more efficient and use less power.
However, lots of different custom designs could make it harder to have common standards and make it more complicated for cloud companies and data centers to buy what they need. People who make rules and industry groups might have to deal with making sure there are enough chips and that they all work with each other as this change happens.
Next steps and market watch points
For now, Anthropic is still looking at all the options and might actually decide not to design chips. People in the field will pay attention to who they hire, what partner deals they make, and where they put their money to get a better idea of what they’ll do.
Whatever Anthropic decides – whether they build chips or continue to rely on partners – it will affect how much computer processing power is available and how much it costs to run large AI models. This decision will impact their competitors, the companies that run the clouds, and customers who want AI services that are fast and cheap.
