Anthropic planned, then abandoned $7 billion purchase of MatX

Anthropic, a leading artificial intelligence lab, had planned to purchase AI chip startup MatX for approximately $7 billion, aiming to bolster its efforts in developing custom hardware for its rapidly expanding AI business. According to two sources briefed on the matter, the talks with MatX have since evolved into discussions about a potential partnership. This move underscores Anthropic's ambition to secure the resources and talent necessary to build its own chips, thereby accelerating its in-house chip development.

Founded by former Google tensor processing unit (TPU) engineers, MatX is now seeking to raise new capital at a valuation of around $4 billion, as reported by Reuters. The discussions on the potential acquisition were for the first time reported by Reuters, but the reason behind the abandonment of the talks remains undisclosed. MatX declined to comment on the deal talks with Anthropic, while MatX also did not respond to a request for comment.

As Anthropic scales its Claude family of AI models, the company is looking to generate its own hardware to cater to the high demand for data processing and reduce its dependence on chips produced by Nvidia. By hiring engineering and executive talent, Anthropic is accelerating the chip design process, which could take years. The company aims to list this year with a valuation of $2 trillion, driven by a projected 2028 revenue of up to $200 billion.

Anthropic's plans to expand its in-house silicon team to design custom chips for faster and more efficient Claude models are well-documented. The company also plans to maintain a multi-chip approach, working with providers from Nvidia to Google. While designing chips is a costly and time-consuming process, taking a year or more to produce a viable piece of hardware with design costs running into hundreds of millions of dollars, an acquisition such as MatX would provide Anthropic with in-house design expertise and potentially lower costs in the long run.

In addition to its discussions with MatX, Anthropic has held meetings with various AI chip startups. The company has not yet decided to make an acquisition but is exploring the range of chip design approaches. Recently, Anthropic hired Google chip veteran Amir Salek to bolster its hardware efforts, and in June, it brought aboard former OpenAI chip engineer Clive Chan.

The leading AI labs, including Anthropic and OpenAI, are increasingly focusing on custom chips tailored to their models and workloads to achieve performance and economic advantages. OpenAI's first custom chip, Jalapeno, recently outperformed a similar processor from Nvidia in terms of energy efficiency during inference calculations.