Anthropic, the company behind the Claude AI assistant, came close to purchasing AI chip startup MatX for approximately $7 billion before the discussions stalled, according to a report from Reuters cited by Channel Insider. The acquisition talks are no longer proceeding, with negotiations shifting toward a possible partnership instead. The episode reveals how seriously Anthropic is considering custom silicon as it seeks greater command over computing infrastructure and costs.
The reported $7 billion price tag is striking because MatX is now pursuing funding at a valuation around $4 billion, the report notes. MatX, established in 2023 by former Google engineers Reiner Pope and Mike Gunter who previously worked on TPU hardware and software, is building processors designed specifically for training large AI models. The startup secured $500 million in Series B funding in February. Reuters couldn't identify why the acquisition collapsed. Meanwhile, Anthropic is expanding an internal chip development team and recently brought on senior hardware talent, including Google chip veteran Amir Salek last week and former OpenAI chip engineer Clive Chan in June.
The MatX discussions are especially revealing because the startup concentrates on training chips, the computationally demanding hardware used to build AI models from scratch. According to the report, this contrasts with OpenAI's strategy, whose newly announced Jalapeno chip targets inference work—running already-trained models to produce responses. Reuters indicated that Anthropic might eventually pursue inference hardware too, but the MatX talks suggest training is a major priority. The report states that developing viable hardware can require a year or longer, with a single chip generation potentially costing hundreds of millions of dollars. Acquiring MatX would have offered a quicker path to specialized capabilities rather than constructing every skill internally.
For companies running models at Anthropic's scale, chips are no longer simply infrastructure bought from vendors—hardware design can become part of the product strategy itself, the report explains. Anthropic isn't abandoning Nvidia, with Reuters saying the company plans to maintain a multi-chip approach involving suppliers including Nvidia and Google. Still, its growing hardware ambitions could serve as protection against chip shortages and mounting computing demand, especially since Nvidia said its processors could stay in short supply through 2027. Even modest efficiency improvements can turn into substantial savings when multiplied across enormous accelerator clusters, the report notes.
The MatX acquisition didn't materialize, but Anthropic's reported interest in the startup, hiring of chip veterans, and conversations with other AI chip companies demonstrate that custom hardware is becoming a progressively more vital element of the Claude maker's strategy. The company is already committing massive amounts to computing infrastructure, including plans to purchase $36 billion worth of Google's AI chips, a $45 billion agreement to lease AI cloud capacity from Nscale, and a commitment to pay SpaceX $1.25 billion monthly through May 2029 for computing capacity. The fundamental tradeoff is velocity versus autonomy: Anthropic can keep purchasing processors for immediate access to proven technology, or invest heavily in custom silicon that might eventually deliver superior economics but carries significant engineering and financial risks. Companies that master both the software and hardware stack may reshape competitive dynamics in ways that current cloud-dependent rivals will struggle to match, particularly as computing costs threaten to outpace revenue growth in foundation model development.