There is a new direction to Anthropic’s efforts to develop its own AI chips.
The company behind Claude had been in discussions to put in an offer of some $7 billion for the chip startup MatX, say those in the know. But those acquisition overtures are off the table now; instead, the two sides are said to be in talks about a partnership. It is a different approach but one that still serves Anthropic’s overriding objective: to have greater command of the hardware that runs its ever-expanding AI workloads.
Observers in the industry are paying attention to the matter. With the expense and compute needed to train and run big models on the upswing, frontier AI firms are less content with off-the-shelf solutions.
Why MatX Was on Anthropic’s Radar
MatX was an attractive target for an acquisition. Put together by ex-TPU engineers from Google, the firm is making chips built for the task of training large-scale AI. Bringing MatX in would have given Anthropic a head start on its custom silicon and weaned it off suppliers like Nvidia. There is also the question of optimisation; with their own hardware, Anthropic could tailor chips to the demands of Claude for better performance and cost control down the line. The lab has already been putting more resources into an in-house silicon team for this purpose.
From $7 Billion Acquisition to a Potential Partnership
What was a possible $7 billion deal has become something else. Sources indicate the conversation has turned to collaboration, though Reuters has not been able to pin down what caused the acquisition to fizzle, and neither party has put any confirmation of the terms out there.
MatX Raises Capital as Its Valuation Takes a Different Turn
It is an interesting turn of events for MatX, which is looking to raise fresh capital at a $4 billion valuation, well under the sum once mooted for an acquisition. The startup has already pulled in a $500 million Series B with Marvell among the backers.
Why Anthropic Wants More Control Over Its AI Hardware
In the end, a partnership may be enough for Anthropic to tap into MatX’s design expertise and speed up its hardware plans. For MatX, having a relationship with a top-tier AI lab as it goes to market for funding is no small thing.
Yet the story is indicative of a larger trend in the chip race. As the requirements for inference and training mount, it is not just a matter of who can put together the finest model. Control of the underlying economics and infrastructure is becoming just as important. Anthropic’s dealings with MatX are a case in point, whether they culminate in a partnership or a multi-billion dollar buyout.
