Anthropic ramps up its self-developed AI chip: It once considered acquiring MatX for $7 billion and is currently in talks with several startups.

Anthropic ramps up its self-developed AI chip: It once considered acquiring MatX for $7 billion and is currently in talks with several startups.

Anthropic is accelerating its chip self-sufficiency strategy, viewing it as a key step in consolidating its competitive advantage ahead of its IPO.

According to a recent Reuters report, Anthropic had been in talks to acquire AI chip startup MatX for approximately $7 billion, but the acquisition plan was ultimately abandoned, and the discussions have since evolved into exploring potential partnerships. Meanwhile, Anthropic has also recently held talks with several other AI chip startups, but no specific path has been finalized yet.

These moves indicate that Anthropic is seeking to reduce its reliance on external hardware suppliers such as Nvidia and establish a long-term advantage in computing power costs and supply security by mastering its own chip design capabilities. For an AI company with a valuation target of $2 trillion and preparing for an IPO, hardware self-sufficiency has become an important part of its business narrative.

Acquisition negotiations take a turn for the worse, and MatX turns to independent financing.

According to the report, two sources familiar with the matter revealed that Anthropic had held in-depth negotiations with MatX regarding a potential acquisition price of approximately $7 billion. A third source indicated that the merger discussions had shifted to exploring a partnership, but Reuters was unable to ascertain the specific reasons why the negotiations had stalled.

MatX, founded by a former Google Tensor Processing Unit (TPU) engineer, is currently seeking to complete a new round of funding at a valuation of approximately $4 billion.

MatX's core technology focuses on training chips—processors used to build large-scale AI models. This differs from the approach taken by some chip startups and competitor OpenAI, which emphasizes inference chips. Sources indicate that Anthropic may also choose to develop inference chips simultaneously.

Extensive outreach to startups, with talent development progressing in tandem.

According to Reuters, Anthropic has held talks with several AI chip startups in recent weeks, a move aimed at helping its engineers and executives gain a comprehensive understanding of the different technological paths in the current chip design field, but the company has not yet made a decision on an acquisition.

In terms of talent acquisition, Anthropic hired Amir Salek, a Google chip veteran, this week, and also recruited Clive Chan, a former OpenAI chip engineer who had participated in the OpenAI chip project, in June of this year.

Anthropic stated that the company is expanding its internal silicon team to design custom chips that will allow the Claude model to run faster and more efficiently, while maintaining a multi-chip supplier strategy and continuing to collaborate with Nvidia, Google, and others.

The strategic logic behind self-developed chips: cost reduction, efficiency improvement, and hedging against supply risks.

Anthropic's push for a custom silicon wafer strategy is driven by multiple real-world factors.

One is cost pressure. Although chip design is extremely expensive—a single generation of design can cost hundreds of millions of dollars, and it often takes more than a year from design to the production of usable hardware—acquiring startups with design capabilities is expected to significantly reduce computing power costs in the long run.

Secondly, there's the issue of supply security. Nvidia stated in its earnings call that its processor supply will remain tight until 2027. Anthropic has already signed several large-scale computing power procurement agreements, including purchasing $36 billion worth of Google AI chips, a $45 billion cloud computing lease agreement with Nscale, and paying SpaceX $1.25 billion per month to use its Colossus 1 data center facility (equipped with over 220,000 Nvidia chips). Developing its own chips can, to some extent, hedge against the uncertainties of external suppliers.

Thirdly, there is the performance competition. OpenAI claimed at a conference this week that its first custom chip, "Jalapeno," outperforms Nvidia's counterparts in inference computing in terms of energy efficiency. Custom chips can be optimized for specific models and workloads, thereby establishing a differentiated advantage in performance and cost-effectiveness.

Anthropic's exploration of chip self-sufficiency reflects the shared strategic orientation of leading AI labs. Google has developed its TPU series of chips, while Amazon has launched the Trainium and Inferentia chips, both aimed at reducing reliance on Nvidia and optimizing their own AI workloads.

Anthropic was one of the first AI companies to run its models on hardware from multiple vendors, and its diversified hardware strategy already has a solid foundation. As the Claude model family continues to expand, the company's demand for computing power is also increasing dramatically.

Against the backdrop of anticipated IPOs, Anthropic's chip strategy has a more direct significance in the capital market.

According to a previous Reuters report, Anthropic's IPO valuation target is $2 trillion, a figure closely tied to its revenue projection of up to $200 billion by 2028. The ability to establish independent and controllable capabilities at the hardware level will be a crucial dimension for investors in assessing its long-term profitability potential.

Risk Warning and DisclaimerInvesting involves risk; please exercise caution. This article does not constitute personal investment advice and does not take into account the specific investment objectives, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Any investment decisions made based on this information are at your own risk.