Anthropic engaged in discussions to acquire artificial intelligence semiconductor startup MatX for approximately $7 billion before talks became inactive, according to reporting from Reuters. The potential transaction highlights efforts by leading frontier AI developers to vertically integrate custom silicon design into their core infrastructure operations.
MatX was founded in 2022 by Reiner Pope, a former Google Brain engineer who contributed to the PaLM language model, and Mike Gunter, a veteran hardware architect from Google's Tensor Processing Unit (TPU) development team. The startup designs specialized application-specific integrated circuits (ASICs) tailored exclusively to the matrix multiplication workloads that dominate modern transformer-based large language models.
Custom Hardware Strategies Across Frontier Labs
While the acquisition negotiations did not result in a completed deal, the discussions underscore how frontier AI companies are exploring proprietary hardware to reduce reliance on Nvidia GPUs. Unlike general-purpose graphics processors that support broader computing workloads, MatX focuses its architecture on extreme memory bandwidth and low-precision arithmetic specifically optimized for dense and mixture-of-experts model execution.
The reported $7 billion valuation would have represented one of the largest private acquisitions in the AI semiconductor sector. Following the conclusion of discussions with Anthropic, MatX is currently pursuing a new private funding round targeting a valuation of roughly $4 billion. The startup previously secured over $500 million in venture financing to support its product roadmap, which includes planned manufacturing through TSMC.
Anthropic declined to comment on the acquisition talks, while MatX did not issue a public response.
