Anthropic in talks with MatX to accelerate custom AI chip development

Anthropic is in discussions with chip startup MatX to accelerate the development of its own custom AI hardware. The move signals a strategic push by the AI lab to secure its own supply chain and optimize the silicon powering its models.

According to reports from Reuters, the partnership aims to leverage MatX's specialized design capabilities to speed up the chip design cycle. MatX specializes in high-throughput AI chips designed specifically for large language models (LLMs). This shift toward a partnership model follows a period where Anthropic reportedly considered acquiring MatX for approximately $7 billion.

The race for custom silicon

This initiative is part of a broader industry trend where major AI players are moving away from a one-size-fits-all hardware approach. Companies including Google, Amazon, and Microsoft have already developed their own AI accelerators to optimize performance and lower operational costs. As a leading developer of LLMs, Anthropic is following this trajectory to ensure its hardware is precisely tuned for the specific requirements of its architecture.

Reducing the Nvidia bottleneck

If successful, the partnership could allow Anthropic to iterate on hardware faster than traditional design cycles. This speed is critical for the efficient training and inference of its Claude models. More importantly, developing proprietary silicon reduces Anthropic's strategic dependency on Nvidia's H100 and B200 GPUs, which currently dominate the market and often face supply constraints.

What to watch

While the discussions are ongoing, the outcome will determine how quickly Anthropic can transition from a consumer of third-party hardware to a producer of its own. Industry observers will be watching to see if this partnership results in a formal joint venture or a long-term licensing agreement, and how quickly the first generation of MatX-assisted silicon can be deployed into Anthropic's clusters.