Anthropic explores designing its own AI chips as demand surges

TL;DR: Anthropic is in early discussions about designing its own AI chips, according to three sources cited by Reuters. The plans are preliminary — no dedicated team or specific design has been committed — but reflect a broader industry shift as frontier AI labs seek to reduce dependency on external chip suppliers. With Anthropic’s run-rate revenue now past $30 billion, the economic case for custom silicon is strengthening rapidly.

The exploration comes as demand for Anthropic’s Claude models has accelerated through 2026. The company’s run-rate revenue has more than tripled from about $9 billion at the end of 2025, creating both the financial capacity and the operational incentive to consider in-house chip development.

Following the vertical integration trend

Anthropic currently relies on a mix of hardware, including Google’s tensor processing units (TPUs) and Amazon’s custom chips, to train and run its AI systems. Earlier this week, the company signed a long-term deal with Google and Broadcom — which helps design TPUs — as part of a broader $50 billion commitment to US computing infrastructure.

Designing an advanced AI chip typically costs around half a billion dollars, according to industry sources, requiring specialised engineering talent and rigorous manufacturing validation. The barrier to entry is substantial, but Anthropic’s revenue trajectory may now make the investment viable.

The move mirrors efforts already underway at Meta and OpenAI, both of which are pursuing custom chip programmes. Google and Amazon have long operated their own chip design divisions, giving them cost and performance advantages in training and serving AI models. For frontier labs still dependent on Nvidia’s GPU ecosystem, the chip shortage of recent years has been a powerful motivator to explore alternatives.

Looking forward

Custom chip development is a multi-year undertaking, and Anthropic may ultimately decide to continue buying rather than building. But the exploration itself signals that the AI industry’s supply chain is entering a new phase, where the largest model developers increasingly view chip dependency as a strategic vulnerability rather than a manageable cost.