Key Takeaways
- Anthropic is reportedly exploring early-stage development of its own AI chips, according to sources familiar with the matter.
- The move is driven by surging compute demand and an annualised revenue run rate that has passed $30 billion, making custom silicon economics increasingly attractive.
- The exploration puts Anthropic alongside Meta and OpenAI in a broader industry shift toward specialised hardware for performance control and supply chain independence. Anthropic is exploring its own custom AI chips — even as it just locked in a major long-term deal for Google TPU capacity. Three sources familiar with the matter told reporters this week that the AI lab is in early-stage deliberations, though no team has been assembled and no design has been committed to. The timing reveals the tension at the heart of modern AI infrastructure: even well-supplied labs are hedging against a hardware landscape they don’t fully control.
Anthropic Eyes Custom Silicon Amid Compute Crunch
The compute pressures driving this exploration are real. Anthropic’s annualised revenue run rate has surpassed $30 billion, up from around $9 billion at the end of 2025 — a trajectory that dramatically scales its hardware requirements. At that growth rate, the economics of custom silicon stop being theoretical. Designing chips optimised specifically for Claude’s architecture could deliver meaningful gains in both performance per watt and cost per inference, the two metrics that matter most at scale.
There’s also a supply chain argument. Dependence on third-party chip suppliers — however capable — creates exposure to allocation constraints and lead times that a lab running frontier models at this scale can’t easily absorb. Custom silicon offers a path to controlling that stack directly.
Early Stages and Existing Partnerships
For now, Anthropic’s chip ambitions remain firmly pre-commitment. No dedicated engineering team exists, and the company hasn’t settled on a design direction. It’s as much a feasibility study as a programme — and it’s entirely possible the economics or execution complexity lead Anthropic to stay a hardware customer rather than become a hardware developer.
In the meantime, Anthropic runs Claude across a mix of Tensor Processing Units (TPUs) — custom processors designed by Google in partnership with Broadcom — alongside Amazon’s Trainium and Inferentia chips and Nvidia GPUs. That hardware diversity is deliberate: it provides negotiating leverage and resilience against any single supplier’s constraints.
The long-term Google-Broadcom deal signed just days before these reports surfaced is the clearest signal of where Anthropic’s infrastructure actually sits today. It secures access to roughly 3.5 gigawatts of TPU-based compute capacity from 2027 — approximately triple what the company was consuming earlier in 2026, according to sources — and builds on a prior commitment to invest around $50 billion in US computing infrastructure. That’s a lot of external silicon to be planning around while simultaneously exploring your own.
Industry-Wide Trend Towards Custom AI Hardware
Anthropic isn’t the first AI lab to think this way. Meta and OpenAI are already deep into custom chip programmes for both training and inference workloads — a sign that at sufficient scale, the vertically integrated hardware model starts to make sense. The pattern mirrors what happened in hyperscale cloud computing, where Google, Amazon and Microsoft all eventually moved to custom silicon rather than relying solely on merchant chips.
The barrier is steep. Industry estimates put the cost of developing an advanced AI chip at around $500 million, covering specialised engineering talent and the extensive validation required before tape-out and production. That’s a serious commitment — but for a company growing at Anthropic’s current pace, it’s a number that sits within strategic reach. The real question isn’t whether Anthropic can afford to build its own chips. It’s whether doing so would actually outperform doubling down on its existing supplier relationships. For more coverage of AI chips and infrastructure, visit our AI Hardware section.
Originally published at https://autonainews.com/anthropic-explores-building-ai-chips-as-revenue-hits-30-billion/
Top comments (0)