Anthropic confirmed on 5 August 2026 that it is building an in-house silicon team, hiring hardware and software engineers to design custom chips co-developed with Claude models themselves. The team is anchored by Clive Chan, previously the second hardware hire on OpenAI’s own chip effort, and Anthropic is reportedly scouting Samsung as a manufacturing partner. The stated goal isn’t a moonshot spec sheet — it’s cost: co-designing silicon around Claude’s specific attention mechanisms, targeting roughly a 50% reduction in per-token inference costs.
That number matters more than the novelty of an AI lab building chips. Every AI coding tool — Claude Code included — runs on a cost structure set by whichever chips it’s renting from Nvidia, AWS, or Google. When usage-based pricing has been volatile all year (Claude Sonnet 5’s launch pricing ends this month; GitHub Copilot moved to usage-based billing in July), the labs that control more of their own inference stack are the ones best placed to hold prices steady, or to keep expanding free usage tiers, rather than passing hardware cost spikes straight through to subscribers.
This is also Anthropic following a well-worn playbook rather than inventing one. Google has run its own TPUs for years; OpenAI already has a hardware team of its own. What’s notable is the timing: Anthropic is making this move while its run-rate revenue has reportedly passed $30bn and enterprise customers spending $1m+ annually have crossed 1,000, which reads less like early-stage experimentation and more like a company sizing its infrastructure to demand it’s already struggling to keep up with — the same demand pressure that’s pushed Anthropic to extend Claude Code’s usage boosts multiple times this year.
So what
If your team has standardised on Claude Code or another Anthropic-backed tool, this isn’t a change to plan around today — it’s a signal about the vendor’s staying power. A lab investing in its own silicon supply chain is one that expects to still be the dominant player in three years, not one hedging its bets. For businesses weighing which AI coding tools to build a long-term delivery process around, vendor infrastructure investment is as relevant a signal as model benchmarks. See how we think about tool selection and where AI genuinely earns its place in a build on our AI-assisted development page, or get in touch to talk through your stack.