Anthropic Is Building Its Own Chips for Claude: Job Listings, Multi-Supplier Strategy, and How It Compares to OpenAI's Jalapeño

The Silicon Engineer role explicitly requires candidates who have "shipped silicon" and can make "complex engineering and business tradeoffs" as Anthropic builds its first-generation custom accelerator . The Research Engineer role sits within the Code RL team and focuses on agentic RTL generation, formal verification, and physical design optimization — effectively creating a flywheel where AI helps design the chips that run AI .

A Multi-Chip Strategy That Adds a Fifth Pillar

Anthropic has long pursued a deliberately diversified, multi-platform compute strategy rather than relying on any single supplier. The company's October 2025 blog post described its approach as a "unique compute strategy focused on a diversified approach that efficiently uses three chip platforms" (Google TPUs, AWS Trainium, Nvidia GPUs) . By July 2026, that portfolio had expanded to four suppliers:

• AWS (Amazon Trainium) — Anthropic's earliest primary compute partner; Trainium 2 chips power "Project Rainier," with commitments of up to 5 GW of capacity
• Google TPUs — An October 2025 deal worth tens of billions of dollars for up to one million custom TPUs, delivering well over 1 GW of capacity in 2026, expanded to multi-gigawatt next-generation TPU capacity starting in 2027
• Nvidia GPUs — Still in the mix, though Anthropic has actively reduced dependence on Nvidia through its other partnerships
• AMD Instinct MI450 — A July 2026 agreement covers up to 2 GW of Instinct MI450-series GPUs, with AMD also investing up to $5 billion in Anthropic as deployment milestones are met

The in-house chip team adds a fifth pillar: custom silicon designed in-house, potentially for inference optimization, that can be co-optimized with Claude's architecture. The company has stated it will maintain its multi-supplier relationships even as it builds internal capability .

How Anthropic's Approach Compares to OpenAI's Jalapeño Chip

OpenAI's Jalapeño chip, unveiled on June 24, 2026, is a purpose-built inference-only ASIC co-designed with Broadcom, manufactured on TSMC 3nm, and slated for late-2026 pilot deployment. Early lab testing shows roughly 50% lower inference cost per token versus current-generation Nvidia GPUs, with performance per watt "substantially better than current state-of-the-art" . It went from initial design to manufacturing tape-out in nine months .

Key differences between the two approaches:

Anthropic is at an earlier stage — still hiring and designing — whereas OpenAI already has working silicon. Reports also indicate Anthropic has been in early-stage talks with Samsung Foundry for a potential 2nm-class chip, which would be a more advanced node than Jalapeño's 3nm .

The Financial Context: Why $500M–$1B Is a Modest Bet

Anthropic's revenue trajectory has been extraordinary, providing the financial firepower to justify custom silicon development:

• End of 2024: ~$1 billion annualized run rate
• End of 2025: ~$9 billion run rate
• Early April 2026: ~$30 billion run rate (confirmed by Anthropic)
• May 2026: Sacra estimates reached ~$47 billion annualized run rate

Industry estimates for designing a cutting-edge AI accelerator range from $500 million to over $1 billion, accounting for engineering teams, EDA tools, mask sets, and initial wafer runs . Given Anthropic's $30B–$47B revenue run rate, that represents roughly 1–3% of annualized revenue — a relatively modest investment for a company of its scale and one that could yield massive savings if the chip meaningfully reduces inference costs for Claude's enormous and growing workload. As analysis site Cryptopolitan noted, the cost to build is a fraction of what Anthropic spends on external compute .