Anthropic Explores In-House AI Chips as Revenue Surges Past $30 Billion
Anthropic is in early-stage discussions to design its own AI chips, according to sources familiar with the matter. The company has not finalized any design, formed a dedicated team, or committed to the project. It may continue purchasing from external suppliers instead.
The exploration comes as Anthropic's annualized revenue run rate jumped to over $30 billion, up from roughly $9 billion at the end of 2025. The threefold increase in three months has intensified pressure to secure adequate computing capacity for Claude's expansion.
Current Hardware Dependencies
Anthropic currently relies on chips from Nvidia, Google, Amazon, and Broadcom to train and run Claude. The company uses Amazon's Trainium and Inferentia processors alongside Google's tensor processing units (TPUs). In recent weeks, Anthropic secured a strategic partnership with Google and Broadcom for 3.5 gigawatts of AI compute capacity starting next year.
Why In-House Development Matters Now
Global semiconductor shortages are affecting all major AI companies. As demand for high-performance chips continues to surge, securing adequate hardware has become a critical bottleneck-one that now rivals model innovation as the primary constraint on growth.
The move mirrors broader industry trends. Meta and OpenAI already have chip development projects underway. Amazon, Google, and Microsoft have all pursued in-house chip development to reduce reliance on external suppliers.
The Economics Problem
AI chip development costs can exceed $500 million and requires specialized talent and years of design, testing, and manufacturing. For a company like Anthropic-even one with substantial funding-the investment represents a fundamental strategic bet on long-term independence from existing suppliers.
The company's current multi-vendor strategy suggests it values flexibility. By diversifying across Google TPUs, Amazon chips, and Nvidia GPUs, Anthropic hedges against any single supplier's capacity constraints.
Timeline and Decision Points
Whether Anthropic proceeds depends heavily on whether existing partnerships can meet scaling demands. The 3.5 GW agreement with Google and Broadcom suggests current discussions may be exploratory hedging rather than imminent product strategy.
If that capacity proves sufficient for Claude's growth through 2026 and 2027, the cost-benefit calculus likely tilts toward remaining a customer. Should hardware constraints emerge despite partnership commitments, Anthropic would face a pivotal decision point around 2027 or 2028.
Establishing the engineering and manufacturing infrastructure for a new processor requires 18 to 36 months minimum. This timeline means any chips Anthropic develops would not enter production until 2028 at the earliest.
Competitive dynamics will also matter. If Nvidia expands capacity or if Amazon and Google's in-house chips prove adequate for most workloads, the urgency for Anthropic to develop proprietary silicon diminishes. Conversely, if supply remains constrained despite multiple vendor efforts, Anthropic may face pressure to commit resources like Meta and OpenAI have.
The company's current position-exploring without committing-amounts to a useful negotiating tactic. It signals to Google, Amazon, and Nvidia that Anthropic will consider alternatives if supply commitments falter, while preserving the option to remain a customer if the economics and supply situation improve.
Your membership also unlocks:
