Breaking

Anthropic's Multi-Chip Strategy Delivers 85% Gross Margin on Model Serving

October 28, 2026 at 11:31 EDT

Foundation Models
Infra & Chips
Sales & Business

Anthropic achieved an 85% gross margin on serving its Claude models in the first half of 2026 by deploying inference workloads across chips from Nvidia, Amazon, Google, and AMD, according to the company's S-1 filing. This diversified approach to silicon—treating processors from multiple vendors as a single fungible computing pool—enables Anthropic to optimize costs and avoid vendor lock-in while serving billions of API requests.

AI Infrastructure Strategy

Anthropic's Multi-Chip Inference Model

85% gross margin through diversified semiconductor partnerships and optimized inference serving.

85%

H1 2026 Gross Margin

11 GW

Contracted Compute Capacity

$60M

Revenue per MW (Current)

3.75×

9-Month Efficiency Gain

Compute Distribution Across Four Vendors

42%

Nvidia GPUs

(4.7 GW)

41%

Google TPUs

(4.5 GW)

11%

AWS Trainium

(1.2 GW)

AMD GPUs

(0.8 GW)

Strategic Advantages of Multi-Chip Approach

Negotiating leverage: Diversified purchasing reduces dependence on any single vendor
Supply resilience: Hardware failures or shortage in one vendor don't disrupt operations
Cost optimization: Dynamic routing of workloads to cheapest available hardware
Future flexibility: Custom chip development (targeting 50% cost reduction) complements vendor partnerships

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.