Inside Amazon's Trainium Chip Lab: The Silicon Powering Anthropic, OpenAI, and Apple

Amazon data center with Trainium AI training chips

Shortly after Amazon announced its staggering $50 billion investment in AI infrastructure, AWS invited TechCrunch on an exclusive tour of the chip lab at the heart of the deal: the facility where Amazon designs and tests Trainium, its custom AI training silicon.

Trainium isn’t just another chip in a crowded market. It’s the silicon that has quietly won over some of the biggest names in AI: Anthropic, OpenAI, and even Apple are now running workloads on Trainium-powered AWS instances. That’s a remarkable customer list for a chip most consumers have never heard of.

Why Custom Chips Matter

NVIDIA dominates AI training with its H100 and B200 GPUs, but supply is perpetually constrained and prices are astronomical. Amazon’s Trainium offers an alternative: custom-designed silicon optimized specifically for AI training workloads, available exclusively through AWS at lower per-token costs.

For AI companies spending millions per month on compute, even a 20-30% cost reduction on training runs translates to enormous savings. That’s the value proposition that convinced Anthropic to build its Claude infrastructure substantially on AWS.

The $50 Billion Bet

Amazon’s $50B AI investment isn’t just about chips — it’s about building the infrastructure layer that every AI company depends on. Data centers, networking, cooling systems, and custom silicon all working together to make AWS the default platform for AI at scale.

The strategy mirrors what Amazon did with e-commerce infrastructure: build it for yourself, then rent it to everyone else. AWS started as Amazon’s internal cloud. Trainium started as an internal training chip. Both are now critical infrastructure for the entire industry.

For the latest in tech hardware, check out our best TVs guide and our best camera phones with AI-powered features.

The Bottom Line

Amazon’s Trainium bet is paying off. When Anthropic, OpenAI, and Apple all choose your silicon, you’ve built something the market needs. The $50B investment suggests Amazon sees AI infrastructure as the next AWS — a platform business that prints money for decades. NVIDIA should be watching closely.