$50 billion. That's how much Amazon just committed to its bet on custom AI chips, and a rare look inside the company's Trainium lab shows why the biggest names in AI keep signing up.
AWS opened its chip design facility to press shortly after announcing its massive investment tied to the OpenAI partnership. The timing was deliberate: Amazon wants the world to know it's not just renting out Nvidia GPUs anymore. It's building the silicon that trains and runs the models from Anthropic, OpenAI, and now Apple.
Trainium is Amazon's custom chip designed specifically for training large AI models (the expensive, months-long process of feeding data into a neural network until it learns patterns). The latest generation, Trainium2, is what these partnerships are built on. For AI companies, the appeal is straightforward: cheaper compute than Nvidia's H100s, tighter integration with AWS cloud services, and guaranteed capacity at a time when GPU supply is still tight.
Who's Actually Using These Chips
Anthropic has been the longest-running Trainium customer, which makes sense given Amazon's multi-billion-dollar investment in the company. OpenAI signing on is the bigger surprise, and signals that even companies with deep Microsoft ties want supply chain diversification. Apple's involvement is the most intriguing - the company has said almost nothing publicly about its AI training infrastructure, and showing up in Amazon's chip lab suggests a significant cloud AI commitment beyond what runs on-device.
The practical question for anyone building on these platforms: does it matter whose chips train your model? Mostly no. You interact with Claude or ChatGPT through an API. But it matters a lot for pricing. More competition in AI chips means lower costs for training, which eventually means lower API prices and cheaper subscriptions. Nvidia's near-monopoly on AI training hardware has been one of the biggest cost drivers in the industry, and Amazon is making the most credible challenge to that position.
Amazon's chip play also explains its willingness to invest so heavily in AI companies that compete with its own Bedrock platform. The real business isn't picking the winning model - it's making sure every model runs on AWS hardware.