Shortly after Amazon CEO Andy Jassy announced AWS's $50 billion investment deal with OpenAI, the tech giant invited me on an exclusive tour of its cutting-edge Trainium lab. The latest in-house chip, Trainium is being hailed for its potential to lower costs and challenge Nvidia’s dominance.

The facility was buzzing with excitement as I met directors Kristopher King and Mark Carroll, who guide the team responsible for this technological marvel. Anthropic, another big player, has already adopted Trainium, with over a million of their Claude AI running on these chips.

While originally designed for faster model training, Trainium is now a powerhouse in inference—a critical process that was previously a bottleneck. This is no small feat: Bedrock, Amazon’s service for building AI applications, relies heavily on Trainium2 to handle its massive traffic demands.

AWS claims that running jobs on Trainium can be up to 50% cheaper than using classic cloud servers. This competitive edge isn’t lost on industry watchers; Apple itself has praised AWS's chip designs in the past, including Graviton and Inferentia.

But Amazon doesn't stop at just designing chips. The company is also crafting the server hardware to host these chips, integrating cutting-edge cooling technologies and virtualization tools into its offerings. Whether this means cheaper, more accessible AI or a new tech battlefield remains to be seen.