Amazon's Trainium AI chips challenge Nvidia and attract $100M startup investment

Amazon Web Services is expanding its Trainium line of AI accelerators, aiming to reduce reliance on Nvidia GPUs in data‑center workloads. The latest Trainium3 chips can be densely stacked in Amazon’s UltraServers, allowing rack‑scale performance comparable to Nvidia’s Blackwell GPUs at lower cost. While Nvidia still dominates the market with its CUDA ecosystem, Amazon’s push for proprietary silicon signals a long‑term competitive shift in the AI chip sector.

Startup TwelveLabs, which builds video‑search AI technology, closed a $100 million Series B round that includes an equity stake from Amazon. The company will run its video‑inference workloads on AWS Trainium chips under a multiyear preferred‑cloud agreement, positioning TwelveLabs against Google’s video‑AI offerings and providing a high‑profile use case for Amazon’s custom silicon.