OpenAI Weighs Google TPUs

OpenAI’s move to test Google’s AI chips—the TPUs—signals more than just window shopping. With TPUs now a “meaningful contributor” to Alphabet’s growth and OpenAI developing its own “Jalapeno” silicon, the grip NVIDIA has on AI compute is being tested. As major labs blend NVIDIA GPUs with TPUs, their procurement muscle grows, pushing for better pricing and supply. For NVIDIA, this isn’t just a headline risk: it’s a real threat to future contract share, especially heading into the 2027 refresh cycle.

But here’s the catch—NVIDIA’s counteroffensive is all about software lock-in. The new Cosmos platform deepens reliance, enabling advanced simulation for robotics and healthcare and pushing buyers toward higher-memory, premium GPUs. Cosmos is optimized for NVIDIA hardware, making it easy for customers to scale but tough to switch. At the same time, supply chain fragility remains: the latest Blackwell chip teardown shows critical dependencies on SK hynix for cutting-edge memory and TSMC for advanced packaging. Any hiccups there, and shipment volumes cap out—regardless of demand.

Featuring reporting from Nikkei Asia, Reuters, TechInsights, Digital Foundry, GameGPU, Light Reading, SCMP, and Digitimes.

Powered by Apisod.com