GPU startup raises $1 billion Series A to build megakernel framework for cheaper AI training

The startup says it has built what it calls the world's most complete megakernel framework, capable of running entire training runs, serving inference, and playing video games—simultaneously in a single kernel. It raised a $1 billion Series A co-led by Index and Lux, with backing from Sequoia, NVIDIA, Kleiner Perkins, GV and others, plus individuals including Steven Cohen and Henry Kravis. The company claims its systems work cuts training and inference costs, enables unusual architectures and optimizers, and has shown internal evidence at trillion-parameter pretraining scale. It aims to unlock markets like robotics and scientific discovery.

✨ AI summary of the linked source — not editorial content.

Explore the topic

• Raises
• GPU
• Series A
• AI