Has the forgotten LPU, neglected for ten years, made a comeback—has a new business emerged?

▤ Summary

This article examines why LPU chips are attracting renewed attention as AI workloads shift from training to inference. It explains how specialized chips can split tasks across different architectures and why Groq’s LPU has gained visibility through Nvidia’s Vera Rubin platform. The piece also weighs the technical tradeoffs of static scheduling and SRAM-based design against competing approaches. It then questions whether independent LPU vendors can build durable businesses or whether larger platform players will absorb this function. Overall, the article frames LPU as both a technical opportunity and a difficult commercial market.

▦ Linked Companies

NVIDIA Corporation

$1B+

Groq

$10M to $25M