Startup software optimizes AMD GPUs for inference cost advantage over Nvidia

A year-old AI startup called Wafer raised $40 million in Series A funding at a valuation exceeding $200 million, marketing software that optimizes machine learning models specifically for AMD hardware. The company demonstrated that its software can tune models to run on AMD's MI355X GPU at 80% of the throughput of Nvidia's B200, but at less than half the cost in testing. Wafer's approach suggests that Nvidia's long-standing software ecosystem advantage might be challenged through layers of optimization aimed at rival chips, potentially making AMD's inference capabilities more cost-attractive to customers despite current performance gaps.

1. Wafer raised $40 million in Series A funding at a valuation above $200 million, according to The Information
2. Wafer said it tuned Z.AI's GLM-5.2 model for AMD MI355X to reach about 80% of Nvidia B200's throughput at less than half the cost in its testing
3. Wafer founders report receiving multiple acquisition offers from larger inference and cloud providers
4. Wafer runs inference on both Nvidia and AMD chips and builds AI agents that optimize models for specific workloads
5. Wafer's pitch centers on using AI-driven software to make alternatives to Nvidia perform well enough, rather than relying solely on superior hardware