Nvidia's (NVDA) grip on AI computing isn't just about raw hardware. Its software ecosystem helps developers squeeze every bit of performance from those chips. Wafer is betting that advantage can be attacked from the other side: use AI to optimize models for rival hardware.
The year-old startup just raised $40 million in Series A funding at a valuation above $200 million, according to The Information. Its founders say Wafer has also received multiple acquisition offers from larger inference and cloud providers.
Wafer's Software Edge
Wafer runs inference on both Nvidia and Advanced Micro Devices (AMD) chips and is building AI agents that optimize models for specific workloads. In July, the company said it tuned Z.AI's GLM-5.2 for AMD's MI355X, reaching about 80% of Nvidia B200's throughput at less than half the cost in its testing.
That's the unusual part of Wafer's pitch: the alternative to Nvidia may not need to be a better chip if software can make existing alternatives perform well enough.
The evidence is still early, and Wafer remains tiny beside Nvidia. But its funding, acquisition interest and growing work with alternative accelerators suggest there is commercial value in a layer of AI infrastructure designed to make the industry less dependent on a single chip ecosystem.
The bigger question is whether inference optimization can become a meaningful counterweight to Nvidia's software advantage. Wafer is still a small experiment, but it offers an early look at a potentially important shift: competition with Nvidia may increasingly happen in software, not just silicon.
