A one-year-old startup that uses AI agents to squeeze more performance out of AMD and Nvidia chips has turned down acquisition offers from multiple cloud and inference providers, choosing instead to raise a $40 million Series A at a valuation north of $200 million.
Wafer’s valuation represents a 50x increase from its $4 million seed round closed just five months earlier in April 2026.
What Wafer actually does
The core pitch is deceptively simple: most GPUs in production environments sit around 20% utilization on average. That means companies are paying for five times more compute than they’re actually using. Wafer deploys autonomous AI agents that continuously optimize inference workloads in real time, squeezing dramatically more performance out of the same hardware.
The results from July 2026 testing tell a compelling story. Wafer’s agents pushed AMD’s MI355X GPU to roughly 80% of the throughput of Nvidia’s B200, the current gold standard for inference performance. More striking: the cost came in at less than half for certain models, including GLM-5.2.
Why turning down buyers matters
The Series A attracted a roster of investors that reads like a who’s who of AI infrastructure believers. AMD Ventures participated alongside Marathon, Chemistry, and Outset Capital. Existing backers Fifty Years and Y Combinator doubled down. The seed round had pulled in notable individual investors including Jeff Dean, Google’s legendary AI researcher, and Wojciech Zaremba, co-founder of OpenAI.
The Nvidia problem Wafer is solving
Nvidia’s dominance in AI isn’t just about making the best chips. It’s about CUDA, the software ecosystem that locks developers into Nvidia hardware. Competitors like AMD can build technically capable GPUs, but if the software stack doesn’t extract peak performance, the raw silicon specs don’t matter much.
By automating the optimization layer, Wafer effectively closes the software gap that has kept non-Nvidia chips from competing on a level playing field. An AMD GPU performing at 80% of an Nvidia B200’s throughput at under half the cost fundamentally changes the math for any company running inference at scale.
Wafer’s focus on heterogeneous compute environments means its technology could optimize workloads across any combination of chip architectures. AMD Ventures’ participation in the Series A is a strategic acknowledgment that third-party optimization software might be the fastest path to making AMD’s inference hardware genuinely competitive in enterprise deployments.