OpenRelay, a San Francisco inference startup that builds a network for running AI workloads across different kinds of chips, said Thursday it raised a $10 million seed round led by Uncork Capital, with participation from Eniac Ventures and Standard Capital. The company announced the round on November 12, 2026. Neither the company's announcement nor Uncork's memo disclosed a valuation or deal structure.
For a company this young, the pitch is ambitious. Uncork general partner Amy Saper laid out the firm's thesis in a memo on its site, arguing that increasingly diverse models, hardware and workloads create demand for infrastructure that handles hardware selection and operations on behalf of developers.
"We believe OpenRelay has the potential to become a foundational infrastructure layer for the next generation of AI companies," Saper wrote. The memo is undated, and it reflects Uncork's rationale for investing, not independent evidence of how large the market might be.
What OpenRelay says it built
The company calls its product an Inference Delivery Network, or IDN. Inference is the stage at which a trained AI model answers requests, and it is commonly cited as a significant source of the ongoing cost of running AI products, though the company's materials provide no cost breakdown. According to OpenRelay's announcement, the IDN combines the company's own deployments with third-party capacity, storage placed near GPUs, networking between GPUs and a custom inference engine. The company says it can start a model from a cold state in seconds.
Prashant Patel, the chief technology officer and a co-founder, described the routing approach in an undated LinkedIn hiring post. "We run production inference across multiple kinds of silicon and route workloads based on real-time latency and cost," he wrote.
The company's homepage claims more than 2 trillion tokens processed per month, autoscaling that is 2.5 times faster, and costs 10% to 20% lower at high throughput. The page does not explain how those comparisons were made. Other OpenRelay materials cite different savings, including "up to 90% cheaper" on LinkedIn, and the baselines behind the figures are not reconciled.
The announcement also cites an annualized revenue run rate, which projects current revenue forward and is not the same as recognized annual revenue. The company did not name a paying customer in its financing announcements. It pointed to robotics, voice assistants, agents and batch processing as example workloads.


