Investing in OpenRelay: Building the Inference Delivery Network

We’ve been paying close attention to the recent surge of new AI models and the diversification of the underlying hardware landscape that supports these models. As open models become increasingly capable, developers are able to choose the model that excels at a particular task (often with far more attractive economics), rather than relying on a handful of proprietary LLMs. Hardware is also diversifying, a trend we believe will accelerate as AI transforms chip design itself. In its S-1, Anthropic highlighted its heterogeneous compute strategy spanning NVIDIA and AMD GPUs, AWS Trainium, and Google TPUs. This approach required years of investment and substantial internal engineering resources, something few AI companies can replicate. 

Meanwhile, inference workloads are growing rapidly and becoming more varied. Real-time voice applications, coding agents, batch document processing workloads, specialized robotics, and other emerging use cases have dramatically different latency, performance, and cost requirements. Together, these trends create both the opportunity and the need for a new infrastructure layer. 

OpenRelay is building inference-native infrastructure that enables developers to take advantage of an increasingly diverse compute ecosystem without having to optimize models for individual chips, source capacity across providers, or manage complex infrastructure themselves. We’re proud to be leading their $10 million seed round, with support from Eniac and Standard Capital.

OpenRelay’s Inference Delivery Network (IDN) is a distributed network of heterogeneous compute spread across many regions and hardware providers, with proprietary orchestration software that optimizes for performance and cost. The concept is similar to how content delivery networks (CDNs) transformed the internet by caching content at the edge, close to end users, dramatically reducing latency. For developers, that means leveraging the right hardware for every workload, whether that’s split-second responses for a voice assistant or the lowest possible cost for overnight batch processing, without juggling providers or optimizing for individual hardware architectures themselves. 

I first met OpenRelay's founders Jaden Wang and Prashant Patel this summer, when the company was two months old and in YC's Summer 2026 batch, and was immediately struck by how complementary their backgrounds were. Jaden started a data center business while still in college and then joined Voltage Park (one of the leading GPU neoclouds) as an engineer. Within months, he owned the whole stack behind their virtualization product, from the metal and networking up to the virtualization software. Prashant spent five years at AWS building production inference infrastructure, including serving as one of the founding engineers on Bedrock. He later joined Voltage Park, where he met Jaden. Together, they bring a rare combination of deep expertise in both physical compute infrastructure and the software required to operate it efficiently at scale.

Their pace of execution and the scale of their ambition have astounded us. When they joined YC in June, they had no customers, revenue, or compute supply. In just five months, they built a distributed network spanning more than 20 regions and 10 hardware types, processing over 5 trillion tokens a month, while growing revenue from $0 to a $10 million annualized run rate. 

As AI models and hardware architectures continue to proliferate, and demand for inference continues to outpace supply, the need to optimize performance and economics will only grow. But taking advantage of these opportunities requires significant investment in physical infrastructure, sophisticated software, and deep technical expertise that most AI application developers simply don't have in-house. We believe OpenRelay has the potential to become a foundational infrastructure layer for the next generation of AI companies.

Congratulations to Jaden, Prashant, and the OpenRelay team! We’re thrilled to be on this journey with you.