OpenRelay funding, valuation and investors

Product & use cases

Illustrative buyer tour derived from the official November 2026 product and pricing documentation: (1) an AI-app team defines model, region/latency, budget and throughput; (2) it creates an API key and sends model requests using an OpenAI- or Anthropic-compatible interface; (3) OpenRelay routes workloads to available GPU or specialized accelerators in the Inference Delivery Network; (4) time-sensitive calls run synchronously, while less urgent requests can use discounted asynchronous one-hour or 24-hour windows, or batch JSONL; (5) the team evaluates latency distributions, output quality, reliability and effective cost per million tokens against another provider; (6) dedicated compute or reserved throughput may be arranged for more predictable production capacity. Published schedule, checked Nov 14: GPT-OSS 120B at $0.15/M input and $0.60/M output tokens for synchronous service; GPT-OSS 20B at $0.05/M input and $0.20/M output. Selected async/batch routes advertise 20%-50% discounts. These are model-specific examples, not a blanket rate or inclusive GPU capacity contract. Supplier claims on autoscaling and relative costs have not been independently reproduced.

• Voice agent routes latency-sensitive inference to available regional accelerators.
• Batch document processing queues nonurgent inference to reduce token costs, accepting later completion.
• AI infrastructure team integrates one inference API to benchmark multiple models and providers.

Key facts

• $10M seed announced November 12, 2026, led by Uncork Capital, with Eniac Ventures and Standard Capital.
• Issuer reports over 20 compute sites, over five trillion processed tokens per month and a $10M annualized revenue run rate within five months of founding; no audited revenue disclosed.
• Live homepage displays 2T+ monthly tokens while funding announcement says 5T+; a likely updating discrepancy, but no reconciliation provided.
• Company offers per-token synchronous, asynchronous and batch inference, reserved throughput and GPU rentals, with workload-specific charges.

Competitive landscape

Edge: Distributed multichip orchestration combines software routing and third-party/owned compute; founders' AWS and Voltage Park backgrounds are supported by Uncork. Advantage depends on usable network supply and service quality, not guaranteed 10-20% cheaper inference for all models.

Uncork's lead-investor note is primary but not independent validation. Source mismatch: the funding blog (Nov12) claims >5T tokens/mo while the homepage currently shows 2T+; retain both as issuer-stated and do not infer a precise current rate. Claimed $10M annualized run rate is an extrapolated pace, not GAAP annual revenue, booked contracts, profit or audited ARR. Provider contracts, gross margin, hardware capital intensity, failover reliability, customer concentration and valuation are not disclosed. Discounts depend on specific model/tier and time windows; compare actual blended input/output rates and service-level targets.

• Together AI direct

  Hosted open-model inference and dedicated GPU endpoints; customers can compare per-token and deployment cost.

• Fireworks AI direct

  Optimized inference endpoints and dedicated deployments; competes on price and latency.

• Baseten direct

  Managed model deployment and inference with dedicated operational tooling; less emphasis on heterogeneous broker network.

• AWS Bedrock incumbent

  Managed model APIs, compliance and enterprise integration; multichip routing transparency differs.

• Self-hosted vLLM or TensorRT-LLM alternative

  Can lower intermediary fees on high stable volume but moves operations and capacity procurement onto buyer.

Market / IPO context