Positron, a chip designer focused on AI model inference, closed $875 million in funding across two tranches, valuing the company at $5 billion, according to an announcement on Wednesday. NEA co-led both Series C and Series C-1 rounds, with Atreides co-leading the Series C and venture investor Jim Clark leading the Series C-1 tranche.
Inference, the computing stage where trained AI models run queries and return answers, consumes less of total workload than training in production systems, yet data centers spend heavily on inference infrastructure once models move to deployment. Training consumes the most attention in venture media but represents a smaller portion of long-term operational spend.
Positron joins a widening list of inference-focused startups raising capital as major cloud providers balance their chip portfolios. NVIDIA maintains dominance in both training and inference. Google, Amazon and Microsoft have developed or funded internal chip alternatives. Inference chips require less raw compute density than training processors but must excel at throughput and latency trade-offs, shifting the engineering problem.
NEA and Atreides have backed multiple semiconductor bets tied to AI workloads. The two firms backed Crusoe Energy's Series C in 2022 and have invested in other chip-stage companies. Jim Clark co-founded Netscape and Healtheon and has backed multiple semiconductor ventures, including GraphCore and SambaNova.
Positron's $5 billion valuation sits below training-chip leaders like Cerebras, which raised at an $8 billion valuation in its last disclosed round. Inference demand is expected to grow as large language models move from experimental deployments into operational use across enterprises. The company has not disclosed revenue or production timelines.
The Series C and Series C-1 combined total of $875 million ranks among the larger single fundraises for an inference chip firm. Graphcore raised $200 million in its largest Series C in 2021; Groq, a competing inference chip startup, has raised approximately $300 million total across multiple rounds. Positron's larger round places it above both in single-round capital.
NEA co-led both tranches. Atreides' involvement in the Series C marks a separate institutional LP commitment to inference chips as distinct from training acceleration. Both tranches closed together, with NEA and Atreides moving simultaneously to reserve allocation.
