Inferact has emerged from stealth with $150 million in funding to commercialise vLLM, betting that open-source inference, backed by managed infrastructure, is the fastest path to scalable enterprise AI.
Inferact Inc. has launched to commercialise vLLM, a widely adopted open-source inference engine for large language models, backed by $150 million in seed funding and an $800 million valuation, underscoring strong investor confidence in open-source AI infrastructure.
The funding round was led by Andreessen Horowitz and Lightspeed, with participation from Databricks Inc.’s venture capital arm, the UC Berkeley Chancellor’s Fund, and other investors. Inferact is founded by a group of AI researchers closely tied to vLLM’s origins, including Ion Stoica, Databricks co-founder, UC Berkeley computer science professor, and director of the Sky Computing Lab, where vLLM was originally developed in 2023.
vLLM has grown into a production-proven open-source project with more than 2,000 contributors, and is widely used to accelerate AI inference through advanced memory and performance optimisations. Its key innovation, PagedAttention, reduces memory waste by storing KV cache data in non-adjacent RAM segments, significantly lowering hardware consumption. Additional techniques such as quantisation further shrink model memory footprints, while multi-token generation improves inference speed.
Inferact plans to introduce a paid, serverless version of vLLM, automating infrastructure provisioning, updates and operations. Job postings suggest the managed offering will include observability, troubleshooting and disaster recovery capabilities, with deployment likely on Kubernetes.
Crucially, Inferact’s founding team includes core vLLM maintainers, who will continue contributing to the upstream open-source project. Planned enhancements include new performance optimisations, support for emerging model architectures, and broader data centre hardware compatibility.
“We see a future where serving AI becomes effortless,” said Woosuk Kwon, Co-Founder, Inferact Inc.
