Inferact

Inferact is an artificial intelligence infrastructure company founded by the creators and core maintainers of vLLM, the world's most widely adopted open-source large language model inference engine. Originating from UC Berkeley's Sky Computing Lab in 2023, vLLM has grown to become the de facto standard for deploying AI models at scale, with over 60,000 GitHub stars, a global community of 2,000+ contributors, and adoption by major technology companies including Meta, Google, Character.AI, Amazon, and the broader AI industry. vLLM addresses a fundamental problem: as large language models have grown exponentially in size and capability, running them efficiently in production has become a critical bottleneck. The company's core innovation, PagedAttention, dynamically optimizes GPU memory allocation, enabling significant reductions in inference costs and latency while maintaining model quality—allowing enterprises to serve more users on fewer servers.

Inferact's dual mission combines open-source stewardship with commercial innovation. The company commits to maintaining vLLM as an independent, vendor-neutral project under the PyTorch Foundation governance, continuously advancing the engine with support for emerging model architectures (mixture-of-experts, multimodal, agentic AI), expanding hardware coverage (NVIDIA, AMD, Google TPU, AWS Neuron, Intel, and custom accelerators), and pushing performance boundaries through distributed inference, speculative decoding, and disaggregated serving. Simultaneously, Inferact is building a commercial platform—a managed inference service that abstracts away operational complexity. This enterprise offering provides serverless model deployment, observability and monitoring, troubleshooting tools, disaster recovery, and multi-tenant scaling on Kubernetes, enabling engineering teams to focus on AI applications rather than infrastructure.

The company's positioning at the intersection of model providers, hardware vendors, and enterprises creates a unique ecosystem advantage: deep technical relationships with both sides of the AI stack, the ability to influence industry standards, and a user base that depends on vLLM's performance and reliability. Inferact's founding team—including CEO Simon Mo (PhD student and original vLLM creator), co-founders Woosuk Kwon, Kaichao You, Roger Wang, Joseph Gonzalez (UC Berkeley faculty), and Ion Stoica (Databricks co-founder and Sky Computing Lab director)—brings 15+ years of collective experience building infrastructure companies that transformed entire technology sectors, particularly Apache Spark and Ray from research projects into industry standards. Backed by $150 million in seed funding from Andreessen Horowitz, Lightspeed, Sequoia, and other leading investors at an $800 million valuation, Inferact is positioned to accelerate the shift from inference as a research problem to inference as a solved, commoditized, and profitable operating layer of the AI stack.

Keywords: open-source inference engine, large language model serving, GPU optimization, PagedAttention, model deployment, serverless inference, enterprise AI infrastructure, multi-hardware support, distributed inference, PyTorch Foundation, managed services, AI infrastructure, generative AI, LLM optimization, production AI systems.