a16z Leads! Inferact Enters with 150 Million USD, Kicks Off the AI Inference Cost Reduction Battle

After academia孵化多个现象级 AI 引擎, the core members of the open-source inference framework vLLM officially announced the establishment of a startup company Inferact. The company successfully completed a seed round funding of $150 million with a pre-money valuation of $800 million.

Robot, AI

Top Capital Support, Accelerating "Inference" Commercialization

This round of financing was led by Andreessen Horowitz (a16z) and Lightspeed Venture Partners. This move confirmed the market's previous speculation about vLLM's commercialization path and marks the shift of the AI industry's focus from "model training" to "application inference."

Technical Background: Inferact was incubated in the laboratory of Professor Ion Stoica from the University of California, Berkeley (co-founder of Databricks). Its core technology, vLLM, significantly improves the speed of large model operations and reduces energy consumption through innovative memory management technology.
Market Position: CEO Simon Mo said that the open-source version of vLLM has been widely adopted by major companies such as Amazon Web Services (AWS) and Amazon Shopping.

The "Berkeley Twins" in the Inference Sector

Inferact's launch followed closely behind RadixArk (originating from the commercialization of another well-known framework, SGLang). The latter recently secured funding led by Accel with a valuation of $400 million. Both companies come from the same Berkeley lab, and their consecutive investments reflect investors' willingness to invest heavily to secure faster and more cost-effective AI deployment technologies.

The Birth of an 800 Million Dollar Computing Power 'Nuclear Bomb': The Original vLLM Team Launches Inferact Aimed at the Global AI Inference Throne

As AI giants focus on large models, a computing efficiency revolution is quietly underway. The top open-source inference engine vLLM team has founded Inferact, an AI infrastructure company aiming to establish a new order in AI inference. This marks not only a technological leap but also a crucial step toward AI's practical application, drawing significant attention from capital markets.....

xLLM Community Unveils Open-Source Inference Engine on December 6th: Supports MoE, T2I, T2V Full Scenarios, Achieves Latency Below 20ms with Mooncake Caching Solution

xLLM Community's first offline Meetup on Dec 6 focuses on building open-source AI Infra ecosystem. Features xLLM-Core inference engine with sub-20ms P99 latency for MoE, text-to-image/video tasks, 42% lower latency and 2.1× throughput vs vLLM. Technical highlights include unified computation graph and Mooncake KV cache optimization.....

Energy Efficiency Increases by 228 Times! Peking University Team Develops a New Analog Computing Chip, Breaking the AI Computing Power Consumption Bottleneck

A research team from Peking University has developed an analog computing chip specifically designed for "Non-negative Matrix Factorization," significantly improving energy efficiency in processing large-scale data, providing an efficient and low-power solution for fields such as image analysis and recommendation systems.

a16z Leads! Inferact Enters with 150 Million USD, Kicks Off the AI Inference Cost Reduction Battle

Top Capital Support, Accelerating "Inference" Commercialization

The "Berkeley Twins" in the Inference Sector

Related Recommendations

The Birth of an 800 Million Dollar Computing Power 'Nuclear Bomb': The Original vLLM Team Launches Inferact Aimed at the Global AI Inference Throne

Samsung Displays Its Giant Screen Ace at CES 2026: 130-inch Micro RGB TV R95H Redefines Home Theater with AI and Color

xLLM Community Unveils Open-Source Inference Engine on December 6th: Supports MoE, T2I, T2V Full Scenarios, Achieves Latency Below 20ms with Mooncake Caching Solution

Energy Efficiency Increases by 228 Times! Peking University Team Develops a New Analog Computing Chip, Breaking the AI Computing Power Consumption Bottleneck

Be a Good Neighbor! OpenAI Promises to Pay for Data Center Electricity and Minimize Water Resource Consumption