❯ Arcee AI Raises at Least $150M Series B After Building an Open-Weight Model Family for $20M

Bounded amountArcee AI completed a Series B but did not disclose its size in the company announcement. Fortune reported the round at at least $150 million, with a post-money valuation above $1 billion. Vista Equity Partners, Cambium Capital and Emergence Capital led, joined by AI10 Ventures, Hitachi, IAG, Microsoft’s M12, P7 and Wipro. Arcee trains open-weight foundation models from scratch and supplies tools for tuning, evaluation, deployment and inference in customer-controlled environments.

From tuning to pretrainingFounded in 2023, Arcee initially focused on model compression and post-training, then raised a $24 million Series A in 2024. Less than a year ago it decided to pretrain models itself, scaling Trinity in six months from a 4.5 billion-parameter dense model to a 400 billion-parameter mixture-of-experts model. The company says its entire 2025 lineup cost about $20 million, including people, compute, data, infrastructure and operations.

An open stackTrinity Large carries a permissive license, while Arcee aims to cover the path from pretraining through production inference and models ranging from laptops to scientific workloads. Funding will expand work with the U.S. Department of Energy and national laboratories; Genesis-Science-1 is an initial scientific deployment. The investor group mixes enterprise-software funds with strategic backers such as Hitachi, Microsoft and Wipro, adding distribution and sovereign-control logic to the deal.

Efficiency betThe round prices model ownership and bets that training efficiency can offset a compute disadvantage. If a $20 million program can keep producing competitive models, enterprises gain an alternative to closed APIs. Open weights, however, weaken pure access fees, forcing Arcee to monetize customization, tooling and deployments. A larger Series B funds more pretraining while making benchmark gaps with leading labs impossible to avoid.

▪ SIGNALThe open-model business is moving from downloads to who can train efficiently and complete deployment inside an enterprise’s own environment.