Arcee AI raises $150M Series B at unicorn status for enterprise small models

Arcee announced the round on September 16 but did not disclose the amount. A source told Fortune it was at least $150M, at a $1B pre-money valuation. That implies roughly $1.15B post-money if all of it is new primary capital. Microsoft's M12, AI10 Ventures, Hitachi, IAG, P7 and Wipro also participated. Emergence led the $24M Series A, and total funding before this round was just under $50M, so the Series B roughly triples the capital raised. The company has moved beyond the small-model framing. Founded in 2023 by ex-Hugging Face employee Mark McQuade, it began by fine-tuning existing models and now trains open-weight models from scratch. Its Trinity family includes the 400B-parameter Trinity Large. Arcee says it trained four models for about $20M in total, and claims results that beat Llama 3 and match Mistral. Proceeds go to new Trinity models, expanded work with the U.S. Department of Energy, and tooling to customize and deploy models across Vista portfolio companies. For founders, the round tests whether a US lab can sell open weights that customers run on their own infrastructure. Arcee's cost claim, if it holds, puts competitive pretraining within reach of well-funded startups. The Vista channel also shows a distribution route: a PE owner with dozens of software holdings as built-in buyers.