Arcee AI, founded in 2023 by Mark McQuade, a former early employee at Hugging Face, has turned a relatively modest training budget into a billion-dollar valuation. The company specializes in open-weight AI models, systems that let businesses run advanced AI locally without sending proprietary data to third parties like OpenAI or Anthropic.
A bet on building from scratch
In 2025, McQuade shifted Arcee's strategy away from simply fine-tuning existing models and toward building proprietary ones from the ground up. The company put roughly $20 million, about 65% to 70% of its available $30 million in capital, into training four models over several months, including Trinity Large, a 400-billion-parameter model released in early 2026. "We can comfortably say we're the most efficient lab in the world based on what we've done," McQuade has said of the approach.
From efficient training to a billion-dollar round
That efficiency argument found real backing: Arcee closed a Series B round in 2026 at a $1 billion pre-money valuation, raising at least $150 million. Vista Equity Partners, Cambium Capital and Emergence Capital led the round, joined by Microsoft's M12, AI10 Ventures, Hitachi, IAG, P7 and Wipro. The company says its models have outperformed Meta's Llama 3 and benchmark competitively against Mistral and Chinese rivals, and it has also begun partnering with the U.S. Department of Energy, extending its reach beyond the commercial AI market it built its name in.