Open-weight model developer Arcee AI reaches $1B-plus valuation with new funding

Open-weight artificial intelligence model developer Arcee AI Inc. said today it has raised an undisclosed amount of cash in a new funding round that lifts its valuation to more than $1 billion.

The Series B round was led by Vista Equity Partners, Cambium Capital and Emergence Capital, and saw participation from A10 Ventures, Hitachi, IAG, Microsoft Corp.’s M12, P7 and the AI consultancy firm Wipro Ltd. One source told Fortune that the round was worth at least $150 million.

San Francisco-based Arcee is pushing to become one of the U.S.’s top developers of open-weight models. Those are AI models that publicly release their core numerical parameters, making it possible for anyone to download them, modify them and run them on their own infrastructure.

Co-founder and Chief Executive Mark McQuade told Fortune that the company is developing open-weight models because he believes that American companies should have more control over the AI systems they use. “Organizations should not have to choose between the capabilities of a frontier model and the ability to control the technology at the center of their work,” he explained.

Arcee is unusual, because most U.S. AI developers have opted for a proprietary approach, in which they keep the underlying code for their models private, and often dictate where those models can be run. In contrast, most Chinese AI developers have embraced the open-weights concept, developing extremely powerful and highly accessible models that match the capabilities of the most advanced systems built by the likes of OpenAI Group PBC, Anthropic PBC and Google LLC.

McQuade told Fortune that his goal is for Arcee to catch up with China’s leading AI developers. “The U.S. is far ahead in closed-source, but it kind of dropped the ball on open-source,” he said.

Arcee has developed a family of models called Trinity, which were all released in the last year. The most powerful is Trinity Large, which has 400 billion parameters and 13 billion active for each token.

What’s especially interesting about the startup is that it developed the Trinity models at a much lower cost than what most of its closed-source counterparts have spent. The company spent about $20 million developing those models last year, with that money covering everything from the compute infrastructure to the data and operations and the developer’s and engineer’s salaries.

According to McQuade, Arcee only had about $30 million available to it when it first decided to start building its own models. “I said, ‘Let’s do it’ and I bet the company on it,” he told Fortune. All told, almost 70% of its total capital went towards developing the Trinity models.

Before developing its own models, Arcee was involved in post-training and fine-tuning AI models, providing infrastructure and tools to companies that want to adapt existing AI models for their own purposes. The startup was founded in 2023 after McQuade left his old employer, Hugging Face Inc., which is now being acquired by Nvidia Corp.

Arcee will use the money from today’s round to train a new generation of its open-weight Trinity models, which are already under development, McQuade said. It will also expand its work with the U.S. Department of Energy and 17 national laboratories. It’s collaborating with them on Genesis-Science-1, which is an open-weight model for scientific computing and research tasks. The startup hasn’t forgotten about its model post-training roots either, and it will continue to invest in its infrastructure and products for companies that want to customize, test and deploy open-weight models.

Monti Saroya of Vista Equity Partners said Arcee and its team have displayed a rare combination of solid technical execution and capital efficiency, building the Trinity models at an impressively rapid pace and low cost. “As enterprises increasingly look for AI systems they can control, customize and deploy on their own terms, we believe Arcee is building critical infrastructure,” he said.

Image: Arcee