Arcee AI Series B Backs American Open-Weight Models

Arcee AI says it spent approximately $20M building its entire 2025 model lineup, including a 400B-parameter open-weight system. The San Francisco model lab has now raised a Series B at a valuation above $1B, turning that capital discipline into a larger mandate: build models that enterprises and public institutions can inspect, adapt, deploy and own.

Vista Equity Partners, Cambium Capital Management and Emergence Capital led the September 16, 2026 financing. AI10 Ventures, Hitachi, IAG Capital Partners, M12, Prosperity7 Ventures and Wipro participated. Arcee will use the round to train the next Trinity models, expand its work with the U.S. Department of Energy and national laboratories, and build products around the open-model stack.

What Arcee AI Announced

Arcee disclosed the round, investor group and valuation, but not the amount raised. Fortune reported that a source familiar with the transaction put the financing at at least $150M, while also reporting that Arcee declined to disclose the figure. The amount should therefore remain an attributed report rather than a company-confirmed deal value.

The distinction matters for honest capital accounting. Arcee's January 2024 seed announcement said the company had reached $5.5M in total funding, followed by a $24M Series A led by Emergence Capital in July 2024. Those disclosures establish $29.5M before the Series B, but they do not support a current cumulative total without a confirmed amount for the new round.

Arcee was founded in 2023 by Mark McQuade, Jacob Solawetz and Brian Benedict. Mark McQuade remains co-founder and CEO, while Lucas Atkins is CTO and Head of Research. The financing announcement did not disclose ownership changes, individual check sizes, security terms, board appointments or hiring targets.

The Bet Behind the Valuation

Arcee began by adapting and merging existing language models for enterprise use. In 2025, Mark McQuade chose to move upstream and pretrain a proprietary model family from scratch. Fortune reports that the decision committed 65% to 70% of roughly $30M then in the bank, a concentrated bet on whether a focused team could build American open-weight models without frontier-lab spending.

The company says the complete 2025 lineup cost approximately $20M across salaries, compute, data, infrastructure and operations. That figure is company-reported, not an audited comparison with other labs, but it gives the financing a clearer logic. Investors are backing a team that used a relatively constrained balance sheet to produce a model family, then made capital efficiency part of the product and market argument.

Trinity Large uses a sparse Mixture-of-Experts architecture with 400B total parameters and 13B active per token. The company's smaller Nano and Mini models are designed to preserve a common skill profile across edge, on-premises and cloud environments. Arcee distributes open weights under Apache-2.0 and also offers hosted access, allowing customers to choose where the models run and how much of the stack they control.

Why Open Weights Became Infrastructure

Open weights give an organization access to the model parameters needed to inspect, adapt and operate a system on infrastructure it controls. That can matter when a buyer has proprietary data, long procurement cycles, strict deployment boundaries or a need to preserve a specific model version. It does not remove the need for evaluation, security, governance or skilled operations, and it does not make every model equally capable.

Arcee's relationship with the Department of Energy turns that ownership argument into an institutional test. DOE confirmed that Genesis-Science-1 is being developed in partnership with Arcee as the first model in the Genesis Open Models Initiative. DOE scientists and national laboratories are expected to contribute reviewed materials, research tasks and evaluation, while Arcee leads model development and the governed execution environment.

Genesis-Science-1 remains in development. The program is intended to support scientific workflows that may involve code, simulation, data, tools, retries and human review, leaving a reproducible record of the work. It is a meaningful partnership and a demanding use case, but it is not evidence that the model has already delivered validated scientific outcomes.

What the Series B Is Funding

Arcee says a new Trinity generation is already in training. The plan spans models small enough for phones and laptops, larger systems for scientific and developer workloads, and software for customization, evaluation, deployment and production operations. That breadth asks the company to behave as both a model lab and an enterprise software provider.

The investor group reflects both parts of that ambition. Emergence Capital backed Arcee's earlier enterprise model work, M12 and strategic participants connect the company to large technology and services ecosystems, and Vista brings a broad enterprise software portfolio. Those relationships can create distribution and operating feedback, but the announcement does not quantify commercial commitments or portfolio adoption.

Public sources also do not disclose Arcee's revenue, customer count, retention, margins, current cash balance or independently audited benchmark results. The company's reported development efficiency is important because it explains how Arcee reached this moment. The Series B will test whether the same discipline survives larger training runs, a wider product surface and support obligations across private companies and public research institutions.

What Arcee AI Must Prove Next

Arcee has already changed its own center of gravity, moving from improving other organizations' models to training and operating a family it controls. The next stage requires proving that model ownership creates enough practical value for enterprises and institutions to accept the work that comes with it, including evaluation, deployment, security, governance and lifecycle management.

The valuation gives Arcee more room to train, hire and build products around that thesis. It also places the company inside a harder operating record, where capital efficiency must coexist with scientific rigor, enterprise reliability and model quality across several sizes. The Series B will be measured through the systems customers and researchers can keep using on their own terms after the announcement has left the room.