Nous Research hits $1.5B valuation, launches enterprise AI agents

Nous Research has closed a $90 million Series B at a $1.5 billion valuation, the open-source AI company confirmed Wednesday, validating a fundraising round TechCrunch first reported.

The three-year-old startup, best known for its Hermes agent, drew a roster of heavyweight backers. Robot Ventures led the round. Nvidia, Union Square Ventures, Menlo Ventures, Samsung, and 1789 Capital — where Donald Trump Jr. is a partner — all participated. The new capital brings total funding to $158 million.

Hermes, an open-source agent developers and hobbyists have cloned more than 24 million times, now drives roughly 2.5% of global AI token usage, according to the startup's own estimates. That footprint has made Nous Research one of the most-distributed open AI stacks outside the major Western frontier labs.

What will the company do with the money?

The capital will fund a push into enterprise. Nous Research plans to launch "Hermes for Businesses," a product that lets companies deploy customized AI agents capable of running multi-step workflows while keeping proprietary data on their own infrastructure.

The company framed the offering as a way to give enterprises the benefits of the open-source Hermes stack without forcing them to host and maintain the models themselves. For Nous Research, the move opens a second revenue line on top of inference usage from its existing developer base.

How much revenue is the company already generating?

The Wall Street Journal reported that Nous Research was running at roughly $36 million in annualized revenue by mid-September 2026. The company expects to pass $100 million in annualized revenue before the end of 2026, the Journal said.

That trajectory implies revenue could nearly triple in roughly three months. Few open-source model vendors have grown at that pace. If the company hits its own forecast, the Series B will look like a bargain. If it misses, the gap between community usage and enterprise dollars becomes the central question for the next financing.

The projection, if realized, would also place Nous Research alongside a small set of open-model companies that have crossed eight-figure annualized revenue. It would still trail the major closed-API vendors, but it would put the company in a different financial category from a typical research-oriented open-source lab.

Why does an open-source footprint translate to a $1.5B valuation?

The 2.5% global token-share figure is the most important number in the announcement, and it underwrites the valuation. Usage, not downloads, is what translates into inference revenue.

A model that already handles 1 in 40 tokens served worldwide has a structural advantage when it sells to enterprises: the workflows developers have built on Hermes do not need to be rebuilt. That distribution also explains the strategic shape of the round.

Nvidia, Samsung, and 1789 Capital are not typical late-stage venture capital participants. Their presence suggests Nous Research is being valued not just as a software vendor but as infrastructure that other companies want access to.

The token-share metric also gives Nous Research a credible answer to a question that has dogged open-source model companies for years: how do you turn free downloads into a business? The company's answer is that its downloads are not really free. They are running, in aggregate, on a workload base that produces real inference revenue.

Who is on the cap table?

Robot Ventures led the round. The firm is known for early-stage bets on AI and crypto infrastructure companies.

Other participants in the round:

• Nvidia
• Union Square Ventures
• Menlo Ventures
• Samsung
• 1789 Capital

1789 Capital is a Washington, D.C.-based firm. Donald Trump Jr. is a partner there. The firm has become a recurring name in 2025 and 2026 AI financings, including deals involving politically connected operators.

Union Square Ventures and Menlo Ventures add traditional venture capital ballast to a round otherwise weighted toward strategic and politically connected investors. Samsung's participation signals interest from the device side of consumer electronics, where on-device AI is becoming a competitive battleground.

What does "Hermes for Businesses" actually offer?

According to Nous Research, the product allows companies to deploy customized AI agents that can handle multi-step workflows while keeping their data private and secure. The company did not disclose pricing, deployment options, or a launch date in its announcement.

The positioning — customizable, private, multi-step — tracks how enterprise buyers describe agent platforms when evaluating vendors. It also gives Nous Research a path to monetize a user base that has, until now, downloaded the models for free.

Enterprises are also the segment where AI vendors have faced the most scrutiny over data handling. A self-hosted or private-cloud deployment option is increasingly a baseline requirement for buyers in finance, healthcare, and government. Nous Research's framing in the announcement maps directly onto those procurement checklists.

What are the stakes for the open-model sector?

A $1.5 billion post-money valuation sets a new floor for well-funded open-model companies outside the top tier of Western frontier labs. It also signals that capital is still available for teams that pair a popular open release with a credible enterprise plan.

The bet is that open-weight agents will take share from closed-API incumbents over the next 18 to 24 months, particularly in regulated industries where data residency matters.

If Nous Research's revenue trajectory holds, the Series B will be remembered as a discount. If enterprise sales stall, the open-source community alone will be hard to monetize at the valuation investors just wrote.

For now, the company is shipping a product, raising on hard usage numbers, and converting its open-source community into a sales pipeline. The next test is whether the enterprise pipeline can convert at the pace the company has forecast.

The funding is in place. The distribution is proven. What remains is the harder problem of turning 24 million clones into paying contracts.