Nous Research has raised $90 million to commercialize an agent that customers can run with models of their own choosing. That design gives the company a different route into enterprise AI: earn money from making an open ecosystem useful and dependable, while allowing users to retain considerable control over the underlying technology.
TechCrunch reported that Robot Ventures led the Series B at a $1.5 billion valuation, bringing total funding to $158 million. Nous confirmed the $90 million raise in its October announcement and named backers including Nvidia, Microsoft's M12, Samsung, Robot Ventures, Union Square Ventures, Y Combinator and Menlo Ventures.
The valuation comes from reporting rather than a public financial filing. Participation by large technology investors provides capital and relationships, but the disclosed round does not establish a material earnings contribution for those investors. Its direct significance is to Nous's ability to develop and distribute its products.
An open license changes the competitive bargain
The company's fundraise note supplies an important detail: Hermes Agent was released under the permissive MIT license in February. Users choose both where the agent runs and which models it uses. Nous says it can retain project knowledge across sessions and create reusable skills as it works.
That architecture can appeal to businesses trying to avoid dependence on one model provider. It also shifts the competitive challenge. If customers can inspect and reproduce the software, a durable commercial advantage must come from execution, integration, service quality and trust rather than exclusive access to the core code.
Nous says the new funding will help deliver Hermes for Businesses, including tools intended to let customers select models according to price and suitability. Lower customer spending is a product promise, not a demonstrated saving across enterprise deployments. Different tasks can have different accuracy and supervision requirements, so the cheapest model per token may not produce the lowest cost per successfully completed task.
Usage is a starting point, not the revenue base
Nous reports more than 24 million repository clones and estimates internally that Hermes drives about 2.5% of global token usage. The company identifies the latter as its own estimate. Neither number should be treated as an independently audited market share.
Clones are also not equivalent to distinct active organizations or paying seats. They can reflect repeated downloads and developer activity. Heavy token consumption demonstrates demand for underlying model computation, but that spending need not accrue to the company providing an open agent.
TechCrunch cited Wall Street Journal reporting of roughly $36 million in annualized revenue in September and a company expectation of reaching $100 million by year-end. Annualized revenue extrapolates a recent pace; it is not the same as revenue recognized over a completed year. The expected increase remains a target.
For the business model, the more consequential issue is how much customers will pay for the enterprise layer after the software is available freely. Revenue can grow with support, managed deployment and organizational controls, but those activities also incur costs. The round disclosed no audited margin or customer-retention evidence that resolves that trade-off.
The ability to choose models can also change how enterprise buyers measure value. A successful task may require several model calls, retries and human checks. Reducing the price of a single call saves little if a cheaper model increases failures or supervision time. Conversely, routing routine work to a less expensive model while reserving stronger models for difficult tasks could improve the total cost of delivery.
Hermes's persistent project knowledge is relevant to that proposition because context gathered once might be reused across tasks. The company's description establishes the intended capability; it does not quantify realized savings or reliability. Enterprise evidence would need to connect repeated usage to completed work and the full cost of supporting it.
The reported valuation is demanding relative to the existing revenue pace. Dividing $1.5 billion by the reported $36 million annualized revenue gives about 41.7 times that run rate. If the $100 million year-end pace were achieved, the same calculation would fall to 15 times. This is BTI arithmetic using reported private-round figures, not an enterprise-value multiple or a comparison with audited annual sales.
The contrast shows how much the financing narrative depends on execution. A faster revenue pace can make an unchanged valuation appear less expensive without establishing better unit economics. The calculation also says nothing about cash, debt, investor preferences or the different rights attached to private securities. It is a scale check, not a fair-value estimate.
The funding gives Nous more room to turn developer distribution into a commercial product. Its enterprise rollout will test whether openness becomes a route to customer trust and recurring revenue, rather than simply a mechanism for spreading software whose economic value is captured elsewhere.