The least interesting thing about Nous Research’s $1.5 billion valuation is the $1.5 billion. Everybody’s treating the number as the headline, as proof that open-weight AI labs have finally earned their seat at the enterprise table. I’d argue the opposite. The valuation tells you what investors believe. It tells you nothing about whether “Hermes for Businesses” works, and right now that’s the only question that matters to anyone who has to actually deploy this stuff.
Let’s lay out what’s confirmed. Nous Research has confirmed a $1.5 billion valuation on a $90 million Series B led by Robot Ventures, with Nvidia, Union Square Ventures, Menlo Ventures, Samsung, and 1789 Capital joining in. That brings total funding to $158 million for a company that, by TechCrunch’s reporting, is three years old. The money goes toward an enterprise push called Hermes for Businesses, where companies deploy customized agents that handle multi-step workflows while keeping data private and secure.
That’s it. That’s the product description. One sentence doing an enormous amount of work.
What “multi-step workflows” actually means in practice
I review agents for a living, which mostly means watching them fail in boring, predictable ways. Multi-step is where almost all of them break. An agent that can summarize a document is a parlor trick at this point. An agent that can read a document, pull three fields out of it, cross-reference those fields against an internal system, make a judgment call, and then write the result somewhere that matters — that’s the thing every vendor promises and very few deliver without a human babysitting each step.
So when the announcement says multi-step workflows, my first question is: how many steps before it drifts? My second is: what happens when step four fails? Does it retry, does it halt, does it hallucinate a success and move on? Those three failure modes are the entire difference between an agent you can trust with invoicing and an agent that generates a very confident mess you’ll spend a week untangling.
None of that is in the announcement. Which is fine, announcements aren’t documentation. But it means the honest review of Hermes for Businesses today is “unknown,” and I’d rather say unknown than pretend a funding round is evidence of quality.
The privacy angle is the real pitch
Here’s where I’ll give Nous genuine credit. The data privacy and security framing isn’t marketing garnish for a company built on open models — it’s the whole strategic logic. If you’re a bank or a hospital system, the blocker on agent deployment has rarely been capability. It’s been that your legal team will not sign off on piping customer records through an API you don’t control, running on weights you can’t inspect, governed by terms that change quarterly.
An open-weight lab selling enterprise agents has a structurally different answer to that objection. You can run it where you want. You can audit what you’re running. You’re not renewing a dependency on a vendor who might reprice you next year. That’s a real differentiator, and it explains why a group like this one attracted Nvidia and Samsung alongside the traditional venture names.
It also explains the valuation in a way the agent demo never could. Investors aren’t paying $1.5 billion for a workflow tool. They’re paying for the possibility that open weights become default enterprise plumbing, and that the company maintaining the most-used open agent gets to charge for the privilege of making it boring and compliant.
A few numbers worth treating carefully
Reporting around this round is messier than the clean headline suggests, and if you’re making decisions based on it, you should know that:
- TechCrunch reported in July 2026 that the round was being finalized at “at least $75 million.” Dealroom logged it as a $75 million Series B. The confirmed figure came in at $90 million. Rounds grow during close — normal, but worth tracking which number a given article is using.
- Third-party funding trackers cite $36 million in annualized revenue by mid-September 2026 with $100 million projected by year-end, attributed to Wall Street Journal reporting. Annualized and projected are doing a lot of lifting in that sentence. A projection is a plan, not a result.
I’m not suggesting anyone’s inflating anything. I’m suggesting the gap between $36 million booked and $100 million hoped-for is exactly the gap that enterprise agent products either close or don’t, and it closes on retention, not on launch-day enthusiasm.
My actual take
Nous has earned attention. An open-source lab reaching this scale on this timeline is a legitimately interesting outcome, and the privacy-first enterprise angle is better positioned than most of the agent startups I test. But a valuation is a prediction, and I don’t review predictions.
Give me pricing, give me step limits, give me a failure-handling story, and give me access. Then I’ll tell you if Hermes for Businesses is worth your procurement cycle. Until then, the number is just a number, and anyone telling you otherwise is reading a press release back to you.
🕒 Published: