Nous Research nears $75 million round at $1.5 billion valuation

Why it matters

Nous is testing whether open-source AI labs can command unicorn valuations before they show public revenue, using developer adoption as the wedge into agents and compute.

Jeffrey Quesnelle (@theemozilla), Karan Malhotra, Ryan "Teknium" and Shivani Mitra's Nous Research is finalizing a funding round of at least $75 million at a $1.5 billion valuation, TechCrunch reported on July 13th, citing three people with knowledge of the deal.

The reported round is led by Robot Ventures, with significant participation from USV and other investors, according to TechCrunch. It has not been announced by Nous Research (@nousresearch), and the useful verb here is "finalizing," rather than "closed." If the financing lands at the reported terms, Nous will have converted an unusually visible open-source agent project into a unicorn-priced company before the market has much public evidence of its revenue, customer base or enterprise adoption.

That is the sharp edge of the deal. Nous is raising against the two things investors are currently willing to pay up for: agent distribution and independence from the closed model stack. Hermes Agent, the company's open-source agent harness, has the kind of developer gravity that normally precedes commercialization. Its GitHub repository listed about 214,000 stars and 39,800 forks when checked. The project describes Hermes as a self-improving agent with a built-in learning loop and persistent memory and skills.

Quesnelle's path into this is worth keeping in view. His personal site lists a master's degree in computer science from the University of Michigan-Dearborn, an undergraduate degree in computer science and mathematics from Oakland University, and older open-source projects including nds4droid, a Nintendo DS emulator for Android. Mitra's site says she was Nous's founding CEO and president, built the team and business vision, and helped raise $70 million from 2023 to 2025. TechCrunch names Quesnelle, Malhotra, Teknium and Mitra as the 2023 founding team.

The money is buying an open-source distribution wedge

Before the reported round, Nous had raised $70 million from investors including Paradigm, Robot Ventures, North Island Ventures, OSS Capital and Balaji Srinivasan, TechCrunch reported, citing Crunchbase. The company's prior financing record also contains a timing wrinkle: TechCrunch describes the new round as coming less than three months after a $50 million Series A, but the linked SiliconANGLE story is dated April 25th, 2025. On the documented record, the Paradigm-led Series A was reported more than 14 months before the July 13th TechCrunch report.

The older round matters because it shows the original investor thesis around Nous. SiliconANGLE reported that Paradigm almost entirely funded the $50 million round, and it framed the company around decentralized AI training rather than consumer-facing agents. Malhotra told Fortune, as quoted by SiliconANGLE, that the crypto incentive mechanism was less a donation model than a transaction model for idle compute. He also acknowledged the baggage around crypto and said Nous did not want to be "bogged down" by that view while positioning itself as a serious research lab.

That tension has not gone away. Nous's public homepage says its mission is to advance human rights and freedoms by creating and spreading open-source language models, supporting their unrestricted availability, and furthering scientific and popular understanding. Its careers page puts the commercial version more plainly: Nous says it is building open-source, humanistic AI to outpace closed tech giants, and lists open roles including forward deployed engineer for Hermes Agent Enterprise.

That job title is a clue. GitHub stars create pull with developers, but enterprise AI budgets flow toward deployed workflows, integrations, support, security and measurable productivity. TechCrunch reported that Hermes can run locally or on a virtual private server, communicate through Telegram and Discord, and use built-in skills such as web search, coding and image understanding. It also reported hosted Hermes tiers ranging from $20 to $200 per month. The reported new funding would give Nous more room to turn a developer phenomenon into a paid product line.

Psyche is the other half of the bet

Hermes is the visible product. Psyche is the infrastructure argument. Nous published its Psyche network architecture on May 14th, 2025, describing a distributed training system that coordinates underused hardware and uses DisTrO and DeMo methods to reduce data transfer. The company says Psyche coordinates through Solana while moving training gradients through a separate peer-to-peer mesh.

In a later Hermes 4.3 post, Nous said it trained the model both on Psyche and through a traditional centralized approach. The company claimed the Psyche run averaged 144,000 tokens per second across 24 Psyche nodes and achieved equivalent throughput to centralized training. Those are company-reported claims, but they explain why Nous can attract capital from crypto-native and open-source backers while still speaking to the mainstream AI market's main constraint: compute concentration.

The market backdrop is forgiving for that story. In May, Cognition said it raised more than $1 billion at a $26 billion valuation for Devin, its software engineering agent. In March, Replit said it raised $400 million at a $9 billion valuation and introduced Agent 4. In February, OpenAI introduced the Codex app. Nous is smaller and less enterprise-proven than those companies, but the reported valuation puts it inside the same market repricing of agent infrastructure.

The open questions are commercial, not technical

Nous has technical credibility, open-source reach and a clear ideological center. It has less public evidence on the metrics that usually support a $1.5 billion private valuation: revenue, retention, named enterprise customers and gross margin profile. TechCrunch's report does not disclose annual recurring revenue, customer count or a signed term sheet status. The company's own public materials emphasize mission, models, distributed training and recruiting rather than operating metrics.

There is also a security caveat around production use. The National Vulnerability Database lists CVE-2026-9350, a Hermes Agent vulnerability affecting versions up to 2026.4.16 in the Batch Runner component, with missing authorization and public exploit availability. The NVD entry is a reminder that local and server-running agents are not ordinary developer toys once they touch terminals, messages and workflow automation.

The best reading of the reported round is that investors are trying to buy into a rare combination before the business is fully legible: a founder-led open-source AI lab, a widely watched agent repo, a distributed training thesis and a user base that could be steered toward hosted and enterprise products. The hard part begins after the money arrives. Nous has to prove that open-source distribution can become durable revenue without dulling the openness that made developers care in the first place.