Nous Research Raises $90M to Take Hermes AI to Enterprise
On October 7, 2026, Nous Research closed a $90 million financing round at a $1.5 billion valuation. The goal is to take Hermes, the most widely used open-source AI agent in the world, into corporate deployments. The round was reported first by the Wall Street Journal and confirmed in an official announcement from Nous CEO Dillon Rol.
The investors read like a who’s-who of AI infrastructure: NVIDIA, Microsoft M12, Samsung, Robot Ventures, Union Square Ventures, Y Combinator, and Menlo Ventures. Together they are making a direct bet that enterprises will route spending toward open-source AI systems rather than locking in to a single proprietary provider.
What Hermes Actually Is
Nous Research was founded in 2023 by a small team that believed open-source AI could compete with much larger labs. Starting from fine-tunes of Meta’s Llama models, the team pioneered synthetic data generation, defined open-source tool-calling standards, and published work showing that small development teams could build models competitive with frontier labs on key benchmarks.
In early 2025, the team shifted focus to an agent stack. The result was Hermes Agent: a local and cloud-based AI agent that handles search, coding, and image understanding. Its defining characteristic is that it learns from use. The more it works inside a session, the more context it accumulates. It remembers projects, writes reusable skills as it completes tasks, and carries that knowledge forward across sessions.
Nous released Hermes Agent under an MIT license in February 2026. The adoption was immediate. By October 7, the repository had been cloned more than 24 million times, and Nous estimates the agent accounts for roughly 2.5 percent of all global token usage. That scale of organic adoption, without a sales team and before a commercial product existed, is what attracted the $90M round.
The Enterprise Product: Hermes for Business
The funding announcement formally introduces Hermes for Businesses. Rol laid out the value proposition in plain terms. The enterprise version will:
- Let companies choose the best models at the best prices for each specific task, rather than routing everything through a single provider
- Scale AI usage safely and privately, with the company controlling where data goes and which models process it
- Build institutional knowledge from the bottom up, converting the work of individual employees and agents into a shared, searchable corpus owned by the company
- Give organizations “dominion over their institutional knowledge,” meaning the company controls what it can do with that knowledge, including fine-tuning open-source models on it
The architecture is deliberately model-agnostic. An enterprise running Hermes for Business is not betting on one lab. It can route a coding task to a specialized code model, a legal review to a fine-tuned legal model, and a customer brief to a cost-efficient general-purpose model. When a better option becomes available, the agent picks it up without vendor negotiation.
How Open-Source Competes With Proprietary in the Enterprise
The common criticism of open-source enterprise AI is that it trades vendor support and reliability guarantees for cost savings. Hermes challenges that framing on two fronts.
First, reliability: the Hermes codebase has thousands of external contributors who have stress-tested it across a range of environments that no proprietary vendor can replicate in a closed development cycle. The MIT license means those contributions accumulate rather than branching.
Second, transparency: closed AI systems cannot be audited in the same way. As regulatory scrutiny of AI systems grows, especially following the Senate’s September 2026 hearings on rogue agent deployments, enterprises face increasing pressure to demonstrate that they understand what their AI systems are doing. An MIT-licensed agent is auditable by definition. A proprietary SaaS agent is not.
The table below compares the main deployment approaches for enterprise AI agents on dimensions that matter to CTOs and AI operations teams.
| Approach | Model Control | Data Privacy | Cost Structure | Auditability |
|---|---|---|---|---|
| Proprietary cloud AI (ChatGPT, Claude) | None | Shared via SaaS | Fixed per-token pricing | Closed |
| Self-hosted open-weight models | Full | Maximum | Variable (compute + engineering) | Full |
| Hermes for Business | Full | User-defined | Best-available model per task | Full (MIT) |
| Enterprise SaaS agents (Salesforce Agentforce, etc.) | None | SaaS-controlled | Per-seat or per-action pricing | Closed |
The Nous bet is that the third row is what large enterprises will increasingly want as they accumulate AI operational experience and begin to feel the costs and risks of vendor lock-in.
Why the Timing Makes Sense
Several trends converged to make October 2026 the right moment for this round.
First, the enterprise AI market has matured past the pilot phase. Most large enterprises are running AI in production. The questions are now about cost, control, and compliance rather than capability. Open-source addresses two of those three directly.
Second, the model landscape has fragmented. A year ago, one or two providers held clear leads. Today, competitive models exist across price tiers from multiple labs, open-weight and proprietary alike. An agent architecture that lets the enterprise choose across this landscape is more valuable than it was when a single model dominated.
Third, the political and regulatory environment has shifted. The Senate’s September hearings on rogue AI agents, the three bipartisan legislative proposals that followed, and the EU AI Act’s accountability requirements have all put enterprise AI governance under a spotlight. Open-source systems that produce auditable logs and allow real-time inspection of reasoning are better positioned in that environment than closed ones.
What This Means for Enterprise AI Teams
If Hermes for Business gains the same adoption velocity in the enterprise that Hermes Agent found among developers, the competitive dynamics for AI operations change in a significant way.
The current state: most enterprises buy AI capability through a combination of platform seats (Microsoft Copilot, Google Workspace), API access to one or two major models, and point solutions. Switching costs are high because data, fine-tunes, and integrations accumulate on those platforms.
The Hermes scenario: an agent layer that runs inside the enterprise’s own infrastructure, learns from the company’s own workflows, and retains all outputs as portable institutional knowledge. Switching models does not mean losing context. Adding a new model does not require a new contract or a new deployment pipeline.
This is consistent with a broader shift in enterprise AI strategy, visible in the recent funding wave for agent infrastructure companies, toward architectures that give the enterprise control over the intelligence layer rather than renting it by the token. Teams rebuilding their operational intelligence approach or evaluating how to make their GTM motion more autonomous will want to track Hermes for Business closely as it moves from announcement to general availability.
For Enera’s clients working on AI-native transformation, the arrival of a credible open-source enterprise AI agent changes the build-vs-buy calculus. The question is no longer whether open-source can match proprietary performance. At 2.5 percent of global token usage, Hermes has already answered that. The question is whether an enterprise is ready to own its intelligence stack, or whether it would rather lease it.
Sources: Nous Research official announcement, Wall Street Journal, The Next Web, Dealroom