Nous Research bets its $90 million Series B on companies paying for controls around free Hermes code

If the $1.5 billion is a post-money figure, the $90 million bought the new investors about 6% of Nous Research ^[17]. The price is roughly 9.4 times everything the company has raised ^[18], and it was set about eight months after Hermes Agent's February launch ^[16]. Spread across the agent's downloads and clones, it comes to no more than about $62.50 each ^[19]. A download is not a user. The other headline figure, that Hermes drives about 2.5% of global token usage, comes from the company's internal estimates, according to Ventureburn ^[11].

The money is earmarked for the agent, a business version and a planned mobile app ^[12]. According to Crypto Briefing, the business version, Hermes for Businesses, adds single sign-on, auditing and workspace controls, the features corporate IT departments insist on before software reaches a company laptop ^[5]. The code underneath is MIT-licensed, so anyone can use, modify and redistribute it ^[4]. Nous says it will stay open source ^[13]. Whatever Nous charges for has to sit on top of free code that customers can already run on their own hardware or through hosted options ^[14].

The case that open-source agents now compete with closed vendors for corporate budgets holds for part of the budget. Hermes lets a company choose its underlying model on price and operational goals, and run it locally or in the cloud ^[9]. The 2.5% share therefore measures work done by models the customer picks, and those need not be Nous's. In my view Nous is bidding for the agent line of an AI budget, while the model line goes to whichever provider the customer chooses. Ventureburn's counter-case is that running models locally keeps proprietary data, and the spending, inside the company ^[9].

If Hermes for Businesses converts part of the clone base into paid seats, that token share becomes a sales pipeline. If companies instead run the free code behind their own login systems, Nous collects usage without revenue. A third route runs through tokens. The roughly $70 million earlier package mixed equity with SAFTs ^[7], contracts that buy rights to tokens a project expects to issue later ^[8]. The company did not disclose pricing, revenue or customer numbers for the business product. I'd weight the first outcome most, because the features being built are the ones IT reviews demand ^[5]. A disclosed customer count that stays small against more than 24 million clones ^[10] would prove that view wrong.

CEO Dillon Rolnick, according to Ventureburn, called $90 million modest by the standards of current AI funding and said the Nous team has always delivered more with less ^[15]. The stated uses fit that description: an agent, an admin layer and a phone app ^[12]. Training a new model is not on the list.