Paris-based Dust raises $40m Series B from Abstract and Sequoia to make enterprise AI a team effort

Dust, the enterprise AI platform with offices in Paris and San Francisco, has raised $40m in a Series B round co-led by Abstract and Sequoia. Snowflake Ventures and Datadog also took part, and the company says its total funding now exceeds $60m.

The round puts a French-founded company at the centre of one of the busiest fights in software: who owns the layer where staff and AI agents actually do their work. Dust is betting that the answer is shared, not personal.

The deal was announced on 18 May. Chief executive Gabriel Hubert told reporters that Abstract and Sequoia led the round together, although some early coverage named only Abstract as lead.

Against "single-player" AI

Dust's pitch starts from a complaint many operators will recognise. In most companies, employees use AI assistants on their own. One person's prompts, documents and results stay in that person's chat history, and a colleague starting the same task tomorrow begins from zero. Dust calls this "single-player AI".

Its product connects a company's knowledge, internal tools and teams in one place. Employees build agents, share them with colleagues and run them inside real workflows, with each agent drawing on the same context as the people around it. Teams can pick different underlying models for different jobs, and humans keep control over what agents are allowed to touch.

Hubert framed the shift as a systems problem rather than a model problem. In his view, the next gains will not come from a better chatbot but from software that gives people and agents shared, governed access to the same information and capabilities.

The numbers Dust is putting forward

The company shared a set of usage figures alongside the round. These come from Dust itself and have not been independently audited:

• more than 3,000 organisations use the platform;
• 70% of users at customer companies are active each week;
• it says it lost no customers in 2025;
• over 300,000 agents have been built and deployed on Dust.

The weekly usage figure is the one investors are likely to watch most closely. Enterprise AI tools are often bought in bulk and then left idle, so a high share of regular users is a stronger signal than seat counts. A zero-churn claim for a full year is also unusual at this scale, though the company has not said how it defines churn.

Where the $40m goes

Dust said the money will fund three areas of product work. The first is agents that improve themselves over time as they are used. The second is what the company calls collaboration primitives: the building blocks that let several people and several agents work on the same task. The third is governance infrastructure, meaning the permissions, controls and oversight that large customers need before they let agents act on company data.

That last item matters for the investor list. Snowflake and Datadog both sell to the same enterprise data and engineering teams that decide whether an AI platform can be trusted with internal systems. Having their venture arms on the cap table gives Dust closer ties to two companies whose products sit next to its own in many customers' stacks.

Founders who have sold to Stripe before

Hubert runs the company with Stanislas Polu, the chief technology officer. The pair previously built Totem, which they sold to Stripe. Hubert later worked at Stripe and at Alan, the French health insurance start-up, before returning to found Dust with Polu.

In past interviews, Hubert has argued that Dust should stay a horizontal platform rather than build narrow tools for one department. He has also said that most companies use AI to decorate old processes instead of redesigning them, and that firms need in-house "AI operators" who can rebuild workflows around agents.

The Series B gives that argument more runway. Dust now has to show that a general-purpose platform can hold its ground as large model providers and established software vendors push their own agent products into the same enterprise accounts. Its next test is whether the self-improving agents and governance tools funded by this round reach customers quickly enough to keep that 70% weekly usage figure where it is.