Dust bets enterprise AI is multiplayer; $16M Series A then $40M Series B, 300k agents

Paris startup bets companies need shared, compounding AI agents, not one private assistant per employee; $60M+ raised, 300k agents deployed

Dust

它在赌什么Enterprise AI compounds only when humans and agents share one workspace: build specialised, company-data-grounded agents instead of one private assistant per employee在扩

做的是什么生意

Dust is a platform for building, deploying and managing AI agents that work inside a shared enterprise workspace, connected to company data and existing tools

启动资金：$5M seed (2023); $16M Series A (Jun 2024); $40M Series B (May 2026); $60M+ total

起因

Dust was founded in Paris in 2022 by Stanislas Polu and Gabriel Hubert, who met at Stanford, built marketing-analytics company TOTEMS (backed by Seedcamp, acquired by Stripe in 2014) and were among Stripe's first 150 employees. Polu then spent years in the US, including as an OpenAI software engineer, while Hubert led healthtech unicorn Alan's product team; they launched Dust in early 2023.

经过

A €5M seed from Sequoia Capital, Seedcamp and Connect Ventures in 2023 was followed by a €15M ($16M) Series A in June 2024 from the same backers. The platform — a Slack-like workspace where non-developers build assistants, 'one assistant for one use case' — passed €1M ARR with more than 250 enterprise customers including Alan, Qonto and Pennylane.

新闻Sifted: founders on €15M raise 新闻TFN: $16M Series A led by Sequoia

结果

In May 2026 Dust closed a $40M Series B co-led by Abstract and Sequoia, with Snowflake and Datadog participating, taking total funding past $60M. It reported 3,000+ organisations, 300,000+ agents deployed, 240% net revenue retention in 2025 and $20M+ ARR, with the US its fastest-growing market; it remains unprofitable by choice.

新闻Sifted: $40M Series B 报道Seedcamp: multiplayer AI thesis

背景

Dust was founded in Paris in 2022 by Stanislas Polu and Gabriel Hubert, longtime co-founders: they met at Stanford, built marketing-analytics company TOTEMS (backed by Seedcamp), sold it to Stripe in 2014 and were among Stripe's first 150 employees. Polu then spent years in the US, including as an OpenAI software engineer, while Hubert led healthtech unicorn Alan's product team; they launched Dust in early 2023.

The thesis was contrarian: most enterprise AI assumed one assistant per employee, which Dust called 'single-player AI' — insights stayed trapped in private chats and never compounded. Its platform let teams build specialised agents connected to company knowledge and existing tools, sharing a workspace so that when one agent improves, the whole organisation benefits.

A €5M seed from Sequoia, Seedcamp and Connect Ventures in 2023 was followed by a €15M/$16M Series A in June 2024 from the same backers. The Slack-like platform, where non-developers build 'one assistant for one use case', passed €1M ARR with 250+ enterprise customers including Alan, Qonto and Pennylane.

In May 2026 Dust closed a $40M Series B co-led by Abstract and Sequoia, with Snowflake and Datadog participating, taking total funding past $60M. It reported 3,000+ organisations, 300,000+ agents deployed, 240% net revenue retention in 2025 and $20M+ ARR, with the US as its fastest-growing market. It is unprofitable by design: 'The raise is about acceleration, not survival.'

这件事要成立，得有什么

The multiplayer thesis was a wedge big-model labs ignored: Dust competed on the collaboration layer around models rather than on model quality
Shared workspaces make agents accumulate context and usage, showing up as retention — 240% NRR, 90%+ monthly active adoption — instead of one-off chatbot experiments
Founder history de-risked the company for investors: ex-OpenAI plus ex-Stripe/Alan covered both the model frontier and enterprise distribution
The risk is real: enterprises could stick with one assistant per person, or platform giants could absorb the collaboration layer — Dust's answer is being the workspace where agents already run

可借鉴之处

Bet on the layer, not the model: Dust's multiplayer thesis — AI compounds when context is shared — made deployments stickier (240% NRR) and grew ARR roughly 20x without owning a foundation model

后续进展

As of September 2026 Dust is scaling: 3,000+ organisations, 300,000+ agents deployed, $20M+ ARR and 240% net revenue retention in 2025, backed by a $40M Series B from Abstract, Sequoia, Snowflake and Datadog. It is spending on three fronts: agents that improve automatically, collaboration primitives that let humans and agents share projects and context, and enterprise-grade governance. The US is its fastest-growing market, with customers like Datadog and 1Password, and profitability is deliberately not the near-term goal.

资料来源

发现哪里写错了？告诉我们。

轮到你了

你刚读完一家。说说你在做什么，看看谁在赌同一件事。

免费账号 · 3 次免费提问 · 不用绑卡