Reference · Market
The Neolabs List (2026)
"Neolab" is 2026’s word for a specific animal: a research-first AI startup founded by defectors from the big labs, funded like an incumbent before it ships anything. This is the sourced list — and, because this is a data publication, the honest state of each lab’s data buying.
The list
| Lab | Founded | Founders from | The bet | Public data program |
|---|---|---|---|---|
| Safe Superintelligence (SSI) | 2024 | OpenAI (Ilya Sutskever) | One goal only: a safe superintelligence; no product before it | None documented |
| Thinking Machines Lab | 2025 | OpenAI (Mira Murati, John Schulman) | Research plus early commercialization — model customization for business | None documented |
| Reflection AI | 2024 | Google DeepMind (Misha Laskin) | Open-weight models, agentic coding — the Western open challenger | None documented |
| Ineffable Intelligence | 2025 | Google DeepMind (David Silver) | Reinforcement-learning-first intelligence; largely stealth | None documented |
| AMI Labs | 2025 | Meta (Yann LeCun) | World models — the explicit bet against the LLM path | None documented |
| Discovery Loop | 2026 | Google (per FutureSearch’s tracker) | Automating the research process itself; newest and least documented | None documented |
| World Labs | 2024 | Google / Stanford (Fei-Fei Li) | Spatial intelligence — models that reason about 3D worlds | None documented |
| Periodic Labs | 2025 | OpenAI / Google alumni | Autonomous laboratories for materials science | None documented |
A note on what is not in this table: valuations and raise sizes per lab. The trackers disagree by billions on the same companies, and a list that prints the numbers anyway is laundering guesses. The aggregate is reported (29 labs, $23 billion combined, fewer than 35% with a commercial product — Dealroom via October 2026 reporting); per-lab figures will join this page when a lab’s own disclosure or a regulator-grade source supports them.
Why a data publication keeps this list
Neolabs are the demand horizon of the data market. A lab with no product and a multi-year research bet consumes a different data diet than an incumbent — evaluations, proprietary expert data, RL environments — and consumes it quietly. When that buying becomes visible, it will become visible the way frontier-lab buying did in 2023–2025: through the AI data labs, expert networks and managed vendors in this directory, not through press releases from the labs themselves. Until then, "none documented" is the accurate entry, and it is the entry every row above honestly carries.
Frequently asked questions
What is a neolab?
A research-first AI startup, usually founded by senior researchers who left an established lab (OpenAI, Google DeepMind, Meta), that raises very large sums to pursue a long-term breakthrough before it has a product. Business Insider’s October 2026 description — "the lavishly funded, research-first AI startups founded by top talent who defected from the likes of OpenAI or Google DeepMind" — is the definition the market settled on.
How many neolabs are there?
Depends who is counting, and the disagreement is the story: Dealroom counted 29 neolabs that had raised a combined $23 billion as of October 2026, with fewer than 35% having released a commercial product; enthusiast trackers list over 100 names under looser definitions. This page lists the labs that appear across the serious counts rather than padding to the longest list.
Do neolabs buy training data?
They must consume data like every lab, but as of November 2026 none of the labs on this list documents a public data program — no published vendor roster, no open procurement, no data partnerships announced. Their data buying, where it happens, runs through the same private vendor market as everyone else’s: the data labs, expert networks and brokers in our directory. Anyone claiming a public "neolab data supplier" list is inventing it.
Sources
- The Neolabs Are a Bet Against Superintelligence (the six-lab forecast table) — FutureSearch (2026)
- What Is an Applied AI Lab? (neolab definition; Business Insider & Dealroom figures, Oct 2026) — Ertas AI
Facts on this page were drawn from the public sources above, checked November 2026. If you spot an error, tell us at the corrections address in the footer — we fix sourced errors and note the change.
Last updated: November 8, 2026