What is an AI neolab? Meaning, examples and how they are funded
An AI neolab is a startup built around a research bet rather than a product. It is usually founded by researchers from large AI labs or universities, raises hundreds of millions or billions of dollars before it has meaningful revenue, and spends that money on compute and researchers to train its own models. Safe Superintelligence, Thinking Machines Lab, Reflection AI, Periodic Labs, AMI Labs and Humans& are the names most often placed in this group.
The term spread through venture capital essays and newsletters in 2025 and 2026, as these companies raised some of the largest early-stage rounds in the industry. It is an informal label, not a legal or accounting category, so lists of neolabs differ from one source to another.
Neolab meaning: the working definition
Radical Ventures, a Toronto-based venture firm that invests in AI, defines a neolab in a June 2026 analysis as "a startup focused on long-term technical breakthroughs", usually started by researchers and engineers who come from top AI companies or university labs. A stricter test is whether the company trains its own models in pursuit of a specific breakthrough, with any product following from the research rather than driving it.
In practice, a company tends to get the label when most of these apply:
- Founders from frontier labs. Senior researchers leave OpenAI, Google DeepMind, Meta, Anthropic or xAI to pursue an idea their former employer is not prioritizing.
- A research thesis first. The pitch is a technical direction, such as world models, reinforcement learning, continual learning or AI for scientific discovery, rather than a finished product.
- Very large early rounds. Seed rounds of several hundred million dollars and valuations in the billions before a commercial product exists.
- Spending on compute and talent. Most of the capital goes to GPU capacity and to researcher pay, which is why the rounds are so large.
- Its own models. Most neolabs train foundation models themselves instead of building on another company's API.
Neolab vs frontier lab vs AI application startup
A frontier lab is an established developer of the most capable models, such as OpenAI, Anthropic or Google DeepMind. These companies have large revenue, products used by millions of people and long-term compute contracts. A neolab wants to reach that level, or to beat it on one technical axis, but starts with a team and a thesis.
An AI application startup builds a product on top of existing models: a coding assistant, a legal research tool or a customer-service agent, for example. It is judged on customers and revenue from the start. A neolab is judged first on research progress, and its investors accept that revenue may come years later.
The boundaries move. Some neolabs already sell products, such as Thinking Machines' Tinker API, and a neolab that succeeds becomes a frontier lab in all but name.
Examples of AI neolabs (as of October 2026)
Safe Superintelligence (SSI)
SSI was founded in June 2024 by Ilya Sutskever, an OpenAI co-founder and its former chief scientist, together with Daniel Gross and Daniel Levy. The company says it has one goal and one product, a safe superintelligence, and that its business model shields the work from short-term commercial pressure. It raised $1 billion in September 2024 and, according to Crunchbase News, about $2 billion in April 2025 at a $32 billion valuation. Gross left for Meta in mid-2025, and Sutskever took over as chief executive.
Thinking Machines Lab
Mira Murati, OpenAI's former chief technology officer, started Thinking Machines Lab in February 2025. In July 2025 it raised $2 billion at a $12 billion valuation in a round led by Andreessen Horowitz, with Nvidia, AMD and Cisco among the investors. Its first product, Tinker, an API for fine-tuning open-weight models, arrived in October 2025, and in July 2026 it released Inkling, an open-weight model, according to its Wikipedia entry. In January 2026 two members of the founding team, Barret Zoph and Luke Metz, went back to OpenAI.
Reflection AI
Former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou founded Reflection AI in 2024. It presents itself as an open-model alternative to closed frontier labs. Reflection raised $2 billion at an $8 billion valuation in October 2025, with Nvidia, Sequoia Capital and Lightspeed among the backers, and was reported in June 2026 to be valued at about $25 billion (Wikipedia).
Periodic Labs
Periodic Labs was founded by Liam Fedus, formerly of OpenAI, and Ekin Dogus Cubuk, formerly of Google Brain. It came out of stealth in September 2025 with a $300 million seed round led by Andreessen Horowitz, TechCrunch reported. Its thesis is that AI systems connected to automated laboratories can speed up discovery in materials science.
AMI Labs
Advanced Machine Intelligence Labs was set up in December 2025 by Yann LeCun after he left his post as Meta's chief AI scientist. Headquartered in Paris, it works on world models, AI systems meant to understand the physical world and plan, and raised a $1.03 billion seed round in March 2026 from investors including Bezos Expeditions, Nvidia and Samsung (Wikipedia).
Humans&
Humans& was founded by Andi Peng, previously at Anthropic, Eric Zelikman and Yuchen He from xAI, early Google employee Georges Harik and Stanford professor Noah Goodman. It raised a $480 million seed round at a $4.48 billion valuation in January 2026, according to TechCrunch, with Nvidia and Jeff Bezos among the investors. The company says it is building AI meant to help people work together rather than replace them.
How neolabs are funded
The size of neolab rounds follows from their costs. Training competitive models requires large GPU clusters, usually contracted for several years, and experienced researchers from top labs are expensive to hire. Radical Ventures counts more than 40 neolabs that together raised about $40 billion over three years. For comparison, it notes that OpenAI and Anthropic together raised $5.3 billion in the seven years between OpenAI's founding and the launch of ChatGPT.
Two patterns stand out in the rounds above. First, chipmakers invest directly, and compute deals often come with the equity: Nvidia is among the backers of Thinking Machines, Reflection AI, AMI Labs and Humans&, and in March 2026 it announced a multi-year agreement to deploy one gigawatt of computing capacity for Thinking Machines. Second, the word "seed" has stretched: seed rounds of $300 million (Periodic Labs), $480 million (Humans&) and $1.03 billion (AMI Labs) are now part of this category.
Risks investors weigh
Radical's analysis lists several risks that apply to most neolabs:
- Capital intensity. A lab can need another multi-billion-dollar round before its research produces a product, and each round depends on investor appetite at that moment.
- Being overtaken. Frontier labs with bigger budgets may reach the same result first, or ship a good-enough version inside an existing product.
- Distillation and open weights. If rivals can copy capabilities by training on a model's outputs, or comparable open-weight models appear, a technical lead loses value quickly.
- Few buyers. Only a handful of companies can afford to acquire a lab valued in the billions, which narrows the exit options.
- Revenue. At some point the research has to turn into a business large enough to justify the valuation.
Talent is a further risk. Researchers who left one lab can leave again, as the Thinking Machines departures in January 2026 showed, and large companies have hired startup teams directly through license-and-hire deals, known as reverse acquihires, that leave the original company in place without its leaders.
Following neolab news
Neolab funding rounds, model releases and departures are part of what we track in AI Business and Models. For the wider picture of who is raising money and at what size, see Funding.
Frequently asked questions
What does neolab mean in AI?
A neolab is a research-first AI startup, usually founded by researchers from major labs, that raises large sums to train its own models before it has a significant product or revenue.
Is OpenAI a neolab?
No. OpenAI, Anthropic and Google DeepMind are usually called frontier labs. The neolab label is used for newer companies that are trying to catch up with them or overtake them in one area.
Which companies are considered AI neolabs?
Commonly cited examples include Safe Superintelligence, Thinking Machines Lab, Reflection AI, Periodic Labs, AMI Labs and Humans&. Lists differ because the term has no formal definition.
How do neolabs make money?
Many have little or no revenue yet. Some sell access to tools or models, such as the Tinker fine-tuning API from Thinking Machines, while others are still purely research operations funded by investors.
Why are neolab seed rounds so large?
Training models at frontier scale requires large GPU clusters and highly paid researchers, so these companies raise enough in their first rounds to pay for years of compute.