No Product, No Problem: AI 'Neolabs' Soak Up $24 Billion in Two Quarters

The strangest number in venture capital this year is $24 billion — and almost none of it is buying anything you can use yet.

According to a tally by Radical Ventures reported by the Financial Times on September 25, AI “neolabs” — research-first startups, many lacking products, markets, or revenue — raised $24 billion in the past two quarters alone. That is nearly five times what OpenAI and Anthropic collectively raised in the years before ChatGPT launched. The figure landed on Techmeme’s front page within hours and is already circulating as shorthand for how far the funding logic of this cycle has drifted from the last one.

What is a “neolab”?

The term describes a cohort of post-2022 founded labs built around a small team of elite researchers — usually alumni of OpenAI, Anthropic, Google DeepMind, or Meta — who raise enormous sums on pedigree and roadmap alone. Deedy Das, whose tracking of the trend helped popularize the label, noted earlier this year that nine of the ten most prominent neolabs carried valuations above $1 billion at the seed stage, often with well under $10 million in revenue and founders who had already made $10-100 million personally at their previous labs.

The archetype is Safe Superintelligence (SSI), Ilya Sutskever’s lab: roughly 50 employees, zero revenue, no shipped product, no published research — and a $32 billion valuation after Nvidia’s $5 billion investment in July. Thinking Machines Lab, founded by former OpenAI CTO Mira Murati, and a long tail of ex-frontier-lab founders round out the category. Trackers now count around 75 such labs.

The numbers that should give you pause

The $24 billion two-quarter figure needs context to land properly:

• ~5x the pre-ChatGPT benchmark. OpenAI and Anthropic together raised a fraction of that before ChatGPT turned AI from a research bet into a consumer product with 800 million weekly users. The comparison is not inflation-adjusted vanity math — it marks how much cheaper capital has made it to skip the “prove a product” step entirely.
• Seed rounds at unicorn scale. Where the pre-2022 generation fought for $10-50 million Series A rounds, neolabs routinely clear $1 billion valuations before writing a line of shipped code.
• Revenue is optional. The FT’s own framing — “no product, no problem” — captures the shift: investors are underwriting research talent and compute access, not traction.

The Financial Times cautions that these numbers “can hide some sleight of hand”: reported round sizes sometimes bundle compute deals, and headline valuations on paper shares can diverge wildly from what later-stage investors actually pay. Radical Ventures, which compiled the tally, is itself an investor in the space, so the figure doubles as marketing for the thesis it measures.

Why investors are doing this anyway

The bull case is brutally simple. If frontier AI consolidates into a handful of winners, the option value of owning a piece of even one additional frontier lab dwarfs the cost of funding a dozen failures. OpenAI’s pre-IPO trajectory and Anthropic’s climb to a $965 billion valuation on $65 billion raised taught investors exactly one lesson: the pain of missing the next one exceeds the pain of overpaying for all of them.

There is also a talent-scarcity argument. The number of people who have actually trained frontier-scale models is in the low hundreds. A team of ten with that experience plausibly holds more option value than a thousand-person company without it — and the neolab structure (tiny team, huge compute budget, no product pressure) is precisely what that talent demands in exchange for leaving labs where they are already wealthy.

The bear case is equally simple: roughly 75 labs are now competing for a prize that history suggests goes to two or three. Most neolabs will never ship a competitive frontier model — the compute bill alone runs to hundreds of millions of dollars per training run — and their valuations have nowhere to go but through the floor. When the FT notes the numbers hide sleight of hand, it is pointing at exactly this: a $1 billion seed valuation is a claim about a future that most of these labs will not reach.

The structural risk nobody prices

The deeper issue is what $24 billion per two quarters does to the ecosystem’s risk profile. This is capital raised against no revenue, backed in many cases by investors who marked up their own earlier AI positions to fund it — a circularity that works beautifully on the way up. The Brookings Institution warned this week that the broader AI buildout now carries $10.3 trillion in projected financing through 2032; the neolab boom is the venture-capital-sized tip of that iceberg.

There is also a subtle signal in who is paying. Nvidia’s $5 billion into SSI is a chip vendor financing its own customer — a pattern that inflates the AI ecosystem’s reported capital formation while concentrating risk in one supplier’s balance sheet. When the enabler of the boom becomes its largest limited partner, the boom’s independence from the boom’s fundamentals becomes an open question.

What to watch

Three indicators will tell us whether the neolab wave is a rational option portfolio or a bubble with a research agenda:

1. First revenue at SSI or Thinking Machines. The moment either ships a paid product, the category gets a real multiple. The longer both stay pre-revenue, the more the $24 billion looks like rotation, not conviction.
2. Down rounds. Watch for any neolab raising flat or down in 2027. One high-profile reset would reprice the entire cohort overnight.
3. The OpenAI IPO. If OpenAI’s public debut stumbles, the public-market appetite that quietly anchors private AI valuations disappears, and the neolabs — the furthest out on the risk curve — fall hardest.

For now, the money keeps arriving. The FT’s tally is a snapshot of a market that has decided the biggest risk in AI is not losing $24 billion — it is standing still while someone else spends it.