Equity for Equations: The Rise of Venture-Backed Physics

It's the first day of 2026, and I'm already two steps through the typical academic checklist: I've received a research council rejection letter and I've started to review nine proposals for fellowships and grants. Each proposal is more than 30 pages of science, consortia, track record, and justification, and with a ~10% chance of success, with funding decisions made on minutiae of technical detail and operational plan.

Meanwhile in California, Periodic Labs have raised $300M in their seed round without a product (or, possibly, employees) to develop a new AI-for-materials paradigm. This sits alongside cusp.ai ($130M), Isomorphic Labs ($600M) and Google Deepmind, who have made a recent move into materials. And that is before discussing pure AI, where Ilya Sutskever's Safe Superintelligence has raised $3B without a path to product...

A recent interview with Yann LeCun really drove this home. He described the environment he wanted to develop in his new post-Meta venture, and I realised: the conventional approach for research funding may be broken.

It seems that the deeptech world is giving birth to a new form of asset class: the neolab. This is neither a startup (they typically do not have a route to product revenue), it is not a university (there are no students, and they receive little if any public funding). It is a privatized Institute for Advanced Studies, powered entirely by venture capital operating in a new mode: placing bets on huge high-risk, high-reward binary science outcomes. In other words, the neolab will solve an existential challenge - artificial general intelligence, AI science, fusion - or it will fail.

This approach contrasts with almost everything I've been taught in an MBA! There is no (short-term) market, no unmet customer need, no growth curve, no P&L focus. Instead, venture capitalists are "putting it on red", effectively buying an option on raw scientific breakthrough. The valuation of Safe Superintelligence hit $32B in April 2025, not only pre-revenue and pre-product but pre-idea giving an indication of the heat in the "ideas" market. So, without a product, what is the exit for the VCs? Mostly likely acqui-hire for Google, Microsoft, or another trillion-dollar tech major.

Historically, universities have been the home for this fundamental research. Yann LeCun famously demanded to retain his academic role when working with Meta to maintain the "freedom to fail", and the backstories of almost all tech giants starts with a university spin-out (or drop-out, for Meta). On the other hand, the myriad reasons for the creep from university labs towards privately funded research have been well trodden; bureaucracy, glacial administration, salary erosion, and IP demands. This is not unique to AI companies either; pure-play quantum computing company PsiQuantum hit $7B valuation pre-product after raising more than $2.3B. Neolabs seem to aim for the sweetspot between freedom-to-fail and VC-level investment.

At the start of 2026, the neolabs are evolving into streams:

• Pure play: Safe Superintelligence, with a single KPI of "not killing humanity with AI". AMI Labs for world models. Fundamental research.
• Hybrid: Thinking Machines, cusp.ai - Research focused, but with a path to software tools that can deliver product
• Hardware: Isomorphic Labs, Periodic, Newcleo (Nuclear); combining software with hardware for "fundamental deeptech".

One of the surprises is that this trend is not simply rehashing the Xerox Parc or Bell Labs approach, of a fundamental division within a major company R&D who are expected to generate IP/patents as a KPI (with the exception of Deepmind). These organizations standalone, as a clean investment in a team and a scientific hypothesis. Furthermore, there appears to be geographic spread, with Paris, London, Cambridge represented alongside California.

The question remains whether the neolab concept represents a frothy idea conceived in the bubble-like valuations in AI, or a fundamental course correction in how we fund research. But before dismissing neolabs as a 21st century phenomenon, it is worth considering previous research funding pivots. In the 19th century, wealthy funders of science enabled Darwin, Joule, and Cavendish to carry out fundamental science by providing time and autonomy. While these early funders may have been primarily motivated by philanthropy (and associated prestige), perhaps neolabs can be best conceptualized as exchanging philanthropic motivation for equity and the dream of a 1000x return on investment.

This transition is clearest in the evolution of OpenAI from a non-profit to a for-profit - where the conflict between the freedom of 19th-century patronage and the discipline of 21st-century VC funding played out in real time.

As researchers and research leaders in the UK, ARIA provides the clearest model of the "big if true" funding concept that seeks to bridge the gap from fundamental to impact. This is an important and evolving route. But within the higher education sector, the more timely question is: do we actively engage, encourage and partner with these new entities, perhaps through innovative IP terms and hosting agreeements, or do we stubbornly "compete" with them, effectively becoming a training ground for their talent and emptying our labs?

Because in a world where a solving differential equation can be valued at $32 billion, the risk isn't that universities 'sell out' - it's that we get priced out.