In July 2025, I suggested to Terry Tao how he might solve the Riemann Hypothesis. He pointed out, very gently, that it was a hard problem.

Let’s back-up a bit for some context. I represented Australia at the International Mathematical Olympiad in 1983. Terry first represented Australia three years later, at the age of ten, and went on to become the greatest living mathematician today. When the 2025 IMO came to the Sunshine Coast, about an hour from where I live, the organisers held a breakfast for Australian alumni. I arrived early. So did Terry. For about ten minutes it was just the two of us. Imagine a physicist getting ten minutes alone with Einstein. That was a lifelong ambition ticked off. This is a post about AI, maths and university commercialisation, but it starts with that breakfast.

Ten minutes with Terry

I knew Terry was interested in AI and in Lean, the proof assistant that lets a computer check every step of a mathematical argument. My suggestion was to build a digital twin of yourself and point it at the Riemann Hypothesis. I’d half thought that Navier-Stokes might be the more accessible target for him, but I wanted Riemann. Everyone who’s ever loved maths wants Riemann.

He replied, in his soft and very humble voice, that the Riemann Hypothesis was a hard problem.

He was right, of course. Riemann is still standing. But hold on to Navier-Stokes.

A parting gift

Late last year, after eight years of service, I stepped down from the board of UniQuest, the commercialisation company of the University of Queensland. UniQuest is the best university commercialisation shop in Australia, by a long way. It was underwritten by Gardasil, the cervical cancer vaccine invented at UQ, and more recently by Sanofi’s acquisition of Vicebio for up to US$1.6 billion, the largest deal ever involving a company built on Australian university IP. Vicebio runs on UQ’s Molecular Clamp, the platform behind UQ’s COVID vaccine candidate.

For a couple of years, I have been loudly advocating for AI at every opportunity, to the point where the phrase “what’s the AI angle, Richard?” was close to a running joke. When it came to my farewell dinner in December, the one thing I felt I’d left unfinished was setting out properly how I thought AI would change university research and its commercialisation. The reasoning models had just arrived, and I used one to help me organise my thoughts. My parting gift to UniQuest was a position paper with ideas for the future.

Everyone knew I was the AI guy, and since I was leaving, I was generously humoured. I suspect the Terry Tao paragraph read as the most far-fetched thing in it.

The University of Estonia

The core argument of the paper was simple. Every academic field carries a backlog of what I called “fill-in” research. These aren’t the big new ideas. They are the lemmas, the extensions, and the open problems that are within reach but not worth a career, so they sit on shelves and in the back of notebooks. I argued that AI would fill them in, all of them, and quickly.

And because someone would do it first, I invented a villain. Picture a researcher at the fictional University of Estonia (the flagship there is the University of Tartu, but I wanted a small, tech-forward country and the point was the country). This university is armed with AI, dropping hundreds of papers in a single week and front-running half the field. Uncomfortable for the researchers, and uncomfortable for universities. But I argued it should be the foundation of a golden age of university commercialisation, because every filled-in proof opens doors that weren’t open before. This would push researchers to spend their time on genuinely new ideas rather than the fill-ins.

I also argued that a commercialisation office using AI to assess, triage, and execute on this flood could plausibly double or triple its throughput of deals. As a slightly cheeky aside, I suggested that the Darling Downs was the natural place to put a gigawatt-scale data centre, and that having one an easy drive from Brisbane and UQ would be a commercial boon.

Then it happened

On 8 September this year, OpenAI announced that an unreleased model, running around 10,000 agents for 88 hours, had found a solution to the Navier-Stokes existence and smoothness problem, one of the seven Millennium Prize Problems. The Clay Institute hasn’t accepted it yet, but I know that Terry called it a remarkable achievement.

Navier-Stokes. The one I’d thought might be more accessible.

Four weeks later came the fill-in. On 6 October OpenAI posted 722 manuscripts covering progress on 372 open problems, from a model that had been pointed at roughly 4,000. Only 162 had their main results checked in Lean, so plenty of that will need scrutiny. But point a machine at everything and keep what lands; that is fill-in at industrial scale. As a mathematician I note the hit rate is about 9%. As an angel, I’d call that a portfolio.

The discomfort arrived on schedule too. One group letter described the release as a show of power rather than scholarship, and an NYU mathematician says he and a collaborator were close to the same Navier-Stokes result when OpenAI got there. It turns out my “University of Estonia” is actually in San Francisco.

And the Darling Downs? In August a Singapore developer proposed a $31.9 billion data centre on a cattle feedlot outside Dalby, now Australia’s largest proposed data centre. I’ll take that one.

What I got wrong

Quite a bit, and the misses are more interesting than the hits.

I pictured one great mind, multiplied. A Terry with a twin. What happened was a company with a data centre and a compute bill that NPR estimated at somewhere between US$6 and US$10 million for Navier-Stokes alone. The genius turned out to sit in the industrialisation.

I thought this would take years. It took about nine months from the dinner.

And I thought the universities would be the ones to do it. So far, it’s a lab. That matters, because whoever owns the fill-ins owns the first option on what’s built on them.

From proofs to products

The next question is what comes out of 722 manuscripts? Proofs aren’t products, but the obvious categories are already being listed:

• Formal verification of chips and code. Proving a microarchitecture correct before tape-out, or a flight-control system before it flies.
• Simulation. Better understanding of fluid dynamics flows straight into aerodynamics, weather and climate models, and plasma containment in fusion reactors.
• Cryptography. Stress-testing post-quantum algorithms, and cutting the cost of zero-knowledge proofs.
• Auto-formalisation. Turning human maths, software specifications or even contracts into something a machine can check.
• Clean training data. A proof that has passed Lean doesn’t hallucinate, which makes it unusually good material for training specialist models.

None of these is a company yet. Each is a category I’d expect to see in pitch decks within a couple of years, usually with a university name somewhere in the founding story.

When every field gets its maths moment

Here’s the part that matters to an angel. Maths went first because maths checks itself. AI-powered tools like Lean give you a yes or a no. Biology, chemistry and materials science don’t have a Lean yet, so their moment will be messier and slower. But eventually they will each have one, and each will produce its own pile of filled-in results.

Let’s think about what that does to the pipeline. For as long as I’ve been an angel, the scarce thing has been good ideas with a real problem behind them. If the backlog clears field by field, that flips. The scarce things become the people and processes that turn ideas into companies: tech transfer offices that can triage at AI speed, founders willing to leave the lab, and early capital willing to back them.

My bet (and not a forecast) is that university commercialisation will become one of the richest sources of deal flow angels have seen. More deals, of variable quality, arriving faster. That argues for breadth rather than conviction. As I’ve written in Twenty-five is a floor, not a ceiling, portfolio size is an infrastructure question, and a flood of fill-in spinouts makes the infrastructure matter more, not less. Plenty of these will be patent-rich and factory-poor, which suits angels perfectly well.

Living in the future

At the time the paper felt provocative. To me it felt obvious, and I’ve wondered why.

I think it’s the seat. An angel spends every working day looking at companies that only make sense if the future arrives. As I wrote in Living in an optimistic future, optimism is the rational position for an angel, whatever your temperament. From that seat, a machine filling in mathematics within a year didn’t look like science fiction. It looked like the next deal.

Terry was right about Riemann, though. It is a hard problem. I’m still hoping.

Richard Moore – MooCoo Ventgures

Richard Moore is co-founder of MooCoo Ventures, an angel syndicate that co-invests alongside Brisbane Angels, one of Australia’s most active angel groups. He has made over ninety personal angel investments since 2013.

moocoo.vc

DISCLAIMER: The information provided is for general informational and educational purposes only and should not be construed as financial, investment, or legal advice. Any decisions you make based on this information are your sole responsibility. You should consult with a qualified financial advisor or other professional before making any financial decisions. Past performance does not guarantee future results.