HPC-AI Advisory Council UK 2026, summary

Conference
HPC-AI Advisory Council
The talks of the HPC-AI Advisory Council 8th Annual UK Conference by subject, the highlights of the two days, and the people I met.
Author

Kolen Cheung

Published

October 15th, 2026

Modified

October 16th, 2026

NoteSeries: HPC-AI Advisory Council UK 2026
  1. Summary: a map of the talks, the highlights and the people
  2. The national compute: the new structure, its systems and its funding
  3. Day 1, 13 October
  4. Day 2, 14 October
  5. UK HPC systems

AI tools, Claude and ChatGPT, helped me digest the material: unpacking a talk’s slides, explaining a term I didn’t know, and researching the systems. I’ve checked what they said against the sources linked.

A map to the talks

The talks of both days by what they are about, rather than when they were given.

National strategy and funding

Open and sovereign AI

Running HPC and AI systems

Storage and data

Science on HPC and AI

People and careers

Highlights

These are the things I found most interesting and worth spending more time to digest.

Day 1

The first keynote, UK-LLM, and the endnote, SOFAIR, are both highlights of the day. I’ll discuss them in more detail in a future post.

The venture scientist is an interesting one. It highlights a problem in academia I know well: PhDs entering academia are not given enough opportunities to stay and eventually become professors (some call this the Ponzi scheme of academia, getting a ton of cheap labour, a.k.a. grad students, to work on research projects), and the rest, most of them, go into industry and spend the rest of their lives there. Her solution? Enable PhDs to become venture scientists, i.e. founders of companies, starting before they finish their PhD, so that they keep access to the (compute) resources and networks to flourish.

I asked her what if these companies are later acqui-hired, back into the big companies. Her answer was simply that it doesn’t impair her vision, as the value has already been created in the process. I tend to agree with her, and the most interesting bit of the answer is that it doesn’t seem to bother her one bit. I personally worry about the big ones getting bigger (start-ups are getting acquired much more than in the past, at least in tech). I suppose this is because it is not a problem for her to solve, so rather than worrying, she directs her attention to her vision, to where she can make a difference. (And this is probably one quality any founder needs. Another reason I’d never be one, besides that being a cosmologist probably wouldn’t lead me to ideas suitable for a venture scientist? Or is this the pessimistic me, the non-founder type, speaking?)

Day 2

Lies We HPC People Tell Ourselves and research data and survival bias are the highlights of day 2. I’ll discuss them in more detail in a future post.

Meeting other people

It was nice catching up with Rich (Richard Gilham), who works on Isambard-AI. I asked him about the agentic workloads he sees on the system, which his talk mentions. He said there are obvious signs that a lot of users are running agents, and it gets things done (in his talk, it makes them superusers). These activities sometimes cause problems, but what surprised me is that it is not worse than in the pre-agentic era, just different. I would have expected AI agents, especially in unskilled hands, to cause all kinds of problems on multi-tenant systems like HPC. After all, as Solomon Hykes put it, quoted by Simon Willison (Willison 2025) and by me in a journal club, “An AI agent is an LLM wrecking its environment in a loop.”

I met Gökmen Kilic, who gave the DiRAC Training Academy talk. He and I both worked on the PyAutoLens project before (my talk on it at Durham HPC Days 2025, and a later one at PyCon UK). We talked about training: his experience of building notebooks and serving them, so that they are easy to deploy and to use. (The Academy’s course pages link to notebooks hosted elsewhere, and offer them as a download too.) That has some similarities with the LFRic Atmosphere training of the Met Office Momentum partnership project I’m working on, with the extra challenge that our software environment is more complex, and we don’t have a hosted, common platform to give users as seamless an experience. I’ll see if there’s anything we can borrow from there.

We also talked a bit about careers. Historically, HPC facilities in the UK are funded one grant at a time, and long-term prospects aren’t guaranteed, so there’s a certain level of instability that comes with it, and it reaches the people who work there too, him included. This doesn’t seem to bother him: he works with the assumption that it will be funded, and keeps focusing on the good work. Hopefully UKRI’s change to a rolling programme would improve this situation.

Funnily enough, both of us ran into exactly the same problem, independently, when booking hotels. We googled the HPC-AI Advisory Council conference, where last year’s, the 7th, came out on top and this year’s, the 8th, second. We clicked into the 7th’s link, and booked the days listed there, 14–15 October, while this conference was a day earlier, 13–14 October. Both of us went through some turmoil to fix it. Frankly, the conference’s website should have a better UI/UX: the year isn’t in the title, the navigation is confusing, the two days of the programme are hidden behind a click, and on refresh it resets to the first day, so there is no permalink to a day. It is a typical web design problem: it looks great, with paper cuts all over. They should hire an RSE to improve it from an accessibility and UI/UX perspective.

I also met a sysadmin from EPCC. Although his job title is sysadmin, he said Research Infrastructure Engineer is a better description. He made one of the comments from the audience in Lies We HPC People Tell Ourselves: of his whole cohort, only he stayed in HPC, and the rest went to quant etc., not happy, but much better paid. We share a lot of similar laments. I mentioned that of the friends I had in the old days who loved mathematics and physics, I’m the only one who stayed in academia, and the others went to Meta, Google, quant, AI startups, etc.

References

Willison, Simon. 2025. “Designing Agentic Loops.” Blog. Simon Willison’s Weblog, September 30. https://simonwillison.net/2025/Sep/30/designing-agentic-loops/.