Expanse

Startup name: Expanse 

Founder: Ismaeel Bashir 

School: School of Informatics 

Describe your entrepreneurial journey in three words: Listen. Build. Iterate. 

What is the idea?

The startup is called Expanse and it focuses on wasted GPU capacity. A GPU is a chip that is really good at doing millions of simple sums at the same time. This AI computing power is a fundamental resource that organisations are paying for but there is significant waste.  

Groups of AI computers at financial firms, research labs and government supercomputers are only doing useful work 30-40% of the time. The main reason is that engineers book 2-3 times more computing power than they actually need, just to be safe in case something goes wrong. 

Expanse fixes this by plugging into a customer’s computer cluster and recovering that wasted capacity in three ways: 

• Right-sizing – predicting how much computing power a job actually needs before it starts, so engineers stop over booking. 
• Improvement tips – suggesting tweaks to code and settings that researchers can apply themselves to run jobs more efficiently.
• Early failure detection – spotting jobs that are going to fail before they waste hours of expensive computing time. 

The mission is simple: get more AI work done from the computers that already exist, so the world doesn’t need to keep building new data centres just to keep up. 

Tell us more about the background and how you came to setup your startup

I studied Informatics (MInf) at the University of Edinburgh and did my 5^th year Minf research at Edinburgh’s Parallel Computing Centre (EPCC), the UK’s national supercomputing centre, under Adrian Jackson. There I built the first multimodal High-Performance Computing (HPC) resource predictor, a model that beat every published baseline by combining job source code, scheduler history and GPU telemetry. That research is what convinced me the problem was solvable in software.

Additionally, I previously worked in industry, at some of the largest banks and quant firms, running HPC and GPU training workloads. In these industries the same patterns play out at a larger scale: researchers over-provisioning by two to three times, jobs failing with no useful logs, and millions of pounds of compute sitting idle every year. The only fix on offer was to buy more hardware.

Expanse started with three of my good friends who lived the same problem at other companies. From our various experiences in HPC facilities and industry we realised this problem is present everywhere. We built this tool as what we wished we had running workloads on data centres. We were accepted into Y Combinator’s Spring 2026 batch and have been full time on it since March 2026.

What motivates you as an entrepreneur?

Well firstly, working as an engineer inside a bank/quant firm/AI lab inherently makes a lot of your impact scoped – the stuff you build mostly used internally and won't be seen on a global scale. With Expanse we can have our technology on every datacenter in the world – the impact is a lot larger. Secondly, I get to work on solving a really cool problem with three of my best friends – I couldn’t ask for a better way for my career to play out.

How do you define success? What success have you seen so far?

Success for us is measured in GPU hours recovered for our customers and the breadth of clusters we run on. Long term, success is every serious GPU cluster on the planet running Expanse.

So far:

• Y Combinator Spring 2026 batch (accepted March 2026).
• Second place at the Edinburgh Innovations Launch Point Competition for the Expanse idea, which was an early signal that the problem was worth chasing full time.
• Readying pilots with national supercomputing centres and large companies (under NDA, names withheld).
• Building the product itself – we beat every published baseline by 34% and can beat frontier LLMs at the same task more than 8x better

What piece of advice would you give to budding entrepreneurs?

Most people think you should build a really cool product and go out there and sell it. But in reality, the best startups I’ve seen, and the ones who’ll really have an impact, understand the problem space deeply.

This doesn’t mean you need to have experienced the problem first hand like we did. But you should at least have 100 hours of deep discovery calls with people in the space to understand what problems they have. Everyone loves to complain about work, or their day, or something. As an entrepreneur in the beginning, your goal is to let everyone vent to you. Once you have spoken to 100 people, you’ll find a cluster of the same problem reappearing, and that’s the one worth solving. Never lead these calls with what your idea is or how you’ll solve X. Have a normal, casual conversation with them on what sucks right now in their life/work and see if they end up talking about the problem you are solving naturally. That’s how you know your idea will stick.

For us, we sat through hundreds of hours of calls with infrastructure engineers and researchers at the biggest AI labs, quant firms and neo clouds, and heard the same problem statement over and over. When we asked them, “if you had a magic wand to solve one thing when running workloads, what would that be?”, we heard our product described back to us. If you find from these calls that a lot of people describe your product at a high level with the magic wand question, complain about the problem your company is solving, and are willing to pay for it, you’ve basically hit gold.

What role has Edinburgh Innovations played in your entrepreneurial journey?

Edinburgh Innovations (EI) were one of the very first people to back the idea. Placing second at the EI's Launch Point competition for Expanse gave us a nice early validation signal that the problem was worth chasing and that we made the right decision to chase it. I always say Edinburgh has a remarkable density of talent and Edinburgh Innovations is the connective tissue that makes that visible to founders.

What's next?

Y Combinator Demo Day is on 16th June 2026. After that we are scaling in San Francisco and London. The bigger ambition is to make Expanse the default intelligence layer for every serious GPU cluster in the world.