Ismaeel Bashir

Expanse 2026 global tour continues: Boston

I'm speaking at the AI Drug Discovery & Development Summit in Boston this Thursday at 12 on the HPC and AI Infrastructure stage (same talk as STAC Summit in New York).

Requested vs. Used, sizing workloads before they run. Everyone who submits a job to a cluster gets asked how much hardware they need and for how long, nobody actually knows, so everyone pads it. The talk is what we measured that costs (in one production deployment it was nearly $8M of idle compute in a single month) and what you can do about it before the job starts.

Chicago next for SC26 in November. If you're at AIDDD, come and find me on the expo floor, I'm there Tuesday to Thursday :)

#AIDDDSummit

I'm speaking at the STAC Research Summit in New York next Wednesday, 21 October, 11:30am

I'm going to speak about sizing workloads before they run to increase effective capacity. This will cover what this can cost if not done correctly on two production clusters, how we solve this with Compute Certainty, and why just asking a frontier model doesn't work.

Come say hi at the start up corner if you're around :)

Agenda: lnkd.in/e4EqTMXR

Every job on a HPC/GPU cluster starts with a guess.

On average (and from clusters we analyse), the industry reserves 2-3x what they use, on average.

Talking about this and how to solve it at London Systems Club, 12 Oct, Modal HQ.

luma.com/86rl7eel

We’ve raised a $5.3M seed for @ExpanseCompute, led by @crane_vc with PXN Ventures.

Expanse predicts exactly what an AI workload needs before it runs, so teams get as much as twice the work from the same hardware.

Why we started it: expanse.sh/blog/expanse-r…

Most cluster monitoring tools can’t tell you, within your allocations, whats being used properly. It tells you a node is allocated. It doesn’t tell you what’s happening inside the allocation. So a job that asked for 400GB and used 60GB reports as busy and you see ‘high utilisation’, but in reality there is 340GB sitting there doing nothing.

Across a cluster this reduces the number of jobs able to flow through your cluster. People then complain they dont have enough capacity, so you buy more, it takes 6-12 months and you still have the same problem.

We’ve measured this problem across various production clusters. Checking every job, what it asked for against what it went on to use. A lot of them reserved 2-3x what they used.

On one cluster the gap came to $8M of capacity nobody could see in a single month.

I’m talking about this and more at @stacresearch in New York in October, drop me a message if you’ll be attending :) - we will also be at the London event as well.