How to Choose an AWS HPC Migration Partner
Mike Thomas
Head of Cloud Practice, OnStak
Mike Thomas writes about the partnerships that make complex cloud migrations work.
Start with the distinction that trips up more migrations than any technical problem. Getting a workload onto AWS and running it well afterward are two different jobs, and treating them as one is where a lot of HPC on AWS migrations go wrong.
A migration specialist gets the workload onto AWS. A managed services partner lives with it afterward, through the quiet months and the bad nights. Some teams want one partner to carry both. Some would rather keep operations in-house and bring in help only for the move. Neither is wrong, but you want to settle which scope you are buying before you compare anyone, because the team that runs the cleanest implementation is not always the team you want answering a page at 2am.
What an AWS HPC migration partner should prove before the first workloadFor a workload that can't fail, ask for evidence, not a pitch. What comparable work have they done in your domain, and who will take your call and tell you how it really went? What would they want to prove on your own estate before you commit, and who on their side will put their name to the readiness call? It is worth asking where they stand in the AWS Partner Network and whether they hold the AWS HPC Competency Partner designation, not as a badge, but as a reference point you can check.
I am not asking you to grade their test method here. That is an engineering conversation, and it belongs with your technical team and ours. My colleague Ahmet covers what to measure and how. For what it's worth, OnStak holds AWS Migration Competency. Check anyone else's claim the same way. What you are checking at this stage is simpler, and more human: that someone credible, with a name and a track record, will stand behind the call that your workload is ready to move.
Any miler knows this part. A partner worth choosing has already done that quiet work long before you meet them.
Who coordinates AWS, your software vendors, and your own team?An HPC migration is never a two-party affair. There are at least four of you involved: you, AWS, your software vendors, and whoever is delivering the move. A good partner earns a lot of their value in the coordination, in making the pieces line up so you are not the one chasing three vendors and holding it together by hand. So ask who owns that coordination. If the answer is vague, it usually turns out to be you.
Who owns it after go-live?This is the question people put off until the morning something breaks, which is the worst morning to be working it out. Before cutover, while everyone is still calm, settle who runs the workload, who handles failed jobs and escalations, and where cost exceptions go. Name the owners. Write the handover down. Agree the escalation path up front.
Think of it as the handoff in a relay. The race is not decided only by how fast each runner is. It is decided in the exchange, in whether the baton passes cleanly from one hand to the next. Most migrations that go wrong do not go wrong because a workload could not run on AWS. They go wrong in the handoff, in the gap between the team that built it and the team left to run it.
Using AWS, or hiring a partner, does not move your accountability somewhere else. If you work in a regulated world, that responsibility stays with you. What a good partner gives you is not a way to pass it on. It is people who share the load, so the baton does not get dropped.
When it comes to choosing, I would put every partner on a single sheet, one row per requirement, with four columns. It is not elaborate, and that is the point. If a partner cannot fill that sheet in cleanly, in front of you, that is your answer.
The four columns: the evidence required, who provides it, who signs off, and who owns the workload after go-live.
Start with a read-only readiness assessment, the kind that hands you the map and the plan to keep whether or not you go further. For the engineering tests behind the readiness call, what to measure and how to read it, Ahmet covers that in his own piece. My job here is the part that decides whether it still works a year from now: who you trust to make the move, and who is still standing there, accountable, when it counts.
If you'd rather have this conversation in person, we'll be at AWS re:Invent 2026, November 30 to December 4 in Las Vegas.
Common questions about choosing a partnerHow do we choose an AWS migration partner for HPC workloads?▾
Do we need a migration specialist, a managed services partner, or both?▾
What should a partner prove before we move our first HPC workload?▾
Who coordinates AWS, our software vendor, and our internal engineering team?▾
Who owns failed jobs, escalation, and cost exceptions after go-live?▾