Many mid-sized employers spent years moving systems to public cloud and building teams to match. Now AI work is changing the maths for some of them. Costs for steady, heavy workloads climb, sensitive data raises compliance questions, and performance can vary when hardware is shared.
The result is a growing number of hybrid setups, with some workloads on public cloud and others on dedicated infrastructure, whether that is a private cloud, a colocation facility or your own data centre. Whether that is right for your company is a technical and financial call. The hiring consequence is not in doubt: the team you built for public cloud is probably not the team you need for this.
Why the team has to change
Public cloud rewards people who are good at configuring services someone else runs. Dedicated infrastructure puts the hardware, the capacity planning and the uptime back on you. For AI work, that adds a few specific demands.
- GPU compute that has to be sized, scheduled and kept busy
- Fast storage that keeps that compute fed with data
- Networking built for heavy traffic between nodes
- Orchestration so several teams can share the same hardware without fighting over it
Each layer needs someone who has done it before. A strong cloud engineer can learn a lot of this, but not in the quarter you need it.
The roles that tend to matter
You rarely need all of these at once. Which ones come first depends on how far along you are.
- Infrastructure and platform engineers with Kubernetes experience, ideally including GPU scheduling
- MLOps engineers who get models from a notebook into production and keep them running
- Data engineers who can move and prepare large datasets inside your own network
- Security and compliance engineers who can document where data lives and who can reach it
- Cloud cost analysts who can compare the true cost of public, private and hybrid over several years
That last role is often missed. Once spend is split across environments, someone in finance and someone in IT need a shared view of it. Controllers should be part of this conversation early, not handed a surprise budget line in month nine.
Decide the workload first, then the hire
A common mistake is to hire an infrastructure team before anyone has agreed what will run on it. Start with the workload.
| Situation | What it suggests |
|---|---|
| Steady, always-on workloads | Dedicated capacity may cost less over 3 to 5 years |
| Data that cannot leave your network | Private or on-premises, with compliance staff early |
| Early experiments, uncertain demand | Stay in public cloud, hire lightly |
If you are still experimenting, a small core team plus contract specialists is usually the sensible route. If the workload is proven and permanent, build permanent capacity.
Contract or permanent
Migrations and builds have a clear start and end. Standing up a cluster, moving workloads and writing the runbooks is project work, and contract engineers suit it well. Running the environment afterward is permanent work, and you want those people to know your systems for years.
A workable pattern:
- Bring in contract specialists to design and build.
- Hire 1 or 2 permanent engineers early so they learn the build alongside the contractors.
- Make documentation and handover part of every contract.
- Let the contract scope shrink as the permanent team takes over.
This avoids the situation where the only person who understands the cluster leaves with their contract.
Regulated employers have less room
Banks, insurers, credit unions, healthcare organizations and public sector bodies face the strictest questions about data location and audit trails. In Canada that includes provincial privacy rules and federal guidance. In the US it includes sector rules that vary by state and regulator. If you work in one of these areas, hire compliance-minded engineers at the start rather than adding a review at the end. Fixing a design after the fact costs far more than building it correctly.
Be realistic about the market
People who have actually run GPU clusters and production ML platforms are not plentiful, and they know it. Expect a longer search than for a general cloud role. Some ways to widen the field:
- Hire for strong Linux, networking and Kubernetes fundamentals, and train the GPU specifics
- Look at people from high-performance computing, research and data centre operations
- Write the job around the actual problem, not a list of 15 tools
- Move quickly once you find a fit
What to do next
Write down the workloads you expect to run over the next 3 years, which of them cannot sit in shared infrastructure, and what each environment will cost. Then list the skills your current team lacks against that plan. That gap list is your hiring brief, and it tells you which roles to fill permanently and which to fill by contract.
A Traverse recruiter can help you scope those roles, find contract or permanent talent in AI, data and operations, and test whether the search is realistic before you post it.