USAF Commits to Azure and Dell EMC: Lessons for Mid-Market Buyers

Public-sector procurement stories usually read like sponsored content dressed up as news. The logos are impressive, the dollar figures are impressive, and somewhere around paragraph four a nameless executive says something about "digital transformation journey." None of it is particularly useful if you're running a 400-user manufacturing company trying to decide what to do about your aging datacenter.

But there are lessons worth pulling out of big procurement announcements if you read past the press release. The multi-year commitments that agencies like the U.S. Air Force have made to Azure and to Dell EMC hardware tell you something real about how the big buyers think about risk, vendor lock-in, and the division of labor between hyperscaler cloud and on-prem infrastructure. Here is the translation for mid-market buyers, with opinions.

Why federal agencies still buy enormous amounts of on-prem hardware

The first thing to notice is that even the organizations writing the biggest cloud checks are also writing very big on-prem hardware checks. The USAF, DoD broadly, the intelligence community, and most of the civilian federal agencies have all made multi-billion-dollar cloud commitments in the last five years. They have also continued to modernize their on-prem infrastructure at roughly the same pace as before.

That is not a contradiction. It is the honest acknowledgment that "cloud-first" doesn't mean "cloud-only," and that the workloads that belong in a datacenter are still best served by a datacenter. The federal buyers have the scale to make that calculation rigorously. When they keep running workloads on Dell EMC (or Pure, NetApp, HPE, depending on the agency) gear, they are telling you something about the unit economics of those workloads.

For a mid-market buyer, the translation is simple: if the Air Force can't make the math work to put everything in Azure, neither can you. The "lift and shift to the cloud" narrative that peaked around 2018 has quietly been replaced by a more mature "right workload, right place" posture across every sophisticated buyer we talk to.

What the hyperscalers actually get used for

The second lesson from the big federal cloud deals is about what the workloads actually are. When you read through the public procurement documents, the hyperscaler work is overwhelmingly concentrated in four categories:

Identity and collaboration. Microsoft 365, Entra ID, Teams, SharePoint. These workloads are SaaS in everything but name, and nobody builds their own email infrastructure anymore if they can avoid it.

Bursty and elastic compute. Analytics pipelines, AI/ML training runs, modeling workloads, anything that needs thousands of cores for a few hours and then wants to give them back. This is where the cloud's elasticity earns its cost premium.

New development. Net-new applications built from scratch are almost always landed in the cloud because the developer tooling, the PaaS options, the managed databases, and the identity integration are all better there than anywhere else.

Disaster recovery targets. Cold-storage, compliance-aligned backup destinations and DR failover targets in a region distant from primary operations. This is one of the cleanest cloud wins in the entire playbook — ransomware-resistant, geo-redundant, and billed by the GB.

Notice what isn't on that list: the steady-state production workloads. The day-to-day file servers, the line-of-business databases, the application servers that run 24/7 at predictable load. Those tend to stay on the on-prem hardware line item, because that is where they are cheapest to run.

Why the Dell EMC (or equivalent) part of the story matters

The second half of most big federal procurement announcements is the on-prem refresh. There's a reason for that — converged and hyperconverged infrastructure platforms from Dell, HPE, Cisco, and the storage-focused vendors have gotten meaningfully better over the last decade. Automation, API coverage, lifecycle management, and integration with VMware, Nutanix, or now Proxmox have all improved to the point where running your own private cloud is a real, competent option again.

The federal buyers have not abandoned the idea of owning their infrastructure. They have upgraded their idea of what that means. Instead of racks of individual servers and SANs cabled to them, they buy integrated systems where the compute, storage, and network layers are provisioned together, managed through a single pane of glass, and upgraded as a unit.

For mid-market buyers, the same trend is accessible and usually a better deal than the hyperscaler. A modern HCI or converged stack from any of the major vendors, sized appropriately and hosted either on-prem or in a colocation facility, will run your steady-state workloads for a fraction of the cost per vCPU that Azure or AWS will charge you over a three-year horizon. The gap is wider now than it was five years ago, not narrower.

The procurement lesson: never negotiate a cloud deal in isolation

The third thing worth paying attention to in the big federal deals is the way they get structured. The USAF and other agencies don't buy cloud, they buy a package — the cloud spend plus the hardware plus the services plus the professional services plus the training. They negotiate all of it together and use the total commitment as leverage on every line.

Mid-market buyers leave money on the table by doing the opposite. Cloud spend gets negotiated with Microsoft's cloud team, hardware gets negotiated with a Dell rep, services get negotiated with the integrator, and none of the parties know what the others are getting. The result is that each vendor optimizes their own line and the customer pays list on everything that isn't the line they pushed hard on.

The fix is to bundle before you negotiate. Walk into the conversation with the total stack on the table and the message that the outcome of one negotiation affects whether the others happen. You don't need Air Force scale to get this leverage. You just need to put all the items on the same page.

What a mid-market buyer should take home

Boil the whole procurement announcement down to three operational truths.

First, the hyperscalers are where you put new development, bursty workloads, SaaS-equivalent workloads like email and identity, and DR targets. That is where the cloud economics work cleanly.

Second, the steady-state production workload — the thing that runs 24/7 with predictable load — is almost always cheaper to run on modern on-prem or colocated hardware. Don't let anyone tell you "cloud is cheaper" without asking them "cheaper than what, over what time horizon, and including which hidden costs."

Third, the procurement leverage is in the bundle. Negotiate the cloud deal and the hardware deal and the services deal as one conversation. The Air Force does. So should you, at your scale.

Three takeaways

1. Even the biggest cloud buyers keep buying on-prem. If the Air Force can't make cloud-only work for its full workload mix, treat anyone who tells you cloud-only is the right answer at your scale with appropriate skepticism.
2. The workload-placement pattern is stable. New dev, bursty, and SaaS-equivalent to the cloud. Steady-state to on-prem or colo. This is not going to flip back the other way.
3. Bundle the negotiation. Separate procurements for cloud, hardware, and services cost more than a single negotiation for the same total spend. Treat your infrastructure stack as a package.