Sold Out Through 2027: What the Cloud Capacity Crunch Does to Your Clients' Bills

Microsoft said the quiet part out loud on an earnings call last year: it couldn't build data centres fast enough to meet demand, and it expected to stay capacity-constrained well into the following financial year. Amazon and Google have made similar noises. Analysts now talk about GPU and power capacity being effectively spoken for through 2026 and into 2027, with the largest AI customers signing multi-year deals that lock up supply before smaller customers ever see it.

For a Nottingham accountancy firm running a handful of virtual machines, or a Midlands manufacturer with its ERP system in Azure, that sounds like someone else's problem. It isn't. When the people who sell you compute can't make enough of it, the price goes up, the discounts dry up, and the assumptions you wrote into your 2025 budget quietly stop being true. Most SMEs won't notice until the renewal lands or the monthly bill creeps past the number they signed off. By then the easy savings have gone.

How we got here

For the best part of a decade, cloud was a buyer's market. AWS, Azure and Google Cloud competed hard on price, cut rates dozens of times, and threw credits at anyone willing to migrate. Capacity felt infinite. You could spin up a server in minutes and nobody worried about whether the data centre had room.

Two things changed that. First, AI. Training and running large models needs enormous amounts of specialised hardware and, crucially, enormous amounts of electricity. The hyperscalers have poured tens of billions into new facilities, but you can't conjure a data centre overnight. Planning permission, grid connections, chip supply and construction all take years. Capital expenditure is up sharply across all three providers, yet demand is still outrunning it.

Second, the big AI firms are willing to pay almost anything for guaranteed capacity. When a provider can sell a rack of GPUs to a model company on a three-year contract at a premium, it has no reason to hand that same capacity to a mid-sized business at a discount. Scarcity hands pricing power back to the seller. We haven't seen the cloud market like this since the early days.

What this actually does to a bill

The crunch doesn't usually show up as a dramatic headline price rise. It's subtler, and it arrives through several doors at once.

Discounts get stingier. Reserved Instances, Savings Plans and committed-use discounts are the main levers for cutting cloud spend. When capacity is tight, providers have less incentive to offer deep commitments, and the terms tilt in their favour. A client used to a certain discount at renewal may find the new offer worse, or tied to a longer commitment than they're comfortable with.

Specific regions and instance types run short. You might go to launch a particular VM size in the UK South region and simply not get it. Capacity errors, once rare, are becoming a real operational issue. The workaround is often a larger, more expensive instance, or a different region with higher latency and data-transfer costs. Either way the bill goes up.

Newer, pricier hardware becomes the default. As older generations of machines get retired or constrained, you're nudged onto current-generation kit that costs more. The performance might be better, but if your workload didn't need it, you're paying for headroom you won't use.

Currency and energy costs pile on top. For UK buyers, a lot of cloud pricing is dollar-linked, so sterling's movements matter. Energy prices feed directly into what it costs a provider to run a data centre, and those costs get passed on. None of this is the capacity crunch exactly, but it lands on the same bill at the same time.

Put together, a client who budgeted for flat or gently falling cloud costs in 2025 could be looking at a double-digit percentage increase by the time everything settles. That's a real line-item problem for a business running on tight margins.

Why the renewal conversation is too late

The instinct for most SMEs is to deal with cloud costs when the contract comes up. That made sense when prices only ever fell. It doesn't now.

If you wait for renewal, you're negotiating from a weak position in a market that already favours the seller. You've had no time to clean up waste, rightsize workloads or decide what genuinely needs to be in the cloud at all. You take whatever's on the table because switching takes months you don't have.

The better moment is now, before the next commitment, while there's room to make changes that actually move the number. This is where FinOps comes in, and where a decent MSP earns its keep.

What a FinOps conversation should cover

FinOps is just a tidy name for treating cloud spending as something you manage actively, with engineering and finance in the same room, rather than a bill you pay and forget. For an SME it doesn't need to be complicated. A few practical areas matter most.

Find the waste first. Most cloud estates carry 20 to 30 per cent of spend that delivers nothing: idle VMs, oversized databases, orphaned storage volumes, dev environments left running over the weekend. We regularly open a client's account and find servers that have been switched on and unused for months. Cleaning this up costs nothing and it's the fastest win available.

Rightsize before you commit. There's no sense buying a three-year reservation for a machine twice the size it needs to be. Measure real usage, shrink workloads to fit, and then decide what to commit to.

Get the commitments right. With discounts tightening, the mix of on-demand, reserved and savings-plan coverage matters more than it did. The right balance depends on how predictable each workload is. Over-commit and you're locked into capacity you don't need; under-commit and you pay the full rate. This is worth modelling properly.

Question what belongs in the cloud. Not everything does. For steady, predictable workloads, the sums for on-premises or hosted hardware can look a lot better than they did two years ago, precisely because cloud pricing has moved. We're not arguing for a wholesale retreat from cloud, but a honest comparison for the right workloads can save real money.

Set alerts and budgets. No business should learn about a cost spike from the monthly invoice. Budgets, anomaly alerts and tagging by project or department turn cloud spend into something you can see and control.

Where Cloudworks fits

The providers aren't going to volunteer that your discount could be better or that half your servers are idle. That's your MSP's job. We go through a client's cloud estate, strip out the waste, rightsize what's left, model the commitment options against the new pricing reality, and then keep watching so costs don't drift back up.

The capacity crunch isn't going to ease for a couple of years. The businesses that come through it in good shape are the ones that treat cloud spend as something to manage, not a utility bill that arrives and gets paid. If your 2025 cloud budget was written on the assumption that prices keep falling, it's worth a conversation now. Leaving it until renewal means negotiating from the back foot in a market that's stopped doing you favours.