AI is making your hardware more expensive, even if you don't use AI

I bought 64 GB of DDR5-6000 a month ago for $300. The same kit today: $1,500.

That’s the part of the AI conversation nobody puts in a press release. The technology that’s costing $300 a desk to play with at the user end is costing the rest of us a few thousand dollars per workstation refresh, in a way that doesn’t show up in any single line item.

What’s actually moving

The headline is RAM. DDR5-6000 2 x 32 GB has gone from an 18-month average around $400 to over $1,000, with spot retail prices well above that. DDR4 is moving in the same direction. SSD pricing has started to follow over the past few weeks. Supply is tight enough that we’re seeing delivery dates slip on confirmed orders, which we hadn’t seen since the 2020 to 2022 chip shortage.

The cause is structural. Hyperscale datacentres being built out for AI training and inference are buying memory and storage in volumes the supply chain wasn’t sized for. Whatever’s left after the datacentre orders are filled is what goes into laptops, desktops, VR headsets, phones and on-prem servers. There isn’t much left.

What this means for your hardware budget

Three things to plan for in the next 12 months.

Workstation and laptop refreshes will land 20% to 30% higher than they did 18 months ago. Not a vendor problem, not a margin grab, just the cost of the components. If you’re budgeting against a quote from late 2024, refresh the quote before you sign anything off.

Server refreshes are worse than that. Server-grade memory is competing for the same factory output as the datacentre orders. If you’ve been planning a hardware refresh for an on-prem server, the AUD figure is going to look bad. We’re routinely advising clients to stretch existing kit with a maintenance contract for another 12 months and revisit when the supply chain catches up, where the box is still serviceable and the workload allows it.

Windows 10 just hit end-of-life and a lot of organisations didn’t make it. A meaningful slice of the Australian SMB market is still on Windows 10 with extended security updates, on hardware that doesn’t meet the Windows 11 requirements. That refresh has to happen, and a lot of it is going to land in 2026, on top of the AI buildout already pressuring the supply chain. If you haven’t started planning yours, now is the moment.

Will it normalise?

Eventually, yes. We saw the same pattern with GPUs during the crypto boom: prices ran up, then settled at a new floor that was higher than pre-boom but well below peak. The post-COVID semiconductor recovery did the same thing.

The honest answer: don’t bank on it normalising in the short term. Once vendors get used to selling at higher prices, the floor moves up. Combine that with tariff uncertainty in the US-China supply chain and the AI buildout continuing through 2026, and the realistic forecast is that current prices are closer to the new normal than to a temporary peak.

What we’re doing about it for our clients

A few practical things, where they apply:

• Running annual hardware planning earlier than usual so we can secure stock at fixed pricing rather than scrambling at end-of-life.
• Recommending stretched lifecycles where the existing kit is still healthy and the workload doesn’t demand new silicon.
• Talking clients out of non-essential upgrades. If the laptop still does the job, the laptop still does the job.
• Planning Windows 11 transitions in waves rather than panic-ordering everything in the same quarter as the rest of the country.

The point that doesn’t get made

Whether or not you’ve ever opened a chatbot in your life, you’re paying for the AI buildout.

Whether or not you have a ChatGPT subscription, whether or not you use Copilot, whether or not you’ve ever opened a chatbot in your life: you’re paying for the AI buildout. It’s in your laptop refresh quote. In your power bill. In water consumption around regional datacentres. In the cost of every component the datacentre supply chain is also bidding for.

The cost of training the next frontier model is being distributed across everyone who buys hardware, electricity, or anything else that depends on the same constrained inputs. That isn’t an argument against AI. It’s an argument for being clear-eyed about who pays for the buildout, and timing your hardware decisions accordingly.

If you’d like a hand reviewing your refresh schedule before the next budget cycle, get in touch.