Something unusual is happening to hardware prices, and most procurement teams found out when a quote came back wrong.
This is not general inflation. It is one component, memory, and one cause.
What actually happened
Memory chips go into everything. Servers, laptops, phones, printers, machine controllers. There are only a handful of companies in the world that make them, and building a new factory takes years.
Those companies have redirected their capacity toward AI data centres, because that is where the money is. Everyone else now competes for what is left.
TrendForce tracks the contract prices that large buyers actually pay. Their figures for the first quarter of 2026 show the scale of it.
That is one quarter. Not one year.
TrendForce gave two reasons. Memory makers are shifting their newest production lines toward server and AI memory. And large US cloud operators have been placing heavy orders since late 2025, taking a bigger share of supply and leaving other buyers to pay more.
By the third quarter of 2026 the increases had slowed but not stopped. TrendForce forecast conventional DRAM up another 13 to 18%, and NAND flash up 10 to 15%. It described the market as remaining "extremely tight".
The reason the rises slowed is worth noting, because it is not good news. TrendForce said buyers in consumer markets such as PCs and phones were "reaching their affordability limit". Prices moderated because people stopped being able to pay, not because supply recovered.
How long this lasts
J.P. Morgan's research team put a number on the whole move.
Their explanation is blunt. AI data centre construction and hyperscaler demand are absorbing a disproportionate share of global memory capacity. Jay Kwon, an equity analyst there, said the industry "will stay in shortage for multiple years".
So do not plan around this ending soon. Plan around it being the condition you buy in.
Why this is bigger than one component
It helps to see how large the AI build-out has become in trade terms. The World Trade Organization publishes a twice-yearly outlook, and its October 2026 update was unusual.
The WTO raised its forecast for world merchandise trade growth in 2026 from 1.9% to 3.9%. Almost all of the upgrade came from one thing.
Read that again. Nearly half of all the growth in world goods trade, across every product and every country, came from chips and servers.
The WTO also expects global AI infrastructure spending to rise by at least 30% in 2026, with a further 10 to 20% in 2027. The demand pulling parts away from you is still growing.
What this does to a normal buying process
Three things break, and they break quietly.
Quotes stop being quotes
In a stable market a quote is good for 30 days and you take it to an approval meeting. In an allocated market the price moves faster than your approval chain does.
If your purchase of a server needs three signatures over two weeks, you are not buying at the price you were quoted. You are buying at whatever the price is when the last signature arrives.
Budgets set last year are wrong
A refresh plan built on last year's unit prices will not buy what it was meant to buy. The gap is not 5%. On memory-heavy kit it can be most of the budget.
This needs to be a conversation with finance early, not a variance explained afterwards.
Lead times become the constraint, not price
When supply is allocated, the question stops being what it costs and becomes whether you can get it at all, and when. A project plan that assumes hardware arrives in four weeks may be built on nothing.
What to actually do
None of this is fixable by negotiation. You cannot negotiate your way past a global shortage. What you can do is change how you buy.
1. Shorten your approval chain for this category
This is the cheapest fix available and most companies never make it. If quotes expire quickly, an approval process designed for stable prices actively costs you money.
Agree a faster route with finance for this category specifically. A pre-approved budget envelope with a named decision maker beats a committee that meets fortnightly.
2. Separate what is urgent from what is habitual
A lot of hardware buying is a refresh cycle rather than a need. Machines get replaced at three years because that is the policy, not because they stopped working.
In a market like this, extending a refresh cycle by a year for office equipment is a legitimate decision. Make it deliberately, with the security and support implications written down, rather than by default.
3. Ask what is actually in the box
Specifications are often inherited rather than chosen. If a standard build carries more memory than the work requires, you are paying the shortage premium on capacity nobody uses.
This is the same discipline as checking whether people use the software licences you buy, applied to hardware.
4. Decide where cloud actually helps
Renting capacity moves the problem to someone who buys at a scale you never will. That is a genuine advantage right now, and it is not free.
Cloud providers face the same component costs and pass them on eventually. The honest framing is a trade between capital you cannot spend today and an operating cost that will drift upward. Do the comparison over three years, not one.
5. Put dates in writing
If delivery timing matters to a project, it belongs in the contract rather than in an email. What happens if the date slips. Whether the price is fixed at order or at delivery. Whether partial delivery is acceptable.
Most purchase orders are silent on all three. In a tight market, silence favours the supplier.
What not to do
- Do not wait for prices to fall
- The research does not support it. J.P. Morgan expects shortage conditions for multiple years. Waiting is a position, and right now it is an expensive one.
- Do not panic-buy either
- Buying three years of stock ties up cash, and hardware depreciates while it sits in a cupboard. The sensible middle is to buy what you can justify needing within a planning horizon you actually believe.
- Do not assume your incumbent is still competitive
- Allocation is uneven. A reseller who was cheapest last year may simply not have stock this year. Check more than one, which in this market is about availability as much as price.
- Do not let the shortage justify every increase
- A genuine component shortage is also excellent cover for a supplier raising prices on things it does not affect. Ask which component moved and by how much. A supplier who cannot answer is not passing through a cost.
The part that lasts after this ends
Shortages are not new, and this one will eventually pass. What is worth keeping is the discovery that most hardware buying runs on assumptions nobody has checked.
That the quote will hold. That the approval chain is fast enough. That the standard specification matches the need. That delivery dates are promises. In a calm market those assumptions are mostly harmless. In a tight one they are expensive.
The companies that handle this well will not be the ones who negotiated hardest. They will be the ones who could answer two questions quickly: what do we actually have, and when does it need replacing.
Common questions
Why have server and memory prices gone up so much in 2026?
AI data centre construction is absorbing global memory supply. Memory makers have shifted their newest production capacity toward server and AI memory, and large US cloud operators have placed heavy orders since late 2025, taking a bigger share of what is made. TrendForce forecast conventional DRAM contract prices rising 55–60% in the first quarter of 2026 alone, with server memory above 60% and NAND flash 33–38%.
How much have DRAM prices risen overall?
J.P. Morgan Global Research estimates DRAM prices will have risen more than 400% from the start of 2024 to the end of 2026. Its analysts attribute this to AI data centre construction and hyperscaler demand absorbing a disproportionate share of global memory capacity.
How long will the memory shortage last?
Longer than most budgets assume. J.P. Morgan equity analyst Jay Kwon said the industry will stay in shortage for multiple years. Building new capacity takes years, and demand is still growing — the WTO expects global AI infrastructure spending to rise at least 30% in 2026 and a further 10–20% in 2027. Plan around it as a condition rather than an event.
Did prices stop rising in late 2026?
They slowed rather than stopped, and the reason is not reassuring. TrendForce forecast conventional DRAM up another 13–18% and NAND up 10–15% in the third quarter of 2026, describing the market as still extremely tight. It attributed the slower pace to consumer buyers in PCs and phones reaching their affordability limit — prices moderated because demand was priced out, not because supply recovered.
How big is AI demand in global trade terms?
Large enough to move world trade figures. The WTO raised its 2026 merchandise trade growth forecast from 1.9% to 3.9% in its October 2026 update, and reported that AI-enabling goods such as semiconductors and servers accounted for 47% of all global merchandise trade growth in the first half of 2026, with trade in those products up 67% year on year.
Should we buy hardware now or wait for prices to fall?
The research does not support waiting. Shortage conditions are expected to persist for multiple years, so waiting is an active position with a cost. Panic-buying is also wrong, because stock ties up cash and depreciates in storage. The defensible middle is buying what you can justify needing within a planning horizon you genuinely believe, while shortening your approval chain so quotes do not expire mid-process.
What should change in our buying process during a shortage?
Five things. Shorten the approval chain for this category, because quotes now expire faster than committees meet. Separate genuinely urgent replacement from habitual refresh cycles. Check whether standard specifications carry more memory than the work needs. Compare cloud over three years rather than one. And put delivery dates, price-fixing and partial delivery terms in the contract — most purchase orders are silent on all three, and silence favours the supplier.
Does moving to cloud solve the hardware shortage?
It moves the problem to someone buying at a scale you never will, which is a real advantage right now. But it is not free. Cloud providers face the same component costs and pass them through over time, so the honest comparison is between capital you cannot spend today and an operating cost that will drift upward. Run the comparison over three years.
How do we tell a genuine shortage increase from opportunistic pricing?
Ask which component moved and by how much. A real pass-through can be traced to a specific part with a published contract price trend behind it. A supplier who cannot identify the component, or who applies the same increase to items without memory in them, is using the shortage as cover rather than passing on a cost.
Does this affect anything other than servers?
Yes. Memory goes into laptops, phones, printers, point-of-sale terminals, machine controllers and most networked equipment. Anything with a processor and storage is exposed to some degree. That is why a refresh plan built on last year's unit prices can fail across several categories at once rather than in one line.
Sources
- TrendForce, Memory Makers Prioritize Server Applications, Driving Across-the-Board Price Increases in 1Q26, Published 5 January 2026. Conventional DRAM +55–60% QoQ, server DRAM above +60%, NAND +33–38%; capacity shifted to server and HBM, heavy US cloud orders since late 2025.
- TrendForce, AI Server Demand Continues to Support Memory Prices in 3Q26, Published 3 July 2026. Conventional DRAM +13–18% QoQ, NAND +10–15%; market "extremely tight"; consumer buyers "reaching their affordability limit".
- J.P. Morgan Global Research, The AI-driven memory shortage, DRAM prices estimated to rise more than 400% from the start of 2024 to the end of 2026; analyst Jay Kwon on multi-year shortage.
- World Trade Organization, Global Trade Outlook and Statistics update, Published 8 October 2026. 2026 merchandise trade volume forecast raised to 3.9% from 1.9%; AI-enabling goods 47% of H1 2026 merchandise trade growth, those products +67% YoY; AI infrastructure spending projected +30% in 2026 and +10–20% in 2027.
Want this run for you?
We take on the transactional half of procurement — invoices, purchase orders, supplier data and indirect spend — inside your own systems and under your approval rules. Start with a free spend audit: we measure your volumes, cycle times and exception rates, and the report is yours whether or not you go further.
Book a free spend audit
