Amazon’s base Echo Dot just went from $49.99 to $79.99, a 60 per cent rise on a device with barely any memory. If a gadget that simple is repricing, your PC fleet and phone rollout are next.
The culprit is a memory premium you can’t see on any shelf price. A PC builder was recently quoted around 500 per cent more for memory than a couple of months ago. This is the endpoint refresh squeeze: a downstream echo of the cost move that repriced servers.
So two questions follow. Refresh now, or extend the life of the kit you run? Is the premium temporary, or does it run to 2028? The supply-side story is the structural repricing behind rising prices.
Why are endpoint prices rising across PCs, phones and smart speakers?
Endpoint prices are rising because AI datacentres are reallocating DRAM wafer capacity toward high-margin HBM and server DDR5, tightening consumer supply. Micron estimates HBM consumes roughly three times the wafers of standard DRAM, so every HBM ramp squeezes general-purpose memory. Counterpoint makes the scale concrete: 16GB of smartphone DRAM went from about $42 to about $181 in a year, more than four times the earlier price. The wafer-capacity mechanics behind it determine how long this lasts.
Most buyers read the wrong signal. They watch street and spot prices, but endpoint quotes track the contract market, where OEMs buy memory on negotiated terms, and quotes are often valid for the same day only. Dell, HP, Lenovo, HPE and Cisco are passing the increase into endpoints, with desktops and laptops up 10 to 50 per cent in 2026 while memory modules rose over 400 per cent, and individual quotes have landed far higher.
The squeeze is not endpoint-only. OVHcloud‘s memory costs rose 6x in a year and are heading for 12x, so it repriced dedicated servers by up to 87 per cent. Hetzner lifted cloud server prices 30 to 37 per cent. It reaches phones too: IDC expects smartphone shipments to fall 16.7 per cent in 2026 as the average selling price climbs 27.6 per cent.
Australia is one case where the move is magnified. US-dollar supply chains, a weaker Australian dollar, freight, margins and 10 per cent GST compound the wholesale increase into shelf prices across Sydney, Melbourne and the other major cities. Magnitude is one question; the harder one is how long the premium holds.
Will the memory premium last until 2028, or is it temporary?
The answer is that nobody knows; forecasts span years rather than quarters. The direction is clear, the date of relief is not. OVHcloud says prices are not expected to return to historical levels before 2028, and SK Hynix expects demand to outpace supply through 2030 while pouring money into new fabs. Read against the broader repricing of software economics, the pattern is structural: capacity committed upstream does not reverse quickly.
The case for a shorter squeeze is softer. Gartner sees no meaningful pricing relief until late 2027, IDC sees the shortage through to early 2028, and Counterpoint won’t commit to relief before late 2028. A slower rate of increase still leaves prices rising; it is not a fall. Analysts expect relief from 2027 and normalisation in 2028 to 2029, but at a higher equilibrium than pre-2025 prices. Whether that reads as structural or cyclical is the debate that decides the duration.
That is why vintage exposure is the practical lens. Hardware bought in 2027 is committed at peak component pricing. Hardware bought in 2028 to 2029 hinges on whether supply normalises on the analysts’ timelines. Treat ‘until 2028’ as a planning scenario; the date is not committed. For budgets, the analyst guidance on normalisation applies the same evidence. Which leaves the question: what do you do with a fleet decision due this year?
Should you refresh hardware now or extend the life of existing kit?
The first signal to weigh is fleet age, alongside the cost of carrying ageing kit and whether refurbished or memory-upgraded hardware beats new-build pricing. Windows 10 support ended in October 2025, raising the question for fleets that can’t meet Windows 11 requirements. Beyond that, scoring devices on performance, battery, thermals and patch status shows which machines genuinely need replacing, regardless of age.
One of the cheapest ways to sidestep peak component pricing is often to not buy new at all. Refurbishment can add three to five years to a device’s life, and buy-back and ITAD pathways keep capability running at a fraction of a new-build quote. Certified refurbished enterprise hardware lands 50 to 80 per cent below new pricing, because its memory was bought before the crunch. Where refurbished won’t fly, a DDR4 upgrade path often beats paying today’s DDR5 premium.
Staging the refresh across your depreciation schedule spreads capex into steady waves and defers the worst of the spike without freezing capability. For Australian fleets, the ATO‘s effective-life and declining-value rules simply mean hardware over $300 is depreciated and anything under is deductible immediately. Lock same-day quotes on in-stock orders, and right-size memory configurations to the workload. Deferred investment carries its own cost, so the goal is to time and stage the spend. The software side is worth a read: how this reprices memory-hungry software.
Plan around the premium instead of waiting it out. Waiting for a street-price drop is miscalibrated. Forecasts span late 2027 through the 2030s, so vintage exposure and carrying cost decide the timing more than any single analyst date. The refresh-versus-extend call turns on fleet age, carrying cost and the refurbished-versus-new-build price gap, staged across your depreciation schedule. For the wider view of how this endpoint squeeze connects to cloud bills and software margins, start with the memory crisis and software economics.
Keep an eye on the analyst houses, Gartner, Counterpoint, Goldman Sachs, Morgan Stanley and IDC, plus TrendForce for contract pricing, and compare them against provider-level moves from OVHcloud and Hetzner. Provider pricing tends to shift before the headline forecasts catch up. When those two lines converge, the premium is easing and you can stop sweating the fleet.
Frequently Asked Questions
What exactly is the “memory premium”?
The memory premium is the extra cost baked into memory-dependent products when DRAM prices rise well above their historical baseline. It is not a single line item on a quote; it is the gap between what components used to cost and what they cost now, passed through by OEMs such as Dell, HP and Lenovo. In practice, it is why two otherwise identical machines can be priced hundreds of dollars apart within a single quarter.
What is HBM, and why does it push up consumer memory prices?
HBM, or high-bandwidth memory, is the stacked memory AI accelerators need, and it is far more wafer-hungry to produce. Micron estimates HBM consumes roughly three times the wafers of equivalent standard DRAM. Because fabs earn more from HBM, capacity shifts toward it, consumer DDR5 supply shrinks, and the cost flows straight into the PCs, phones and smart speakers you buy.
Why is a cheap device like the Echo Dot affected by memory prices?
Even low-memory devices are not immune, because they still contain DRAM that OEMs now buy at elevated contract prices. Amazon repriced the Echo Dot from $49.99 to $79.99, roughly a 60% jump, on a device with modest memory. The lesson is that the premium tracks component exposure, not how premium the product feels to the buyer.
Is this just “RAMageddon” and “chipflation” by another name?
Broadly, yes. RAMageddon and chipflation are the colloquial labels for the same DDR5 cost surge this article calls the memory premium. The informal terms spread faster on consumer channels, but they describe one phenomenon: wafer capacity shifting toward AI datacentre demand and repricing everything downstream. For procurement, the memory premium is the more useful term, because it maps to contract pricing you can actually track.
How much more will a new PC or laptop cost in Australia?
There is no single figure, but Australian buyers tend to feel the increase more sharply than US buyers. US-dollar supply chains, a weaker Australian dollar, freight and 10% GST compound the wholesale rise into local shelf prices in Sydney, Melbourne, Brisbane, Perth and Adelaide. A modest component increase can therefore become a noticeably larger retail increase once it lands here.
Can I avoid the premium by upgrading the RAM in my existing machines instead of buying new?
Often, yes, and it is one of the cheapest ways to sidestep peak component pricing. Adding DDR4 to older machines, or refurbishing existing units through buy-back and ITAD pathways, can extend capability by three to five years without a new-build purchase. Quantify the trade-off rather than defaulting to new hardware, because the price gap is wider than it has been in years.
Refurbished or memory-upgraded hardware versus new-build pricing?
Refurbished and memory-upgraded hardware usually wins on cost when new-build pricing is inflated by the component premium. Buying back, refurbishing and upgrading existing units can add three to five years of useful life at a fraction of a new purchase, while new builds lock in today’s peak memory costs. The exception is workloads where performance or warranty requirements rule out refurbished kit.
2027 hardware vintage versus 2028 to 2029 vintage: which is more exposed?
The 2027 vintage is the more exposed, because it is bought into peak component pricing with little chance of a near-term reprieve. Forecasts span late 2027 to 2030, so 2028 to 2029 vintages hinge on whether supply normalises on the analysts’ timelines. In practice, treat 2027 as a committed premium and the later vintages as the window where vintage choice genuinely starts to matter.
Where can I find analyst guidance on when memory prices are expected to normalise?
Track the analyst houses: Gartner, Counterpoint, Goldman Sachs, Morgan Stanley and IDC all publish normalisation views, and TrendForce is the reference for contract-market pricing data. The most useful habit is to contrast contract-market forecasts with provider-level price moves from OVHcloud and Hetzner, since provider pricing often shifts before the headline forecasts catch up.
Does the memory premium affect servers and cloud as well, or only endpoints?
It is not endpoint-only. OVHcloud and Hetzner show the same squeeze at provider level, with OVHcloud’s memory costs moving from 6x to 12x and its CCX13 plan rising from €15.99 to €42.99. Endpoint pricing and cloud pricing are downstream echoes of one contract-market cost move, which is why the squeeze surfaces in server, cloud and consumer line items at the same time.
Will rising memory costs push up my software and subscription bills too?
Possibly, because software vendors whose economics depend on memory-intensive infrastructure face the same cost pressure and often pass it on. This is why the crunch is repricing memory-hungry software and shifting build-versus-buy calculations. Review your subscription and licensing renewals alongside hardware, so a memory-driven increase does not arrive as an unexamined line item.
How do I know whether my own fleet is badly exposed to the endpoint refresh squeeze?
Score your exposure on three signals: fleet age, the cost of carrying ageing kit, and whether refurbished or memory-upgraded hardware beats new-build pricing. High failure rates, rising support costs and a Windows 10 end-of-support trigger all raise exposure. If your devices are old and expensive to maintain, the premium hits hardest at the exact point you are most likely to replace them.
