For the first time in roughly 30 years, NVIDIA isn’t releasing a new gaming graphics card. Not because anything went wrong in engineering. Because the memory those cards need is worth more to AI accelerator customers.
The AI infrastructure buildout has turned semiconductor wafer capacity into a zero-sum competition — and consumer electronics are losing. IDC projects PC shipments down 11.3% and smartphone shipments down 12.9% in 2026. LPDDR5x mobile memory prices are up 90% quarter-on-quarter. Counterpoint Research calls it “the steepest on record.”
This article maps the damage across four markets and identifies which brands and buyers are most exposed. For the background on how we got here, the AI memory shortage and its causes are covered in our pillar overview.
Why Did NVIDIA Delay Gaming GPUs for the First Time in 30 Years?
NVIDIA has shipped new gaming graphics cards every year for roughly three consecutive decades. As of early 2026, TrendForce confirmed both the planned RTX 50-series Kicker refresh and the RTX 60-series launch have been paused — making 2026 the first year in modern NVIDIA history with no new gaming GPU.
The economics are pretty straightforward. Gaming GPU revenue was approximately 35% of NVIDIA’s total revenue in 2022. By late 2025 that had collapsed to around 8%. NVIDIA’s AI compute segment runs at approximately 65% operating margins versus roughly 40% for gaming graphics. So SK Hynix, Samsung, and Micron — who together account for more than 90% of global memory production — have reallocated accordingly. They’re prioritising HBM for AI accelerators over the GDDR6 and GDDR7 that gaming cards require.
There’s a geopolitical angle here too. US export controls restrict Chinese researchers from buying NVIDIA’s advanced AI chips. One workaround that emerged was buying GeForce gaming GPUs for AI training and inference instead. Pausing production closes that channel — a second-order effect that goes well beyond consumer gaming.
How Is the Memory Shortage Hitting the PC Market at the Worst Possible Time?
The PC industry entered 2026 expecting a tailwind. Microsoft ending support for Windows 10 was supposed to drive one of the largest enterprise refresh cycles in years. Instead, peak memory pricing arrived at exactly the same moment as the Windows 10 EOL deadline. Terrible timing.
The cost mechanics are stark. HP‘s CFO disclosed that memory now accounts for 35% of a PC’s build cost — up from 15–18% the prior quarter. Dell‘s COO reported DRAM at $2.39 per gigabit, a 5.5x increase over six months. Lenovo, Dell, HP, Acer, and ASUS have all warned of 15–20% unit price hikes. Gartner projects PC prices rising 17% in 2026. IDC projects shipments falling 11.3% — “more negative than even our most pessimistic scenarios suggested just a few months ago.”
Gartner estimates higher memory prices are causing enterprise refresh cycles to lengthen by 15%. For organisations that can’t defer, the math is painful: you’re paying 15–20% more per device in a supply environment where vendors aren’t guaranteeing prices beyond two to three weeks. IDC expects actual PC shortages to start in Q2 2026 as smaller vendors struggle to secure components. If you want to understand the price mechanics underlying these market declines, the DRAM contract pricing trajectory explains both the scale and the duration.
What Is Happening to Smartphone Prices and Shipments in 2026?
The smartphone market is experiencing what Counterpoint Research calls “the sharpest decline on record.” Global shipments are projected to fall 12% in 2026 — down to volumes not seen since 2013. IDC projects a 12.9% decline.
The driver is LPDDR5x — the mobile memory variant powering mid-range and flagship smartphones. Prices have risen approximately 90% quarter-on-quarter. That’s the steepest in the technology’s history. The reason is direct competition from AI infrastructure: NVIDIA’s most powerful AI rack systems contain 54 terabytes of LPDDR5x each — the same variant that goes into your phone. As Counterpoint’s Tarun Pathak put it: “A lot of these memory companies are asking smartphone vendors to stand in line behind the hyperscalers.”
Smartphone average selling prices are expected to rise 3–8% in 2026 (IDC) or as much as 14% (Gartner). The problem is that the buyers most exposed — those shopping in the sub-$400 tier — are also the most price-sensitive. When a $250 Android phone needs to be priced at $290 to maintain margins, unit volumes collapse faster than the price increase percentage suggests. Qualcomm is already absorbing the downstream damage as OEM customers order fewer application processors.
Why Is the AI PC Rollout Stalling Despite All the Industry Momentum?
The AI PC story heading into 2026 was coherent. Intel integrated Neural Processing Units into its Core Ultra line. AMD followed with Ryzen AI. Microsoft defined Copilot+ PC — requiring a minimum of 16GB DRAM — as the hardware standard for on-device AI. Lenovo, Dell, HP, Acer, and ASUS all launched Copilot+ lines.
Then the memory cost shock arrived.
The Copilot+ specification requires 16GB DRAM minimum; 32GB is recommended for serious AI inference workloads. Under current DRAM pricing, those 32GB configurations have become “economically punishing” — IDC’s framing, not ours: “Just as the industry is seeing a need to add more RAM, it has become prohibitively expensive to do so, even if they can get supply. This will result in higher prices, lower margins, or a potential downmix in the amount of RAM in new systems at the worst possible time.”
Here’s the bind that creates. If OEMs ship Copilot+ at 16GB minimum rather than 32GB recommended, those devices can’t run the AI workloads that justify the Copilot+ premium. You pay more for a device that delivers less. Gartner confirmed enterprises will continue investing in AI PCs but at a slower rate, “likely to purchase devices with reduced memory.” The momentum isn’t gone — it’s been pushed back by at least 12–18 months. For actionable guidance on evaluating AI PC procurement against current memory pricing, the cost framework requires more than a simple timing call.
Which Brands Are Most Exposed — and Why Apple Is Different?
The memory shortage doesn’t affect all brands equally. The clearest illustration: Apple versus Android mid-range OEMs.
Apple: well-positioned. Apple secures memory supply through long-term agreements covering 12–24 months in advance. Its 2026 flagship iPhone models are expected to hold at 12GB RAM rather than increasing to 16GB, keeping BOM exposure manageable. Apple’s premium pricing means even a 3–5% cost pass-through doesn’t threaten demand.
Android mid-range OEMs: exposed. Honor, Vivo, Oppo, Xiaomi, TCL, Transsion, and Realme operate on thin margins with little supply hedging and large portfolios in the sub-$400 tier. Counterpoint Research forecasts these brands will be hit hardest.
Samsung: the dual-role position. As a vertically integrated memory manufacturer, Samsung produces the LPDDR5x it uses in its own phones — supply security that no pure-play OEM can match. SK Hynix and Samsung together surpassed the combined market cap of Alibaba and Tencent in early 2026. That tells you who’s winning from this shortage.
PC OEMs: scale matters. Lenovo’s scale gives it supplier leverage that smaller regional OEMs simply can’t match. Gartner has said directly that “consolidation isn’t off the map here — it’s survival of the fittest.” For broader context on the broader memory supply crisis driving these dynamics, the structural supply constraints explain why this asymmetry isn’t closing quickly.
What Does This Mean for Companies Making Hardware Decisions Right Now?
The memory shortage translates directly into decisions that need to be made in the next 90 days: PC refresh timing, AI PC adoption, smartphone fleet management, server hardware planning.
Here’s what you’re working with:
Memory prices won’t stabilise until 2027 at the earliest. The top four cloud providers are projected to spend $600 billion on AI infrastructure in 2026 — 70% more than 2025 — sustaining demand pressure well into next year.
Hardware costs are 15–20% higher than 12 months ago. Dell’s data ($2.39/Gbit DRAM, up 5.5x over six months) and HP’s BOM disclosure (memory at 35% of build cost, up from 15–18%) set the baseline. If your budget was built on 2025 hardware cost assumptions, it’s wrong.
The Windows 10 EOL decision has no good timing. Organisations that can’t run Extended Security Updates are refreshing at peak memory prices. That’s the constraint — not the strategy.
A practical framework to work through:
-
Assess your true Windows 10 EOL exposure. Can you purchase Extended Security Updates to buy 12–18 months while prices normalise?
-
Evaluate standard PC versus Copilot+ AI PC. For most buyers, the Copilot+ premium is hard to justify at current memory pricing when normalisation isn’t expected before 2027.
-
Tier your vendor selection by supply chain position. Prioritise Lenovo, HP, Dell — suppliers with established long-term agreements and scale. IDC expects shortages to hit smaller vendors in Q2 2026.
-
Build a 15–20% hardware cost buffer into your 2026–2027 technology budget.
For the full response strategy covering AI workload cost management during the shortage, the framework goes beyond hardware procurement. And for when pricing will actually normalise, the memory shortage end-date analysis covers what the production data suggests about the 2027–2028 horizon.
Frequently Asked Questions
Why is NVIDIA not releasing a new graphics card in 2026?
Gaming GPU revenue collapsed from approximately 35% of NVIDIA’s total revenue in 2022 to around 8% in 2025. The GDDR memory gaming cards require is more profitably allocated to HBM for AI accelerators, where operating margins run at approximately 65% versus 40% for gaming. TrendForce confirmed the pause covers both the RTX 50-series Kicker refresh and the next-generation RTX 60-series launch — making 2026 the first year in approximately 30 years with no new NVIDIA gaming GPU.
How much will PC prices increase because of the DRAM shortage in 2026?
HP disclosed that memory now accounts for 35% of a PC’s build cost, up from 15–18% the prior period. Dell’s COO cited DRAM at $2.39 per gigabit, up 5.5x over six months. Gartner projects PC average selling prices rising 17% in 2026; IDC projects ASP increases of 4–8%.
What is the AI memory shortage and how does it affect everyday devices?
Semiconductor fabrication lines that previously produced standard DRAM for laptops and phones have been retooled to produce HBM for AI data centre chips. Every wafer allocated to an HBM stack is a wafer not available for the LPDDR5x in a smartphone or the DDR5 in a laptop, creating simultaneous supply shortfalls across all consumer memory types.
Is now a good time to buy a new PC or should I wait?
Gartner advises that memory prices will not stabilise until 2027. If Windows 10 EOL compliance requires a refresh, deferral may not be possible. If it can be deferred, waiting 12–18 months for prices to normalise would reduce hardware costs meaningfully.
Why are smartphone prices rising in 2026?
LPDDR5x — the mobile memory variant — has seen prices rise approximately 90% quarter-on-quarter, the steepest in the technology’s history. Smartphone manufacturers either absorb the cost through margin compression or pass it to buyers through ASP increases of 3–8% (IDC) to 14% (Gartner). Memory companies are prioritising hyperscaler AI orders over smartphone vendor allocations.
What is HBM and why does it matter for the shortage?
High Bandwidth Memory (HBM) is DRAM chips stacked vertically and placed adjacent to the GPU die — delivering very high data transfer rates, essential for AI accelerators. SK Hynix, Samsung, and Micron, who collectively account for more than 90% of global memory production, are diverting fabrication capacity from conventional DRAM to HBM, creating the shortage across all consumer memory types.
Which smartphone brands are safest to buy from during the shortage?
Apple and Samsung are hedged via long-term supply agreements covering 12–24 months in advance. Apple’s 2026 flagship models are expected to hold at 12GB RAM rather than increasing to 16GB, limiting BOM exposure. Budget Android OEMs — TCL, Transsion, Realme, Honor, Oppo, Vivo, Xiaomi — face the worst margin pressure and are most likely to raise prices sharply, reduce memory configurations, or exit lower-end segments.
Will the AI memory crunch affect AI PC availability?
Yes. Microsoft Copilot+ AI PCs require a minimum of 16GB DRAM, and 32GB configurations — recommended for serious AI inference workloads — are significantly more expensive under current pricing. The shortage has raised the price floor for AI PCs at the exact moment the market was building adoption momentum.
How does the Windows 10 end-of-life situation make the memory shortage worse?
Microsoft’s end of Windows 10 support was expected to trigger a large enterprise PC refresh wave in 2026. Instead, memory costs are causing businesses to extend refresh cycles by approximately 15% (Gartner). Organisations that cannot delay face paying peak memory prices for hardware that costs 15–20% more per unit than 12 months ago.
Is the 2026 memory shortage worse than past chip shortages?
In terms of price velocity, yes. DRAM and NAND prices are projected to rise 130% year-on-year by end 2026. The 2026 shortage is distinctive because it is being deliberately sustained by economically rational wafer reallocation decisions — not COVID disruptions or factory fires — making it more persistent and predictable.
Can gaming cards be used for AI inference instead of data centre GPUs?
Yes, and this is already occurring in China, where US export controls restrict access to NVIDIA’s advanced AI chips. Chinese researchers have been using GeForce gaming GPUs for AI training and inference. NVIDIA’s pause therefore has a geopolitical dimension well beyond consumer gaming.
What does the memory shortage mean for server and cloud infrastructure costs?
Server DRAM has risen sharply alongside consumer memory variants. The top four cloud providers are projected to spend $600 billion on AI buildouts in 2026 — 70% more than 2025. For CTOs evaluating on-premises AI infrastructure versus cloud deployment, the full cost framework is covered in the companion article.
