Micron’s Memory Squeeze Is Raising PC Costs, but Waiting Until 2040 Makes No Sense

Micron says AI demand has turned memory into a strategic asset, while PC buyers face rising RAM, SSD, and graphics memory costs in 2026. That conflict has inspired an exaggerated question: should frustrated buyers simply wait until 2040?

The question surfaced in a PC buying debate on Zhihu and reached the platform’s hot list in early September. Its wording is playful, but the frustration behind it is real. Waiting for the next discount once worked because consumer hardware followed relatively familiar inventory cycles.

That strategy has become less reliable. Micron, Samsung, and SK hynix are directing more manufacturing capacity toward server memory and high-bandwidth memory, known as HBM. HBM packages stacked memory beside AI accelerators, giving those processors much faster access to data.

PC buyers now compete indirectly with cloud companies signing long-term agreements for enormous AI deployments. The central conflict is no longer impatient shoppers versus temporary retail markups. It is consumer affordability versus a supply chain that earns more by serving data centers.

The good news is that nobody needs to postpone a computer purchase for 14 years. The uncomfortable news is that the next broad price decline will depend on supply, demand, and product category. A calendar date alone cannot settle the decision.

What Changed in the Memory Market During 2026

Memory prices rose because AI infrastructure absorbed supply while manufacturers kept capacity disciplined, not because consumer demand suddenly became exceptional.

TrendForce sharply increased its memory forecasts early in 2026 after finding a worsening imbalance between supply and demand. During the first quarter, the firm forecast conventional DRAM contract prices to rise between 90% and 95% sequentially. It projected NAND Flash contract prices to increase between 55% and 60%.

Contract prices describe negotiated component transactions between suppliers and large buyers. They do not equal retail prices, but they influence the costs carried into finished computers, memory modules, phones, and SSDs.

The increases continued during the second quarter. By July, however, the direction remained upward while the pace had slowed. TrendForce’s third-quarter forecast projected conventional DRAM contract prices to increase another 13% to 18%. NAND Flash contract prices were expected to rise 10% to 15%.

This distinction matters. A slower increase is not a price decline. Buyers who interpreted moderating inflation as an approaching return to older retail levels were reading the signal too optimistically.

By September 2, the spot market was showing an even more complicated picture. Spot transactions cover relatively immediate purchases rather than longer supply contracts. TrendForce reported sluggish trading for branded DDR5 and DDR4 chips, yet firm supplier quotations kept prices elevated.

Its September spot update found that a mainstream DDR4 chip indicator increased 2.08% over one week. In contrast, the tracked 512Gb TLC NAND wafer price declined 0.90%.

Those movements show why “hardware prices are rising” is too broad a diagnosis. DRAM, NAND, and graphics memory share manufacturing relationships, but each market has different inventories and customers. Retail products also include processors, controllers, cooling systems, circuit boards, logistics, and retailer margins.

A consumer SSD can therefore become cheaper during a promotion while its underlying NAND market remains historically expensive. One graphics card can fall because inventory is weak, while another rises because supply is limited. The headline trend matters, but it never replaces product-level comparison.

The current event is the collision between elevated supplier pricing and weakening consumer tolerance. PC and smartphone manufacturers cannot keep accepting higher component costs without raising device prices, reducing specifications, or sacrificing margins.

That creates the article’s real tension. Weak consumer demand would usually push component prices down. In 2026, it has mainly slowed further increases because AI infrastructure continues supporting the supplier side.

Why AI Servers Get Memory Before Consumer PCs

The strongest customers now buy complete computing systems, and memory suppliers have greater incentives to serve them before price-sensitive PC shoppers.

An AI server requires more than an accelerator from Nvidia or another chip designer. It also needs HBM for the accelerator, DRAM for its CPUs, and enterprise SSDs for models, retrieval data, checkpoints, and application records.

Inference, the process of using a trained model to produce an answer, spreads this demand beyond model-training clusters. Every large deployment needs systems that can prepare data, coordinate accelerators, and serve many simultaneous requests.

TrendForce says general-purpose x86 servers using registered memory remain central to agent-based AI workloads. Registered DIMMs, or RDIMMs, add circuitry that helps servers manage large memory configurations reliably.

That means the AI boom pressures several memory categories at once. HBM receives much of the attention, yet conventional server DRAM and enterprise storage also consume manufacturing resources.

Micron has described the resulting environment in unusually direct strategic terms. Its latest quarterly outlook says memory has become a strategic asset for customers. The company also points to multi-year customer agreements designed to make its financial performance more predictable.

These agreements matter because cloud providers can commit demand far ahead of ordinary shoppers. A memory producer gains better visibility when a large customer reserves future output. Consumer module makers and smaller system builders have less bargaining power.

Memory factories cannot instantly switch between every product. HBM needs advanced packaging and specialized production steps. New DRAM or NAND capacity also takes years to approve, construct, equip, qualify, and ramp.

Manufacturers remember what happened during earlier memory downturns. Excess production caused severe price collapses, inventory write-downs, and investment cuts. Suppliers therefore have reasons to add capacity gradually, even when current margins encourage expansion.

This cautious response separates 2026 from a brief retail shortage. More supply is coming, but manufacturers are trying to avoid rebuilding the conditions for another destructive glut.

The allocation choice reaches consumer PCs through several paths. Less conventional DRAM capacity can raise module costs. Expensive NAND affects SSD bills of materials, meaning the list of components needed to build them. Higher GDDR costs can add pressure to graphics cards, although memory represents only one portion of a card’s cost.

Not every extra unit of AI memory directly removes one consumer module. Production lines, process generations, and packaging constraints differ. Still, companies make portfolio decisions across limited investment budgets, clean-room space, equipment, and engineering resources.

Those decisions favor products with durable demand and higher returns. In 2026, cloud infrastructure usually makes a stronger claim than an entry-level laptop or a discounted gaming upgrade.

For consumers, this changes the usefulness of waiting. A short delay can capture a sale or allow channel inventory to improve. Waiting indefinitely assumes suppliers will soon prioritize cheap consumer capacity again, which the available evidence does not support.

The 2040 Waiting Strategy Confuses Cycles With Structural Change

Waiting works when retail inventory is the problem, but it works poorly when manufacturers are reallocating capacity toward a different customer class.

PC enthusiasts have long called themselves “waiters” when launch prices look unreasonable. The tactic has a sensible foundation. New products often receive discounts, competition improves, and early demand eventually fades.

Memory markets are also famously cyclical. Producers invest during profitable periods, new capacity reaches the market, supply overtakes demand, and prices fall. Lower prices then discourage investment, preparing the next shortage.

The current cycle has not abolished that pattern. It has altered the forces inside it. AI infrastructure introduces buyers with larger budgets, longer planning horizons, and growing memory requirements.

A prediction tied to 2040 ignores those mechanisms. Fourteen years contain several product generations, recessions, factory expansions, export-policy changes, and new computing architectures. No responsible analysis can map a current memory shortage that far forward.

A more useful question asks how long the buyer can wait and which component is blocking the purchase. Someone replacing a failed workstation has a different decision from someone chasing an optional gaming upgrade.

The first buyer faces downtime, lost productivity, and security risks from aging hardware. A moderate component premium can be smaller than the cost of postponement. The second buyer can track promotions and delay without sacrificing necessary work.

Product age matters as well. A discounted last-generation graphics card can offer acceptable performance even when the newest model remains constrained. An SSD with slightly lower peak throughput can feel identical in ordinary office work.

The same logic applies to memory capacity. A buyer who needs more RAM for compilation, video work, or local AI inference should compare the benefit today against uncertain future savings. Waiting has an opportunity cost.

The strongest case for delay exists when a complete system still performs its required tasks. In that situation, buyers can watch several weeks of inventory and promotional data instead of committing to an arbitrary year.

There is also a danger in waiting for the exact bottom. Retail prices rarely announce that they have reached it. Attractive products can disappear as vendors replace older models, even while the broader component market improves.

An industry forecast discussed in a 2028 market scenario suggests DRAM and NAND conditions could improve late in that year. That scenario depends on supply catching demand and remains an analyst estimate, not a guaranteed buying date.

Even if component prices decline then, finished devices need not retrace every increase. Manufacturers can preserve margins, upgrade specifications, or introduce new products at different market positions.

The practical opponent in this story is therefore not one manufacturer against another. It is the promise of a familiar consumer cycle against the reality of structural capacity reallocation.

Waiting until 2040 would win only by avoiding many years of useful computing. That is not a purchasing strategy. It is an expensive form of market timing.

RAM, SSD, and GPU Prices Do Not Move Together

A shared memory shortage can influence all three categories without producing identical retail outcomes or identical reasons to wait.

DRAM modules provide working memory for a computer’s processor. Their retail prices respond to chip contracts, spot availability, module inventory, product generation, and competition among sellers.

In September, weak spot-market transactions suggested buyers were resisting supplier quotations. Elevated quotations nevertheless prevented a decisive retreat. That combination points to a standoff, not a settled recovery.

PC manufacturers also face inventory timing. Devices sold today can contain memory bought under earlier contracts. A contract increase might therefore reach stores after a delay, while an individual retailer discounts existing stock.

NAND Flash behaves differently. It stores data without power and forms the main capacity inside consumer and enterprise SSDs. Suppliers can sell related technology into phones, memory cards, removable drives, and data centers.

Consumer SSD buyers received one encouraging signal during the third quarter. PC manufacturers had accumulated substantial client SSD inventory during the first half of 2026. That reduced their willingness to accept further increases and forced more flexible negotiations.

Enterprise SSD demand remained firmer because AI systems need high-capacity, high-performance storage. Some enterprise products also require DRAM for caching and mapping, exposing them to both DRAM and NAND constraints.

The split explains why a consumer SSD sale does not prove that the wider storage shortage has ended. Inventory can temporarily outweigh upstream pressure in one channel.

Graphics cards add another layer. Their retail cost includes the GPU die, graphics memory, board components, cooling, software support, and channel conditions. Supply decisions by Nvidia, AMD, and board partners can dominate short-term pricing.

TrendForce reported that Nvidia’s RTX PRO 6000 Blackwell had not created the anticipated wave of GDDR7 demand by July. Weaker notebook shipments also reduced demand for GDDR6 and GDDR7.

Yet graphics memory prices still rose with the wider DRAM trend because suppliers could redirect capacity toward other products. Soft demand in one graphics segment did not automatically create plentiful, inexpensive GDDR supply.

This evidence supports a narrower claim than the social discussion implies. Memory conditions can add pressure to graphics-card costs, but they do not prove that every GPU increased for the same reason.

A consumer affordability analysis described the third-quarter slowdown as resistance from buyers rather than a major supply improvement. It also noted weaker demand for graphics memory and consumer storage.

That tension gives buyers leverage in selected products. Retailers must still move inventory, and device makers cannot ignore declining unit demand. However, the leverage is uneven.

A buyer should therefore separate the shopping list before deciding to wait:

• For RAM, track the exact generation, capacity, speed, and kit configuration. DDR4 and DDR5 markets can behave differently.

• For SSDs, compare capacity, endurance, controller quality, warranty coverage, and sustained performance. A temporary discount can outweigh modest future declines.

• For GPUs, judge total system performance and local inventory. Memory-market headlines alone cannot determine value.

This category-by-category approach lacks the simplicity of declaring all hardware overpriced. It produces better decisions because it follows the mechanisms that actually set retail prices.

Weak Consumer Demand Is the Biggest Challenge to the Shortage Story

The bullish supplier narrative faces a real limit: PC and phone buyers are already resisting higher costs, while some spot and storage indicators have softened.

Memory manufacturers have strong evidence for tight supply. AI deployments continue, server demand remains healthy, and production allocation favors products with better returns.

Still, tight supply does not guarantee unlimited pricing power. Customers can delay purchases, reduce memory capacity, reuse existing devices, or choose cheaper configurations.

TrendForce explicitly identified affordability as a constraint in July. It expected notebook prices to rise as higher component costs moved through inventory, which would weigh on shipments.

Smartphone brands faced a similar tradeoff. They could increase device prices to offset higher low-power DRAM costs, but doing so risked weakening sales. Some vendors responded with more conservative production plans.

By September, sluggish DRAM spot trading showed that resistance had not disappeared. NAND spot prices were largely range-bound, and the tracked TLC wafer measure had moved slightly lower during the week.

These data do not establish a broad reversal. Spot markets can be volatile, while contract markets follow separate schedules. They do show that the shortage narrative has boundaries.

Client SSD inventories provide another counterweight. PC manufacturers bought aggressively during the first half, leaving them better positioned to reject another large increase. Suppliers then softened their negotiating stance to preserve shipments.

Consumer substitution also matters. Buyers can choose a smaller SSD, reuse storage, purchase a previous-generation system, or upgrade only one component. These actions reduce the number of sales available at elevated prices.

Graphics markets face their own uncertainty. GDDR supply remains influenced by wider capacity decisions, but disappointing demand for a professional GPU family weakens claims of uniform scarcity.

The largest unknown is the durability of AI spending. Cloud providers currently have incentives to expand, and inference creates recurring infrastructure demand. Yet project delays, power constraints, better model efficiency, or weaker returns could change memory consumption.

Technical efficiency cuts both ways. Models that need less memory per request reduce demand for a fixed workload. Lower operating costs can also encourage many more requests, creating greater total demand through increased usage.

Supply forecasts carry uncertainty too. New factories can arrive late, yield fewer usable chips than expected, or ramp faster than planned. Governments can alter incentives, trade rules, and equipment access.

Manufacturers also control how aggressively they expand. Their public investments indicate more output, but disciplined additions can keep the market tighter than consumers expect.

This is why a precise date for normalization deserves skepticism. An estimate for late 2028 is more useful than a 2040 joke, yet it still depends on assumptions about demand and capacity.

The defensible conclusion is narrower. Price growth was slowing by the third quarter of 2026, but the slowdown reflected consumer resistance and high comparison levels. It did not demonstrate that abundant supply had returned.

Buyers should treat declarations of a permanent shortage and promises of an imminent crash with equal caution. Both simplify a market divided by product, customer, contract, and inventory position.

Three Signals Matter More Than Waiting Until 2040

The next buying window will become visible through contract prices, consumer inventory, and supplier capacity, not through a distant calendar target.

The first signal is fourth-quarter contract pricing for conventional DRAM and NAND Flash. A move from slower increases to flat or declining contracts would indicate that customer resistance is finally overpowering tight allocation.

One quarter would not establish a durable trend. Consecutive periods of softer contracts would provide stronger evidence that buyers can wait for lower retail costs.

The second signal is consumer inventory. Watch whether PC makers continue carrying elevated client SSD stocks and whether notebook shipments weaken further.

High inventory gives manufacturers and retailers reasons to discount products even when upstream memory costs remain uncomfortable. Falling inventory without stronger sales would remove that pressure and strengthen suppliers again.

For graphics cards, channel availability deserves separate attention. Sustained stock across several vendors and models matters more than isolated promotions or reports about GDDR contracts.

The third signal is capacity guidance from Micron, Samsung, and SK hynix. Buyers should focus on expected output, qualification schedules, and the balance between HBM, server DRAM, conventional DRAM, and NAND.

Factory announcements alone are insufficient. Semiconductor plants need time to ramp, and announced investment does not immediately become saleable consumer memory.

Long-term customer agreements are equally important. If suppliers keep signing large AI customers before new capacity arrives, additional output can be committed before it reaches the open market.

These three signals can strengthen or weaken the case for waiting. Falling contracts, stubborn consumer inventory, and faster conventional-memory output would support a delay.

Continued contract increases, clearing retail inventory, and AI-first allocation would favor buying necessary hardware during a targeted promotion. They would undermine the assumption that patience automatically produces a better deal.

A practical decision starts with a deadline. Buyers who need hardware within weeks should compare complete systems and exact components, then act when a suitable discount appears.

Buyers with functioning equipment can review the signals monthly. They should set a performance target and an acceptable configuration instead of trying to predict the bottom.

Businesses should also include downtime, staff productivity, warranty coverage, and deployment effort. Component cost is only one part of a workstation’s economic value.

Developers running local models must consider memory capacity carefully. Insufficient RAM or graphics memory can prevent a workload from running, while slower storage mainly affects loading and data movement.

Knowledge workers usually have more flexibility. Many office workloads gain little from premium graphics hardware, so a balanced system or selective upgrade can avoid the most constrained components.

The 2040 framing captures understandable frustration, but it encourages the wrong mental model. Today’s squeeze is neither a universal retail increase nor a permanent end to semiconductor cycles.

It is a reordering of customers. AI infrastructure receives priority, consumer buyers resist the resulting costs, and manufacturers add supply cautiously.

Do not wait for the market to issue an all-clear. Decide what performance you need, watch the three measurable signals, and compare the opportunity cost of delay. If your current computer still works, patience remains useful. If it blocks important work, waiting until 2040 would cost far more than a carefully chosen upgrade.