SSD Prices Surge 220% as AI Demand Squeezes DIY Buyers

Tom Hardware has measured a 221.5% annual increase for one mainstream SSD, exposing how severely AI infrastructure demand is squeezing consumer storage.

The comparison covers the Samsung 990 EVO Plus 2TB. A major sale made it an attractive component in July 2025. Twelve months later, its current discounted price was more than triple that earlier deal.

This is not simply one retailer producing a strange listing. Other consumer drives have risen sharply, while NAND suppliers increasingly favor enterprise customers. The pressure has spread from premium NVMe products to older SATA models and ordinary capacity upgrades.

The conflict now shaping storage is straightforward. AI data centers can justify high component costs because storage supports revenue-generating services. DIY builders, home-server owners, and upgraders usually cannot.

That imbalance gives enterprise buyers first claim on constrained NAND flash, the nonvolatile memory inside SSDs. Consumer buyers receive tighter allocations, weaker discounts, and fewer genuinely affordable choices.

The Tom Hardware analysis captures the retail result. Industry data explains why the problem extends far beyond one product or promotion.

Tom Hardware Finds the Cheap SSD Era Has Reversed

The defining change is not that SSD prices increased, but that a routine PC component became a major budget constraint within one year.

Tom Hardware compared current listings with deals recorded around the 2025 summer sales period. Its clearest example was the Samsung 990 EVO Plus 2TB, which increased by 221.5%.

That drive matters because it represents a practical mainstream choice, not an obscure professional product. It offered a familiar brand, useful capacity, and performance suitable for gaming or general desktop work.

A 2TB drive once fit comfortably into a balanced gaming build. The same component now competes with processors, graphics cards, and memory for a much larger portion of the total budget.

Higher-capacity products show the same direction. Tom Hardware found that a 4TB WD Blue SN5000 had roughly doubled over the comparable period. Its cost per gigabyte also more than doubled.

SATA has not provided a dependable escape route. The publication tracked a Samsung 870 EVO 2TB listing that was about 160% above its approximate level one year earlier.

That product uses an older interface and cannot match current NVMe performance. Its increase shows that buyers are paying for scarce flash capacity, not only faster controllers or newer specifications.

Retail listings still require caution. An inflated list price can make an ordinary selling price look like an enormous discount. Temporary stock gaps can also produce offers that do not reflect sustainable market value.

Tom Hardware explicitly noted this distortion. Some displayed list prices appear disconnected from normal transaction levels, making comparisons between discounts especially unreliable.

The year-over-year examples remain useful because they compare identifiable products across a consistent period. Together, they show that affordable storage has disappeared across several performance categories.

This reversal affects more than enthusiasts chasing benchmark records. A developer may need space for virtual machines, containers, datasets, and local models. A creator may need working storage for high-resolution footage.

Home-server users face the same problem at larger capacities. Local media libraries, backups, security recordings, and network-attached storage can consume several terabytes before redundancy is considered.

Those buyers can postpone purchases, but postponement has a cost. Existing drives continue accumulating writes, projects remain split across devices, and backup plans become harder to maintain.

The earlier consumer SSD market rewarded patience. New NAND generations and excess supply regularly delivered more capacity at lower costs. Seasonal promotions reinforced the expectation that waiting would produce a better deal.

That relationship has reversed. Waiting through the last twelve months exposed buyers to tighter supply and higher replacement costs.

The result is a different DIY calculation. Storage is no longer the easy component that builders select after settling the processor and graphics card. It can determine whether the project proceeds at all.

AI Storage Demand Is Pulling NAND Away From Consumers

AI demand matters because it changes which customers receive scarce NAND, not merely how many total SSDs the industry ships.

An SSD combines NAND flash with a controller, firmware, and supporting components. The NAND stores data without power, while the controller manages placement, error correction, and communication with the host.

Consumer and enterprise SSDs serve different workloads, but they draw from the same broad manufacturing base. Suppliers therefore decide how wafers, production lines, and finished components are allocated across markets.

Enterprise customers currently present the stronger economic case. Cloud providers require high-capacity storage for training data, checkpoints, retrieval systems, inference caches, and continuously generated AI content.

These buyers purchase at scales unavailable to an individual builder. They also negotiate long-term agreements, value predictable supply, and can absorb costs inside larger infrastructure budgets.

TrendForce said North American cloud providers accelerated enterprise SSD procurement from late 2025. It expected first-quarter 2026 enterprise SSD contract prices to increase between 53% and 58%.

The same NAND price outlook projected overall NAND flash contract prices to rise between 55% and 60% during that quarter.

Supply cannot respond instantly. New fabrication facilities require long planning, construction, equipment installation, qualification, and customer validation cycles.

Manufacturers can increase bit output through process migrations, meaning each generation stores more data within a similar physical area. However, migrations cannot immediately close a large demand gap.

Capacity decisions elsewhere also influence NAND. TrendForce reported that memory manufacturers were prioritizing attractive DRAM opportunities, limiting how much existing factory space could support additional NAND output.

AI inference adds persistence to the demand. Training a model can look like a concentrated project, but operating agentic systems produces ongoing storage and memory requirements.

Longer contexts create larger working sets. Repeated agent actions generate logs, intermediate files, embeddings, checkpoints, and cached data. Multimedia generation increases storage needs further.

TrendForce’s May forecast described this as a structural expansion rather than a short purchasing burst. It projected that the nine largest cloud providers would increase combined capital expenditures by 79% in 2026.

Its AI memory forecast also raised the expected 2026 NAND flash market to 270.6 billion, representing 280.7% annual growth.

Those market projections remain forecasts, not guaranteed outcomes. Cloud capital spending can change, and AI companies continue searching for more efficient storage architectures.

However, the allocation signal is already visible. Micron decided to exit its Crucial consumer business while continuing to serve enterprise customers.

Micron said AI-driven data-center growth had increased demand for memory and storage. Its December 2025 announcement stated that reallocating support toward larger strategic customers motivated the Crucial decision.

The company continued consumer shipments through the end of its fiscal second quarter in February 2026. It also committed to ongoing warranty service for existing Crucial products.

The Crucial exit removed a familiar SSD and memory brand from retail. More importantly, it provided a direct explanation for the industry’s priorities.

Consumers did not suddenly need fewer drives. They became less attractive customers than data-center operators willing to secure large volumes at higher margins.

That is the mechanism behind the retail surge. Enterprise demand pulls supply upward, suppliers favor strategic accounts, and consumer channels compete for the remaining output.

The DIY Market Is Losing Both Supply and Bargaining Power

DIY buyers face a two-sided squeeze because scarce NAND is moving toward data centers while finished consumer drives are moving toward PC manufacturers.

Silicon Motion designs controllers used in many SSDs. Its position gives the company visibility into where module makers send their finished products.

At Computex 2026, company executive Nelson Duann said the retail SSD market had “almost disappeared” during the first half of the year. His explanation revealed an important second layer of competition.

PC manufacturers could not obtain enough NAND or finished SSDs directly from memory suppliers. They increasingly turned to module makers that had traditionally supplied retail products.

Those module makers still purchased controllers and built drives. However, more output flowed into systems from Acer, Asus, Dell, HP, and other manufacturers instead of reaching store shelves.

The retail SSD shift means strong controller shipments do not necessarily indicate a healthy aftermarket. Products can move through the supply chain while bypassing DIY buyers.

Prebuilt computers may therefore offer more storage value than separately purchased components. Manufacturers can secure drives in volume and spread storage costs across an entire system.

That option carries tradeoffs. A retail SSD normally exposes the exact model, controller generation, NAND type, endurance rating, and reviewed performance.

A prebuilt specification can identify only capacity and interface. Two systems advertising the same storage size may contain drives with different sustained write behavior, thermal limits, or endurance.

The distinction matters when manufacturers substitute components during shortages. A system may retain the same public model name while its internal storage changes between production batches.

Buyers should focus on workload requirements before accepting that compromise. Most games do not need the highest sequential transfer rates available from premium PCIe 5.0 drives.

Sequential speed measures performance when reading or writing large contiguous files. Everyday responsiveness often depends more on latency and random access patterns.

PCIe 4.0 remains adequate for many gaming, development, and productivity workloads. Even older interfaces can serve secondary storage when the alternative is abandoning needed capacity.

However, lower specifications no longer guarantee lower costs. The Tom Hardware examples show severe increases across both NVMe and SATA products.

Used enterprise drives present another possible route, especially for servers. These products can offer strong endurance, but they may require adapters, additional cooling, or interfaces absent from ordinary desktops.

Used storage also demands careful health verification. Buyers should inspect total writes, error counts, power-on hours, firmware history, and available warranty coverage.

For a primary boot device, uncertainty around prior use can outweigh the savings. For replicated server storage, the risk may be manageable if the system has backups and replacement capacity.

Capacity reduction is the most direct response. A builder can select a smaller boot drive, reuse existing storage, and add capacity when conditions improve.

That approach works only when data has somewhere safe to go. Moving files to a single aging drive creates a hidden reliability problem rather than a genuine solution.

Cloud storage can absorb some pressure, but it changes the cost and control model. Upload speeds, recurring commercial terms, privacy requirements, and account access become part of the storage plan.

Local AI users face a particularly sharp contradiction. Running models locally offers greater control over data and availability, yet model files and related datasets require substantial storage.

The AI investment cycle is therefore making some local AI projects harder to build. Infrastructure demand from centralized services raises the component costs paid by people pursuing local alternatives.

Developers can limit duplication by documenting environments, datasets, and model versions before downloading them repeatedly. A searchable engineering knowledge base can help teams identify which local assets remain necessary.

Documentation cannot replace storage, but it can reduce waste. Forgotten disk images, duplicate model checkpoints, and abandoned build artifacts become expensive when every terabyte carries a larger opportunity cost.

The pressure target is clear. DIY builders and independent professionals lack the purchase volume needed to secure preferential allocations.

Their forced response is equally clear. They must delay upgrades, accept smaller drives, reuse hardware, choose less transparent prebuilts, or reconsider local workloads.

The 220% Headline Does Not Describe Every SSD

The market is clearly distressed, but one extreme product comparison should not become a universal inflation rate for all consumer storage.

The Samsung example began with a promotional price from a major sales event. Comparing that deal with a current retail listing captures a real buyer experience, but not a neutral annual market index.

Promotional inventory can disappear quickly. Retail algorithms can raise prices when stock becomes scarce, and third-party sellers can remain visible after mainstream offers sell out.

Manufacturer list prices add more noise. A retailer can display a dramatic percentage discount even when the resulting transaction level remains far above earlier street prices.

That is why the 221.5% figure should be treated as a product-specific observation. It demonstrates the severity of the market without proving that every SSD rose by the same amount.

The broader evidence still supports the central conclusion. Tom Hardware found large increases across multiple interfaces and capacities, while industry researchers recorded extraordinary contract-price growth.

Independent supply-chain testimony also points in the same direction. Silicon Motion described NAND being redirected toward data centers and finished drives moving from retail channels toward PC manufacturers.

Micron’s consumer exit adds a corporate decision to those price and allocation signals. Together, these sources make a stronger case than any isolated online listing.

Several uncertainties remain. Weak consumer demand can eventually reduce purchasing pressure, leaving module makers with inventory they must discount.

Technological progress can also increase the number of bits produced per wafer. Newer 3D NAND stacks place additional memory layers vertically, raising density without requiring proportional factory expansion.

Chinese suppliers are another variable. TrendForce expects their production capacity to grow as new equipment enters service, expanding their share of global NAND bit output.

However, increased bit supply does not guarantee immediate retail relief. New output must qualify for products, reach module makers, and arrive in consumer channels before it affects street availability.

Trade policy and regional market access can also determine where that supply appears. A drive available in one country may not have equivalent distribution, support, or warranty coverage elsewhere.

Demand presents its own uncertainty. AI infrastructure spending has remained strong, but it is concentrated among large cloud providers and platform companies.

A procurement pause by even a few major buyers could change inventory expectations. Suppliers might then redirect more capacity to client devices or accept lower contract prices.

The opposite outcome is also possible. Agentic AI services can increase storage consumption through repeated inference, larger caches, and growing volumes of generated media.

TrendForce’s July assessment found that servers already represented more than 40% of total NAND bit demand. Smartphones and notebooks together accounted for nearly another 40%.

This composition gives weak consumer electronics demand an important role. Falling notebook and smartphone production can help balance the NAND market, even while hurting those industries.

TrendForce forecast a 15% to 20% decline in global smartphone production during 2026. It also estimated that notebook shipments would decline around 10%.

Those contractions limit how far suppliers can push consumer pricing. At some point, buyers delay replacements and manufacturers reduce orders.

The skeptical interpretation is therefore not that the shortage is fictional. It is that retail extremes can overstate a uniform market move and hide differences between products.

Buyers should compare current transaction levels across several reputable sellers. They should also separate genuine availability from listings that remain online at unrealistic terms.

Historical price trackers are useful, but sale events require context. The proper question is whether an available drive fits the workload and current market, not whether it matches the lowest promotion ever recorded.

Nobody can identify the exact bottom while supply remains constrained. A failed drive or blocked project can make indefinite waiting more costly than purchasing a reasonable alternative.

The evidence supports caution, not panic. Consumer SSD conditions are poor, yet that does not make every displayed discount credible or every premium drive necessary.

Meaningful SSD Price Relief Is Still a 2027 Story

The strongest available forecast places broad supply relief in the second half of 2027, leaving buyers exposed through the coming year.

TrendForce’s July NAND update estimated a 4% to 5% supply deficit for 2026. It attributed the shortage to strong AI demand, limited factory space, and restrained capacity expansion.

Suppliers are relying mainly on process migrations to increase output. Existing facilities remain constrained, while manufacturers continue prioritizing DRAM expansion.

The firm expects Korean, American, and Japanese producers to upgrade existing lines and selectively expand output. Chinese suppliers are expected to add more substantial capacity.

That combination should eventually push supply growth above demand. TrendForce’s 2027 supply forecast places the turn toward balance in the second half of that year.

Three signals will determine whether that timeline holds.

First, buyers should watch NAND contract prices and the estimated supply deficit. Smaller quarterly increases would show that customer resistance and bit-output growth are weakening supplier pricing power.

A persistent deficit near the current estimate would support the view that retail relief remains distant. A move into surplus would weaken it and improve the chances of renewed promotions.

Second, cloud capital spending and enterprise SSD orders require close attention. Continued procurement growth would keep data-center customers at the front of the allocation queue.

Any broad reduction in enterprise orders would matter before it reached retail shelves. Module makers and NAND producers would need alternative channels for unused output.

Third, buyers should track actual consumer availability rather than launch announcements. More in-stock models, stable listings, and narrower gaps between capacity levels would indicate healthier competition.

The return of meaningful retail volume matters more than a new controller standard. Faster interfaces offer little benefit when the drives remain unaffordable or unavailable.

For immediate purchases, workload discipline offers the best defense. Gaming systems rarely need flagship sequential performance, while archive storage should prioritize capacity, reliability, and backup design.

Developers should separate active projects from cold assets. Keeping current repositories, virtual machines, and datasets on faster storage can reduce the capacity required at premium performance levels.

Creators can use a similar hierarchy. Active media belongs on responsive storage, while completed projects can move to slower local tiers with independent backups.

Buyers considering a prebuilt system should request the exact SSD model when possible. They should also confirm whether the manufacturer reserves the right to substitute an equivalent component.

A vague capacity specification deserves scrutiny. Equivalent size does not guarantee equivalent endurance, cache behavior, or sustained performance.

Those who can wait should monitor several models instead of one favorite product. The current shortage can produce temporary availability differences between brands with similar real-world capabilities.

Those who cannot wait should avoid paying for unused performance. A well-reviewed PCIe 4.0 drive can remain a better decision than a scarce flagship selected only for benchmark results.

The larger lesson from the Tom Hardware price check is uncomfortable. Consumer storage no longer sets the market’s priorities.

AI infrastructure buyers now influence allocation, product strategy, and the timing of new capacity. DIY users experience that shift through fewer choices and weaker purchasing power.

Relief will require more than one sale or one new SSD family. Supply growth must catch demand, consumer channels must receive more inventory, and retail competition must return.

Until those changes appear together, treat storage capacity as an early design constraint. Audit what the system genuinely needs, preserve reliable backups, and compare available interfaces without assuming newer is essential.

Watch the three signals over the next several quarters: NAND contract prices, enterprise procurement, and genuine retail availability. If all three soften, the shortage thesis weakens. If they remain tight, the Tom Hardware findings will look less like an extreme snapshot and more like the consumer market’s new baseline.