SSD prices in 2026: what a terabyte of flash costs

What a terabyte of flash costs, by capacity

Flash pricing is not a ladder. It is a curve with a floor in the middle: the smallest drives carry a controller-and-packaging penalty, the largest carry a scarcity premium, and the cheapest gigabyte sits at 2TB on NVMe and 4TB on SATA. That shape is the single most actionable fact on this page, and it is the opposite of how memory kits price.

Cheapest new listing per capacity tier, not an average of the tier. The basket is Amazon US for every row, so these are single-basket readings rather than a market median; where a second tracker covering eight retailers reads the same tier, the row carries a range instead of a point. Price per TB divides the listing price by raw capacity.

Buy now or wait? The verdict by situation

Storage is a different call from memory. The RAM tracker says buy, flatly, because nothing in the supply picture improves. Flash has one honest counter-signal that RAM does not: the contract increase has decelerated sharply and the spot market has already turned. What changes by situation is which drive, not whether you are being gouged.

Where the bits went

Every price above is downstream of one decision, and it is measurable. Counterpoint Research puts enterprise SSDs at 48% of all NAND bits shipped in 2Q26, against 26% in 2Q25. That is 22 percentage points of the entire world supply of flash reassigned to datacenter drives in twelve months, or 1.8x the share it held a year ago. Nothing about consumer demand had to change for a consumer drive to double in price. The bits simply went somewhere else.

Enterprise SSD against every other destination combined (client SSD, mobile, memory cards, embedded). The residual is stated as a residual: Counterpoint publishes the enterprise share, and everything else is what is left.

Who ships the bits now

The other half of the Counterpoint reading is who supplied them, and it produced a first. China's YMTC took 14% of NAND bits shipped in 2Q26, entering the global top three by volume for the first time, on bit shipments up 22% year over year. Samsung, still first, is down to 25% from 32% two years earlier. The detail that matters more than the ranking is the mismatch: YMTC is number 3 by bits and number 5 by revenue. Shipping volume and earning money from it have come apart, which is Counterpoint's own conclusion: profitability is now "decided less by who ships the most bits, but who ships the right mix". The right mix means enterprise, and that is exactly why the consumer aisle is short.

Bit shipment share, which is volume rather than money. It is the cleanest measure of where flash physically went, and it is a different ranking from revenue share: the same quarter had YMTC third by bits and fifth by revenue.

Enterprise SSD became the whole business

Follow the money to the same conclusion. TrendForce put combined revenue at the top five enterprise SSD brands at $18.46 billion in 1Q26, up 86.1% in a single quarter, with contract prices up around 80% over the same three months and supplier inventories at historic lows. Consumer drives are not competing with each other for supply. They are competing with that.

How the 2TB ladder moved in four months

This is the only free before-and-after with a stated method. GamersNexus collected Newegg listing prices through PC Part Picker price history for named drives between November 12, 2025 to March 12, 2026. The average tracked 2TB NVMe drive went from $169 to $358, a rise of 113.7%. SATA moved less, 76.3%, which is what you would expect of the parts furthest from an AI server.

The channels disagree, and that is the news

Spot, contract and retail are three different markets for the same gigabit, and right now they point in different directions. Contract is still rising because it is negotiated in advance and lags. Spot is a thin, opportunistic market and it has already rolled over. Any claim that flash prices are simply going up in a straight line is blending the two.

Supplier contract prices, the channel OEMs buy on. TrendForce publishes these as percentage moves rather than absolute levels in its free releases, so no dollar figure is available here. Mainstream NAND means 3D TLC and QLC, the flash in ordinary SSDs. Compounded, those three forecasts imply contract prices 2.49 to 2.78 times their level going into 2026, if each quarter landed at the bottom and the top of its published range. That is an implication of three forecasts, not an observation.

SLC NAND is not the flash in your SSD

The eye-catching NAND percentages circulating in 2026 mostly belong to SLC NAND and NOR Flash, which are industrial, automotive and embedded parts on separate supply lines. The flash in an ordinary consumer or datacenter SSD is 3D TLC or QLC. Over the first half of 2026 the gap was smaller than the headlines imply: SLC NAND rose 130 to 150% while mainstream NAND contract prices compounded to roughly 126 to 142% across the same two quarters. The divergence is in the second half, where SLC is forecast up another 70 to 75% and mainstream NAND +10 to 15% in 3Q26. Two errors follow from ignoring that: quoting a half-year figure as though it were a quarter, and pricing a consumer drive off an automotive part.

SLC NAND and NOR Flash are industrial, automotive and embedded parts on their own supply lines. Their price moves are far larger than mainstream 3D TLC and are routinely quoted as if they described consumer SSDs. They do not.

Enterprise flash: the premium almost vanished

Enterprise NVMe drives normally cost far more per terabyte than consumer ones, because they are rated for orders of magnitude more writes. Not any more. The cheapest tracked enterprise row is $117 per TB, only 12% above the cheapest consumer NVMe per terabyte. Consumer flash did not get cheap; consumer flash caught up to the datacenter, which is what happens when the datacenter is the customer suppliers are optimising for.

Live listed prices for enterprise U.2, U.3, E1.S and E3.S NVMe drives from server-parts resellers, a different channel from both consumer retail and supplier contract pricing. New and refurbished are never averaged together; only new rows are listed here. Across all 32 tracked drives the range runs $117 to $1,151 per TB, with the tracked average around $364 per TB and reported up 51.5% over seven months to August 2026.

Against spinning rust

The question a storage shortage actually creates is whether to go back to hard drives. For bulk cold data the answer is still yes on price, and the shortage widened rather than narrowed the case: $28 per TB at the 14TB tier against $93 for the cheapest tracked flash. The part most advice leaves out is that hard drives are on allocation too.

The lead-time row is the one that changes behaviour. TrendForce reported that "lead times for nearline HDDs have ballooned from just a few weeks to over 52 weeks", driven by the same inference demand pulling flash into the datacenter, and it also puts high-capacity QLC SSD power draw about 30% below a nearline drive. So datacenter buyers are substituting toward flash rather than away from it, which is why both sides of this comparison are short at once. Treat a bulk hard-drive purchase as something to schedule, not something to do on Friday.

Capacity-tier median price per TB across eight retailers, not the cheapest single listing, so it is not directly comparable to the consumer flash rows above. Included because the buying question a storage shortage creates is whether to go back to spinning rust.

What it did to the price of a console

Consoles are the clearest read on pass-through anywhere in consumer electronics, because they are normally sold at or below cost, so a component increase has nowhere to hide. Both of the notices below are first-party. Only one of the two companies names storage as the cause, and the difference is worth noticing.

That sentence is the most useful forward-looking statement on this page, and not because Microsoft can see the future. It is a company with long-term supply agreements telling its own customers what it has already been quoted. A contract forecast is an analyst's estimate. A price increase a platform holder has committed to is a purchase order that already happened. Nintendo, by contrast, cites only "various changes in market conditions, which are expected to extend over the medium to long term" and does not name memory at all; the memory attribution in coverage of the Switch 2 increase comes from president Shuntaro Furukawa and is secondary, so it is labeled that way here.

Why: the suppliers chose this

A shortage caused by demand looks like flat revenue on constrained volume. A shortage caused by allocation looks like this: revenue up 77% in a quarter on higher prices rather than higher shipments, while the capital that would add flash capacity goes to high bandwidth memory instead.

TrendForce attributes the increase primarily to higher NAND Flash average selling prices driven by AI server demand, not to higher bit shipments. That is the cleanest evidence that the price move is an allocation decision rather than a volume shortage.

Planned 2026 capital expenditure as reported by TrendForce in November 2025. The asymmetry is the point: DRAM and HBM budgets went up across all three large suppliers, while NAND investment is flat to cut at two of them. New flash capacity is what ends a flash shortage, and it is the line nobody is funding.

What would actually end this

Not a demand collapse, and not the incumbents. New flash capacity ends a flash shortage, and the only supplier building it aggressively is the one that cannot charge much for it. YMTC shipped 14% of the world's NAND bits in 2Q26, third by volume, on 267-layer flash built with its Xtacking 4.0 architecture, while ranking 5th by revenue. That gap is the whole story: a supplier winning on volume and losing on price is the one force in this market with a reason to keep adding wafers when everyone else is adding HBM instead.

The NAND analogue of CXMT: the one supply-side force large enough to change the price of flash. The share and growth figures are Counterpoint. The capacity and roadmap figures are trade press and are labeled as reported until a primary confirms them. The share and growth figures above are Counterpoint. The capacity and roadmap rows are trade press and are labeled as such until a primary confirms them, which is the same treatment the memory cluster gives CXMT. The honest read either way: a fab that starts producing in 2027 does not change what a drive costs in 2026. Full detail in what YMTC is and whether it can break the flash shortage.

The spot readings

Spot is a thin, opportunistic market and a different channel from contract. The two 2025 to early 2026 readings were compiled by GamersNexus from archived DRAMeXchange snapshots; the current reading comes directly from a TrendForce weekly spot update. Both track the same 512Gb TLC session average, but the series is not continuous and no free source publishes every week of it.

From $2.70 in October 2025 to $21.13 now is 7.8x on a wafer of flash, which is the number underneath every retail price on this page.

Every figure on this page, with per-row source attribution and verification dates, is available as one machine-readable file: ssd-prices.json.

← Back to Capital & Compute