Expansion Is Already on the Way, So Why Won't Memory Get Cheaper Anytime Soon

Wed, 12 Aug 2026 20:35:00 +0800

The pressure from consumer electronics is showing up in end-device pricing and shipment expectations. Both IDC and Gartner’s models suggest that rising memory costs will push up the average selling price of PCs and smartphones while reducing shipment volumes. The specific figures and models differ between the two, so they cannot be combined into a single unified conclusion. This naturally raises a question: aren’t memory manufacturers all expanding capacity? When will memory prices come down?

As of August 12, 2026, the answer based on publicly available information is not straightforward: The second half of 2026 looks more like a slowdown in price increases rather than a broad-based price decline; the second half of 2027 is one of the earliest windows in which NAND Flash supply and demand may gradually ease, but it is neither a DRAM price-cut schedule nor a guarantee for retail prices of PCs, smartphones, and SSDs.

The scope of discussion here is limited to DRAM, NAND Flash, and the PCs, phones, and storage products whose costs are pulled by them. It should not be extrapolated into “when will all semiconductors drop in price”: mature process nodes, logic foundry, MCUs, analog chips, and advanced packaging each have their own supply and demand cycles.

“Capacity Expansion” is not a number that can be directly added

What most commonly appears in the news are investment amounts, factory groundbreaking ceremonies, cleanroom floor space, or the overall production capacity of a wafer fab. But what consumers end up buying are memory modules, flash chips, and modules that have gone through manufacturing, yield ramp-up, packaging and testing, and product allocation. At least three different layers of supply stand in between.

Layer	What It Describes	Why It Does Not Directly Translate to Consumer-Level Price Drops
Wafer or Clean Room Capacity	The physical foundation of fabs and manufacturing capability	It may be foundry-only or not yet in production; it is not equivalent to finished DRAM or NAND output
Bit Output from Process Migration	The increase in storage bits producible on the same wafer	It requires yield, product mix, and market deployment to align; fab count alone is not sufficient
Effective Sellable Supply	Qualifiable finished products allocable to PCs, phones, and SSDs	Allocation can be shifted by priority demand such as HBM, high-capacity DDR5, and enterprise-grade products

TSMC is a good counterexample. Its 12-inch equivalent capacity for 2025 exceeds 17 million wafers, and its capital expenditure for 2026 is projected at 52 to 56 billion USD, with a focus on 2nm, 3nm, specialty processes, and advanced packaging. This is a company-wide wafer metric and cannot be used to calculate the increase in DRAM or NAND output, let alone serve as evidence that memory prices for phones and PCs will fall. The semiconductor industry does not have a unified “total capacity table,” nor does a single unified price curve exist.

Samsung’s materials likewise remind us not to mistake long-range construction plans for near-term supply. The company stated in 2023 that, under its planning at the time, cleanroom capacity by 2027 would reach 7.3 times that of 2021; this is a long-term foundry cleanroom goal, not the number of wafers already in production and available for memory. By 2026, Samsung had additionally announced HBM4 mass production and a supply collaboration with AMD, which can be read as a sign that high-end AI memory is still receiving prioritized advancement, but this is not a complete capacity allocation ranking.

These projects are really about memory, and their timelines are much longer than just “breaking ground.” Micron’s new Singapore NAND fab, announced in January 2026, is scheduled to begin wafer output in the second half of 2028; HBM advanced packaging at the same site is expected to start generating meaningful supply in 2027. Among its U.S. DRAM expansions, the first Idaho fab is expected to produce its first wafer by mid-2027, the second fab by the end of 2028, and the New York project comes even later. A first wafer does not equal mature yields and shippable finished products, and the product mix can be adjusted according to the market; none of these milestones can be equated with new sellable supply of consumer-grade memory in the current quarter.

SK Hynix has announced the construction of the M15X DRAM fab in Cheongju and completed the M16 fab. However, the public fact sheet does not provide current monthly wafer output figures that can be aggregated horizontally with other manufacturers. Rather than piecing together a seemingly precise “capacity ranking of each company,” it is better to acknowledge that the units, timing points, and product calibers publicly disclosed by different companies are inconsistent, and forcing them to be aggregated would only produce erroneous conclusions.

Where Supply Went Is More Important Than Total Supply

What makes this round special is not just the “limited supply,” but the shift in where new resources are being directed. IDC forecasts that DRAM and NAND supply growth in 2026 will be 16% and 17% respectively, while also noting that capacity is being prioritized for AI data center products such as HBM and high-capacity DDR5. Samsung’s mass production of HBM4, along with Micron’s decision to place HBM packaging at earlier nodes at its Singapore facility, also corroborate this phenomenon.

This means that even if the total bit output increases, how much incremental supply consumer-grade conventional DRAM and NAND can obtain still depends on the manufacturers’ product mix and profit considerations. The high-end memory and advanced packaging capabilities required for servers are not fully interchangeable with consumer-grade chips; therefore, “significant AI-related investment” cannot be automatically translated into “256GB SSDs will be cheaper.”

Institutional Assessment: First Distinguish Among the Three Kinds of “Price Reductions”

Market discussions often combine three different signals into a single term: contract price decline, easing supply-demand in the industry, and a decrease in end retail prices. These are three signals that need to be observed separately, and their transmission timing and sequence may not be consistent, nor necessarily occur in the same quarter.

TrendForce’s forecast for Q3 2026 shows that conventional DRAM contract prices will still rise 13% to 18% quarter-over-quarter, while NAND Flash contract prices will increase 10% to 15% quarter-over-quarter. This already falls within a “narrowing increase” context, yet it is still an increase rather than a decrease; moreover, contract prices represent the quarterly settlement benchmark between upstream buyers and sellers, and are not equivalent to retail shelf prices.

Regarding longer-term supply and demand changes, TrendForce projects that NAND will still have a supply gap of 4% to 5% in 2026; with the combined effect of process migration, growth in bit output, and weak consumer electronics demand, the tight situation will gradually ease in the second half of 2027. This is the judgment closest to an inflection window in the available materials, but the key word is “gradual easing.” It cannot be used to derive that a certain brand’s SSD will drop in price on a particular day, nor can it be extended to conclude that DRAM will necessarily decline over the same period.

IDC’s timing assessment is more conservative: price acceleration will slow in the second half of 2026, but prices will still rise and remain elevated, with the baseline model not pointing to a return to 2025 levels. Its judgment on PCs is that the global average selling price of PCs in 2026 will rise by 17%, and even with memory capacity expansion over the next two years, prices are unlikely to return to 2025 levels. For mobile phones, IDC forecasts shipments will fall to 1.09 billion units in 2026, a 13.9% year-over-year decline, drop another 1.1% in 2027, and rebound by 5.5% in 2028 once memory supply normalizes. However, this mobile phone forecast also incorporates geopolitical factors and transportation costs, so the entire decline cannot be attributed to memory.

Gartner’s model presents another measure of pressure: it estimates that the combined price of DRAM and SSD will rise 130% by the end of 2026 compared with 2025, while PC and smartphone prices will increase by 17% and 13% respectively, with corresponding shipment volumes declining by 10.4% and 8.4%. These figures differ from the specific numbers later updated by IDC, and the two should be understood as scenario estimates from different institutions using different models, rather than precise numbers that can be pieced together into an “industry consensus.”

What Long-Term Contracts Lock In Is Supply Visibility, Not Five-Year Net Profit

If we push the question one step further: since these vendors have begun signing multi-year long-term agreements, can we calculate the minimum profit over the next five years based on this? The answer requires more restraint than “when will prices drop.” Public disclosures currently only support a very limited but important conclusion: Micron has disclosed minimum revenue under contracts; the other three have disclosed multi-year partnerships, orders, or capacity plans. None of them have publicly disclosed complete economic terms sufficient to calculate the minimum net profit over five years.

Company	Publicly Disclosed LTAs / Multi-Year Arrangements	Verifiable USD Floor	Can 2026—2030 Minimum Net Profit Be Calculated From This
Micron	16 SCAs typically covering 2026—2030, automotive agreements typically three years; includes minimum purchase volumes and price bands	Cumulative minimum contracted revenue from 14 signed SCAs is approximately $100 billion	No; revenue floor is not net profit
SK Hynix	Completed multi-year LTAs with approximately ten customers	Not disclosed	No
Samsung Electronics	HBM customer POs already in place for 2026, with customers pursuing discussions for multi-year supply in 2027 and beyond	Not disclosed	No
TSMC	Capacity planning jointly based on customer multi-year demand	Not disclosed	No; its logic foundry profits cannot substitute for memory contract profits

Micron is an exception, but only as far as “revenue.” Its management stated that 14 signed agreements have cumulative revenue of approximately $100 billion over the remaining period at contract minimum prices, with the agreements typically extending through the end of 2030; these agreements cover roughly 20% of DRAM shipments and one-third of NAND shipments over the same period. Large contracts typically include price ceilings and floors, and are take-or-pay. This disclosure takes a portion of demand and pricing out of the spot cycle, thereby increasing visibility into future revenue.

But it cannot derive that “Micron’s minimum net income over the next five years is X USD.” Net income also depends on revenue outside the contract, manufacturing costs, yield and product mix, depreciation, R&D and selling expenses, interest, taxes, and other items; the strong gross margin cited by management does not provide a substitutable lower bound on net margin. Dividing 100 billion USD by five and multiplying by the current quarter’s or historical profit margin merely disguises a cyclical peak as a contractual commitment.

SK hynix has confirmed multi-year LTAs with about ten customers; Samsung has confirmed that customers are driving multi-year supply discussions and that the HBM models mass-producible in 2026 have already received orders; TSMC is allocating capacity based on customers’ multi-year demand plans. These developments explain why vendors are more confident in investing, and why consumer electronics cannot simply wait for one additional production line; however, contract amounts, minimum purchase volumes, pricing terms, and costs have not been disclosed. Based on currently public information, only Micron can confirm partial guaranteed revenue of roughly $100 billion from its contracts; none of the four companies can have their five-year minimum net profit calculated from the disclosed materials.

This is not an evasion of valuation, but rather a delineation of one. Long-term contracts are sufficient to reduce part of the uncertainty in sales volume and price, yet they do not turn memory cycles, cost curves, or corporate operating expenses into fixed values. For the reader, it reinforces the original judgment: supply tightness is no longer determined solely by short-term spot quotations; however, any table that translates long-term contracts directly into a five-year “guaranteed profit” goes beyond the public data.

If we adopt the consensus expectations from institutions, we can obtain a valuation table

If one is careful not to confuse “digits” with “long-term guaranteed minima,” the analyst consensus averages compiled by FactSet can be used. The table below shows each company’s “net profit (Gewinn)” series as publicly verifiable in this round; U.S. dollar figures are in billions, rounded to one decimal place. Micron and TSMC were originally denominated in USD; SK Hynix was originally in EUR, and Samsung was originally in KRW — all converted at the European Central Bank reference rate of 23 July 2026 (1 EUR = 1.1392 USD; consequently 1 USD ≈ 1,472.7 KRW). The currency conversion is solely for uniform presentation and does not alter the original analyst forecasts.

Company (Fiscal Year End)	2026E	2027E	2028E	2029E	2030E	Minimum Value in Disclosed Series
Micron (August)	83.4	175.4	192.5	211.1	300.9	83.4 (2026E)
SK Hynix (December)	171.2	229.6	255.1	348.9	533.6	171.2 (2026E)
Samsung Electronics (December)	207.0	294.4	295.5	—	—	207.0 (2026E)
TSMC (December)	86.4	113.5	142.9	177.9	—	86.4 (2026E)

This table answers the question “What is the lowest net profit among the disclosed years in the brokerage average scenario?"—not “the pessimistic lower bound for the next five years,” and certainly not a contractual guarantee. Samsung’s publicly available consensus net profit estimate extends only through 2028, while TSMC’s extends only through 2029; no extrapolation is made for the blank entries. Micron’s 2026E refers to its fiscal year ending August 2026, and cannot be treated as exactly the same observation window as the other three companies’ calendar-year fiscal years.

Notably, the minimum of all four average expected series falls on 2026E. This reflects the current shared assumptions among analysts regarding AI memory demand, the high-end product mix, and industry supply constraints; it does not imply that the next five years will only trend upward. Insufficient coverage by long-term agreements or orders, a renewed weakening of consumer electronics demand, process and yield issues, competitive capacity expansion, and exchange rate fluctuations could all cause these numbers to be revised. They can serve as a thermometer for readers to observe how the market is pricing future earnings, but they cannot replace judgments based on contract terms or consumer-grade memory prices.

What Signals Should Consumers and Buyers Wait For

If the goal is to find a realistic basis for “whether one can wait for a better price,” the first things to observe are not the groundbreaking news of a new factory, but rather the changes at each of the three levels separately, rather than treating them as a fixed buying clock.

First, whether the contract prices settled quarterly by upstream manufacturers and major customers for DRAM and NAND shift from “narrowing increases” to quarter-over-quarter declines; second, whether the industry supply-demand balance moves from a shortfall to equilibrium or even looseness, particularly distinguishing between NAND and DRAM; third, whether the inventory, promotions, and configurations at OEMs and retail channels transmit upstream changes into actual transaction prices. Brands may absorb part of the cost, adjust capacity configurations, or delay price adjustments, so the transmission at the retail end may be slower and may also move out of sync with the first two types of signals.

Based on the available information, the supply-demand easing window for NAND is being pointed to the second half of 2027, while the contract price forecast for Q3 2026 still points upward. For DRAM, the existing materials do not give a credible and明确的 starting quarter for decline; for retail prices of mobile phones, PCs, or SSDs, there is even no verifiable specific date for price reductions. Stating these unknowns clearly is more useful than giving a seemingly tidy but unsupported timeline.

Weakening consumer electronics demand does cap the pace of further price spikes, but it is not a switch that determines prices on its own. The push for high-end AI memory, the bit-growth brought by process-node advances, the speed at which capacity expansions are realized, and manufacturers’ product allocation together determine the effective available supply. As of August 12, 2026, expansion is underway; for consumers, what is truly worth waiting for is not news of factory construction, but the moment supply begins to enter the regular market in the right products and the right quantities.

This article is based solely on publicly available information for industry observation and does not constitute procurement or investment advice.

References

TSMC’s 2025 Form 20-F (verified 2026-08-12)
Samsung Electronics: Samsung Foundry Forum 2023 (verified 2026-08-12)
Samsung Electronics: Samsung and AMD Expand Strategic Collaboration on Next-Generation AI Memory Solutions (verified 2026-08-12)
Micron: Groundbreaking of Advanced Wafer Fabrication Facility in Singapore (verified 2026-08-12)
Micron: Accelerating U.S. Investments, First Concrete Poured at New York Project (verified 2026-08-12)
Micron FY2026 Q3 Conference Call Prepared Remarks (verified 2026-08-12)
FactSet Consensus Estimates: Micron (displayed via finanzen.net) (verified 2026-08-12)
SK hynix Fact Sheet (verified 2026-08-12)
SK hynix 2Q26 Earnings Release (verified 2026-08-12)
FactSet Consensus Estimates: SK hynix (displayed via finanzen.net) (verified 2026-08-12)
FactSet Consensus Estimates: Samsung Electronics (displayed via finanzen.ch) (verified 2026-08-12)
FactSet Consensus Estimates: TSMC (displayed via finanzen.net) (verified 2026-08-12)
ECB Reference Exchange Rate (2026-07-23) (verified 2026-08-12)
TrendForce: 3Q26 Memory Prices Supported by AI Servers, but Consumer-Side Pressure Widens to Constrain Gains (verified 2026-08-12)
TrendForce: NAND Flash Supply Growth to Outpace Demand in 2027 (verified 2026-08-12)
IDC: Global Memory Shortage Crisis (verified 2026-08-12)
IDC: Higher ASPs, Lower Unit Volumes (verified 2026-08-12)
IDC: PC Market Enters Volatile Territory (verified 2026-08-12)
IDC: Worldwide Smartphone Market to Decline 13.9% in 2026 (verified 2026-08-12)
[Gartner: Surging Memory Costs Will Reduce Global PC

写作附记

Original Prompt

$blog-writer search online for production capacity data and expansion plans of various semiconductor companies, as well as estimates from various institutions on the subsequent price trends of semiconductors, when price reductions can begin, and the current significant pressure on consumer electronics.

Add a paragraph to the article: based on published data, estimate the minimum profit data of each company over the next five years under their current long-term agreements, in USD. Evaluate whether the article needs to be rewritten, and remember to append the content I supplemented to the prompt.

No specific calculated data is required; data estimated by institutions is also acceptable.

DRAM on Uncle Xiang's Notebook