Uncle Xiang's Notebook Collecting Moments, Marking Time
← Article list

Capacity Expansion Is Underway. Why Won't Memory Get Cheaper Soon?

Consumer electronics are under pressure, but when can more memory supply actually lower retail prices?

Codex (GPT-5) v8.11.1

Consumer electronics are already feeling the cost of more expensive memory, even as announcements of new capacity keep arriving. Put the two together, and it is tempting to think that phones, PCs, and SSDs should become cheaper once a few more factories open.

The difficulty lies in what happens after they open. New manufacturing capacity must turn into qualified products, those products must reach the consumer market, and only then might prices on store shelves change. Expansion adds manufacturing capability; consumers are waiting for usable supply of the products they need. The distance between the two deserves more attention than the size of the investment.

As of August 12, 2026, the available material supports a limited conclusion: the second half of 2026 looks more like a period of slower price increases; the second half of 2027 is one possible window for NAND supply and demand to ease gradually. There is no equally clear window for DRAM, and no promise of cheaper PCs, phones, or retail SSDs. This discussion concerns memory and storage chips, not all semiconductors.

On this page

Factories, bits, and products people can buy

“Capacity” has at least three meanings. Buildings and wafers provide the manufacturing base; process migrations can increase bit output per wafer; yields, packaging, testing, and product allocation determine the supply that can actually be sold. Figures in the news may all be rising while referring to different points in this chain.

TSMC’s annual capacity in 2025 exceeded 17 million 12-inch-equivalent wafers, and its projected 2026 capital spending was US$52 billion to US$56 billion, focused on advanced processes, specialty technologies, and advanced packaging. These are company-wide figures. They cannot be converted into additional DRAM or NAND supply. Samsung once planned, as of 2023, to expand cleanroom capacity to 7.3 times its 2021 level by 2027. That was a foundry target, not a measure of finished memory output.

Even for actual memory projects, the meaning of each milestone matters. Micron’s disclosed plans include several different stages:

Project Expected milestone What the milestone means
Singapore HBM advanced packaging 2027 Beginning to contribute meaningful supply
First Idaho DRAM fab Mid-2027 First wafer output
Singapore NAND fab Second half of 2028 Start of wafer output
Second Idaho DRAM fab Late 2028 First wafer output

Yields still have to ramp after the first wafers, and the product mix can change. This table is not a consumer supply schedule. SK hynix has announced construction of M15X and completed construction of M16, but its public fact sheet does not provide current monthly wafer figures on a basis that can be added to those of other manufacturers. I would rather leave that gap visible than assemble an artificially precise capacity ranking.

Another question is easier to miss: once production starts, who receives the increase? IDC projects 2026 supply growth of 16% for DRAM and 17% for NAND, while noting that resources are being directed first toward AI data-center products such as HBM and high-capacity DDR5. Samsung’s HBM4 mass production and Micron’s earlier push into HBM packaging fit that direction, but do not establish each manufacturer’s full allocation ratios.

Total supply can therefore grow while consumer supply remains tight. High-end memory, advanced packaging, and consumer chips are not fully interchangeable, so more investment need not reach ordinary SSDs sooner. Equally, blaming all the pressure on AI goes too far: processes, yields, construction schedules, and consumer demand matter as well.

Slower increases are still some distance from falling prices

TrendForce’s forecast for the third quarter of 2026 is a quarter-on-quarter increase of 13%–18% in conventional DRAM contract prices and 10%–15% in NAND Flash prices. Even if increases moderate, settlement prices are still rising. Consumers hearing that conditions are “easing” may not find a lower price on their next shopping trip.

Signal What it tells us What it cannot promise
Falling contract prices Whether upstream quarterly pricing has turned A simultaneous retail price cut
A narrowing supply-demand gap Whether the industry is approaching balance Ample supply of every product
Lower actual purchase prices Whether consumers are paying less That one promotion marks an industry turning point

TrendForce estimates a NAND supply shortfall of 4%–5% in 2026, with process migrations, bit growth, and weak consumer-electronics demand together easing conditions gradually in the second half of 2027. In this outlook, capacity growth and weakening demand both contribute to relief. The date cannot be attributed entirely to new fabs, nor can the NAND forecast be transferred to DRAM.

Why has weaker demand not immediately pushed prices down? At least in these forecasts, demand pressure has not yet eliminated the supply gap. IDC expects price acceleration to slow in the second half, while prices continue to rise and remain elevated; it projects a 17% increase in average PC selling prices in 2026. Its smartphone forecast is 1.09 billion shipments, down 13.9% year on year. Geopolitical conditions and transport costs also enter that forecast, so memory inflation cannot take all the blame.

Gartner estimates that combined DRAM and SSD prices will be 130% higher by the end of 2026 than in 2025, with PC and smartphone prices up 17% and 13%, and shipments down 10.4% and 8.4%, respectively. Its model differs from IDC’s. Both identify pressure, but their figures cannot be spliced together. An average device selling price is also not the price of a particular model: brands absorbing costs, changing configurations, and running channel promotions all affect the increase consumers actually experience.

Long-term agreements support expansion, but do not lock in profit

Multiyear contracts make some future demand more visible and help explain manufacturers’ willingness to invest. But prices, costs, and expenses still separate “customers are willing to buy” from “a minimum net profit over five years.”

Micron’s 16 disclosed Strategic Customer Agreements, or SCAs, generally cover 2026–2030; automotive agreements generally last three years. Fourteen signed agreements represent approximately US$100 billion in cumulative revenue over their remaining terms at minimum contractual prices. The 16 agreements together cover roughly 20% of DRAM shipments and one-third of NAND shipments over the period. Large contracts generally include price bands and take-or-pay arrangements.

That is a substantial revenue disclosure, but it covers only part of the business. Revenue outside the contracts, yields, product mix, depreciation, research and selling expenses, interest, and taxes can still change. Dividing US$100 billion by five and multiplying it by a current profit margin merely treats a margin from one point in the cycle as a contractual term.

The other three companies disclose less. SK hynix has confirmed multiyear LTAs with around ten customers; Samsung has confirmed 2026 HBM orders and discussions about subsequent multiyear supply; TSMC plans capacity around customers’ multiyear needs. Public information does not contain the complete economic terms needed to calculate a profit floor. TSMC’s logic-foundry earnings cannot stand in for profits on memory contracts either.

None of the four provides a basis for calculating guaranteed minimum net profit over five years. Long-term agreements improve visibility into some supply and revenue, but neither eliminate the cycle nor independently prove that consumer prices will keep rising for years.

Analyst forecasts describe scenarios, not minimum profits

The absence of a contractual profit floor does not justify giving an analyst consensus another name and using it instead. Consensus generally describes average estimated earnings at a particular date and under a particular accounting definition. The smallest figure in a sequence of forecast years is simply the smallest figure in that sequence—not the worst outcome the company could experience.

To compare manufacturers, one must retain at least the original currency, amount units, fiscal year-end, definition of net profit, and forecast publication date. Dynamic valuation pages update over time. Today’s figures cannot be backfilled as yesterday’s judgment, and missing years cannot be completed to create a five-year “guarantee.” Earnings forecasts are an especially poor substitute for evidence about individual consumer products: manufacturers may make money from high-end products while ordinary customers still face high costs.

There are differences of view as well. TrendForce identifies the second half of 2027 as a scenario for gradual relief in NAND; Micron’s June 2026 materials judge that tight conditions may persist beyond 2027. Their product scopes and perspectives differ, so they cannot be combined into a single countdown to cheaper memory.

Returning to the consumer’s question, waiting needs a specific object. First distinguish DRAM from NAND. Then watch whether contract prices turn from smaller increases to declines, whether the supply gap narrows, and what people actually pay for the capacity and configuration you need. These three signals may move at different times; they are not a mechanical purchasing countdown.

At the dates covered by these sources, NAND has a possible easing window, DRAM has no clear quarter for the start of declines, and retail prices have no confirmed reduction date. When the next capacity announcement arrives, the useful questions are: when will it produce qualified output, and how much of that output will reach the kind of product I need?

This article is an industry observation based on public information, not purchasing or investment advice.

References

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.

Rewrite the articles from these three months. Please note: articles not written by AI do not need to be rewritten.

Table of contents

Related content

05
  1. 01

    The Old Economics of the Memory Industry
  2. 02

    Xiaomi drops back to hardware ledger: Storage price hikes, auto subsidies, and AI spending hit simultaneously
  3. 03

    The renewed surge of A-share semiconductors should not be bought based on industrial logic alone; caution is needed as this may constitute a crowded trade.
  4. 04

    Relying solely on stock prices to gauge the semiconductor cycle is insufficient; SK Hynix's earnings report serves as a better barometer.
  5. 05

    The big model development has indeed drawn the internet giants into the same competitive arena.

Comments

Comments will load when you scroll here.