When Will RAM Prices Go Down? Should You Buy Now or Wait in 2026?
- 21 hours ago
- 29 min read
RAM has quietly become the most painful line item in any PC build or upgrade. A 32GB DDR5 kit that cost around $110 to $140 a year ago now runs roughly $390 to $400, and DDR4 has not been spared either. The cause is not a single bad quarter. It is a structural shift: AI data centers are absorbing memory capacity that used to go into ordinary laptops and desktops. This article gives you the direct, evidence-based answer to when relief might actually arrive, and whether buying today makes more sense than waiting.
TL;DR
Contract DRAM prices are still climbing in Q3 2026, but the rate of increase is slowing as consumer buyers hit an affordability ceiling.
Samsung, SK hynix, and Micron all describe the memory market as undersupplied through at least 2026, with SK hynix's CEO calling 2027 the industry's "worst year" for shortage.
The earliest plausible window for real, sustained retail relief is late 2027 to 2028, tied to new fabs from SK hynix and Samsung. Some forecasts push full normalization to 2030.
DDR4 is not a safe budget escape. Manufacturers are winding down DDR4 production, which has pushed some DDR4 chip spot prices to record highs.
If you need a working PC now, buy now and buy only the capacity you actually need. Speculative buying for "future-proofing" is expensive insurance in this market.
Waiting 1 to 3 months is unlikely to help. Waiting until late 2027 might, but is not guaranteed.
When Will RAM Prices Go Down?
RAM prices are unlikely to fall in 2026. Contract prices are still rising, just more slowly. The earliest plausible window for a real, sustained decline is late 2027 into 2028, when new fabs from SK hynix and Samsung begin adding meaningful supply. If you need a PC now, buy the capacity you need; waiting a few months is unlikely to help, and waiting years is the only strategy with real evidence behind it.
Table of Contents
When Will RAM Prices Go Down? The Short Answer
No credible forecaster currently expects RAM prices to fall in 2026. TrendForce's latest survey, published July 3, 2026, still projects conventional DRAM contract prices to rise 13% to 18% quarter-over-quarter in Q3 2026, and a further 3% to 8% in Q4 2026, after revising both figures upward from earlier, lower estimates on June 30, 2026.
That is a slowdown in the pace of increases, not a price decline. Samsung told investors on July 30, 2026 that it expects the memory market to stay undersupplied through the second half of 2026 as server DRAM, enterprise SSD, and HBM demand keep outpacing supply. SK hynix's CEO went further on July 10, 2026, telling Reuters that 2027 would be the "worst year" in the industry's history for supply, with demand likely exceeding capacity "even beyond 2030."
Putting the evidence together: the earliest plausible window for a real, sustained decline in retail RAM prices is late 2027 to 2028, when the first meaningful new fab capacity from SK hynix and Samsung is scheduled to come online. Some forecasts, including SK hynix's own, extend the shortage well past that. Nothing in the current data supports betting on a 2026 price crash.
What RAM Costs Right Now
Prices checked: August 24 to 25, 2026, based on Tom's Hardware's PCPartPicker-sourced RAM price tracker and TrendForce spot-price data. Retail prices change daily and vary by retailer, so treat these as a representative snapshot rather than a fixed price.
Configuration | Representative price, Aug 2026 | Approx. price, Aug 2025 | Change |
|---|---|---|---|
16GB DDR5 (2x8GB) | ≈ $210–$240 | ≈ $55–$70 | +200% or more |
32GB DDR5-6000 (2x16GB) | ≈ $392–$400 | ≈ $110–$140 | +180% to 260% |
64GB DDR5 (2x32GB) | ≈ $849 | ≈ $220–$260 | ≈ +230% |
96GB DDR5 | ≈ $1,799 | Not widely sold in 2025 | N/A |
128GB DDR5-6400 | ≈ $3,399 | ≈ $450–$500 | +600% or more |
32GB DDR4-3200 (2x16GB) | ≈ $188 | ≈ $60–$90 | +110% to 210% |
DDR4 1Gx8 spot chip | $42.45 (record, Aug 7, 2026) | ≈ $3–$4 | Over 1,000% |
Two things stand out. First, DDR5 kit prices scale badly with capacity right now: the jump from 32GB to 128GB is far steeper than it would be in a normal market, because every additional gigabyte competes for the same constrained wafer supply. Second, DDR4 has not been a refuge. The benchmark DDR4 spot chip hit its highest recorded price on August 7, 2026, even though DDR4 is the older, supposedly less desirable standard.
How Much Have RAM Prices Increased?
The scale of the run-up is easiest to see quarter by quarter. TrendForce reported that conventional DRAM contract prices rose roughly 93% to 98% quarter-over-quarter in Q1 2026 alone, contributing to an 81% jump in overall industry revenue to about $97 billion for that quarter. Prices kept rising through Q2 2026, then TrendForce projected a further 13% to 18% increase for Q3 2026, on top of an already record-high base.
At retail, that translated into 32GB DDR5-6000 kits moving from roughly $110 to $140 in mid-2025 to about $392 to $400 by late August 2026, according to Tom's Hardware's tracking. A 64GB DDR5 kit that cost a few hundred dollars in mid-2025 was selling near $849 by August 2026. It is worth separating two ideas here: DDR5 climbing from a genuinely low starting point, and DDR4, an older and theoretically cheaper technology, rising just as sharply in percentage terms because of production cuts described in the next section.
This is not the first DRAM up-cycle. The 2017–2018 boom, driven by smartphone growth and cryptocurrency mining demand for GPUs, also saw sharp price spikes before a steep correction within about two years as new capacity came online. The 2026 cycle looks structurally different for reasons explained under "Could High RAM Prices Last Into 2028 or Beyond."
Why Are RAM Prices So High in 2026?
The short version: AI infrastructure spending created a sudden, enormous appetite for high-value memory products, primarily HBM and high-capacity server DRAM. Manufacturers have limited wafer and cleanroom capacity, so building more of those profitable products means building less of the ordinary DDR5 and DDR4 that goes into consumer PCs.
The chain runs roughly like this: AI infrastructure growth creates enormous demand for HBM and server memory, which leads manufacturers to prioritize those higher-margin products, which leaves less cleanroom and wafer capacity for conventional PC DRAM, which raises the cost module makers pay for chips, which raises the price retailers pay for finished modules, which raises the price you pay at checkout.
Micron's own earnings materials make this explicit. In its fiscal Q3 2026 prepared remarks, the company cited "cleanroom constraints and long construction lead times, higher HBM trade ratio, higher HBM growth rates, and declining bits per wafer growth from node migrations" as the structural reasons supply cannot simply expand quickly to meet demand. Building new fab capacity, Micron noted, faces "long lead time for fab construction across the world, shortage of workers with critical trade skills, complex regulations including permitting, and the need for enhanced energy infrastructure."
A second, less-discussed factor: inventories across the industry were unusually low heading into this cycle. Samsung told investors that DRAM and NAND inventory levels were "significantly low" as of its Q2 2026 report, which limits suppliers' ability to smooth out price swings by simply selling from stock.
How AI and HBM Are Affecting Desktop RAM
It is tempting to say "AI uses all the RAM," but that is not quite accurate, and the imprecise version misses why your desktop DDR5 kit is affected at all. AI accelerators like Nvidia's GPUs use High Bandwidth Memory, or HBM, not standard desktop DIMMs. HBM is a stack of DRAM dies connected through silicon interposers, built for extreme bandwidth rather than affordability.
The connection to your DDR5 kit is manufacturing capacity, not the chip itself. HBM and conventional DDR5 both start as DRAM dies made in the same kinds of fabs, on related process technology, competing for the same limited cleanroom space, wafer starts, and engineering talent. Micron has stated repeatedly that HBM requires roughly three times the wafer area per bit compared with conventional DDR5, and that this ratio increases with each new HBM generation. One analysis citing TrendForce data put AI infrastructure's share of global DRAM wafer output at approximately 23% in 2026, up from about 19% in 2025.
Think of a bakery with one oven. A cake (HBM) uses far more oven time and ingredients per unit than a loaf of bread (conventional DDR5), and sells for a much higher price. If demand for cakes explodes, the baker reasonably shifts oven time toward cakes, and there is simply less oven time left for bread, even though nobody baked a single cake out of bread flour. That is the practical relationship between HBM and your desktop RAM: not identical products, but direct competitors for the same limited manufacturing resource.
Server DRAM adds a second, related pressure. General-purpose x86 servers using RDIMM memory are the primary platform for agentic AI workloads, according to TrendForce's July 3, 2026 memory report, and that demand competes with PC DRAM for the same conventional (non-HBM) production lines, not just with HBM.
Why DDR4 Is Not Necessarily the Cheap Escape
DDR4 looks, on paper, like the obvious workaround: older technology, presumably cheaper to make, presumably less in demand. The evidence does not support treating it as a safe budget option in 2026.
Samsung and SK hynix are winding down DDR4 production to redirect capacity toward DDR5 and HBM, though South Korean reporting indicates both companies plan to continue some DDR4 output into 2027 rather than stopping immediately. China's ChangXin Memory Technologies (CXMT) reportedly ended most of its DDR4 production as well. The result is a shrinking supply of an increasingly scarce part, at the same time as builders squeezed by DDR5 prices are turning to older AM4 and LGA1700 platforms specifically to use DDR4, adding demand on top of falling supply.
That combination shows up directly in the data: the benchmark DDR4 1Gx8 3200 spot chip hit a record $42.45 on August 7, 2026, according to TrendForce and DRAMeXchange figures, a level higher than at any point during the 2017–2018 boom. A 32GB DDR4-3200 kit that cost roughly $60 to $90 in October 2025 was running around $150 to $188 by mid-2026 in retail tracking.
None of this means DDR4 is a bad choice for every buyer. If you already own a DDR4 platform and simply need more capacity, DDR4 is very likely still cheaper per gigabyte than DDR5 today, and buying into your existing platform avoids the much larger cost of a full new build. What the evidence rules out is the idea that switching to DDR4 is a reliable way to dodge the shortage altogether, or that a brand-new DDR4 build makes sense purely to "save on RAM," since a new DDR4 platform is now a legacy choice with a shrinking parts supply of its own.
What Memory Manufacturers Are Saying
The clearest evidence about future pricing comes directly from the three companies that make almost all of the world's DRAM.
Samsung Electronics
Samsung's Q2 2026 results, announced July 30, 2026, reported record revenue of KRW 171.5 trillion and operating profit of KRW 89.5 trillion, driven substantially by memory. Samsung told investors that server DRAM, enterprise SSD, and HBM demand is expected to accelerate in the second half of 2026, keeping the market undersupplied "despite partial demand moderation in mobile and PCs." Management separately indicated it expects memory shortages to persist through 2028.
SK hynix
SK hynix listed on Nasdaq on July 10, 2026 in the largest-ever U.S. listing by a foreign company, raising $26.5 billion. On the same day, CEO Kwak Noh-jung told Reuters that 2027 would be the industry's "worst year" for supply, and that customer demand is likely to exceed the company's supply capacity "even beyond 2030." On August 7, 2026, SK hynix's board approved roughly $38 billion for two new fabs, Y2 in Yongin and M17 in Cheongju, whose cleanrooms are not scheduled to open until December 2028 and June 2029 respectively.
Micron
Micron's fiscal Q3 2026 prepared remarks reiterated that supply-demand conditions for DRAM and NAND are expected to "remain tight beyond calendar 2026," attributing this to cleanroom constraints, HBM's growing wafer trade ratio, and slower bit growth from node migrations. Micron completed its calendar 2026 HBM supply agreements, including HBM4, in advance, and separately confirmed that it wound down its consumer-facing Crucial brand by the end of February 2026 to redirect output toward larger strategic and enterprise customers.
The consistent theme across all three companies: none forecasts a return to pre-shortage pricing within the current forecasting horizon, and each has taken concrete steps (long-term agreements, capacity reallocation, or brand exits) that only make sense if they expect the shortage to persist for years, not months.
RAM Price Forecast for the Rest of 2026
TrendForce's most current published guidance, from July 3, 2026, calls for conventional DRAM contract prices to rise 13% to 18% quarter-over-quarter in Q3 2026, and 3% to 8% in Q4 2026, both figures raised from earlier, lower estimates published on June 30, 2026. Server DRAM contract prices are separately projected to rise 13% to 18% in Q3 2026, driven by strong hyperscaler demand and multi-year long-term agreements.
Two forces are pulling in opposite directions for the rest of 2026. On one side, inventories remain very low across the industry, server and AI demand keeps growing, and suppliers face little incentive to discount while every unit sells. On the other side, TrendForce and Tom's Hardware both reported in early July 2026 that consumer electronics buyers were reaching the limit of what they will pay, which is why the pace of increase is slowing even as the direction stays upward. That is a real, verifiable trend, but it describes decelerating inflation, not falling prices; the distinction matters and is explained further below.
Bottom line for the rest of 2026: expect retail prices to keep climbing, likely at a somewhat slower pace than the extraordinary jumps seen in the first half of the year, with no credible forecaster currently projecting an outright decline before year-end.
RAM Price Outlook for 2027
2027 looks less like a recovery year and more like the peak of the current cycle, based on the weight of current evidence. SK hynix's CEO specifically called 2027 the "worst year" for the memory shortage. TrendForce's initial 2027 guidance, discussed in its July 9, 2026 report, estimates total server RDIMM bit supply will grow only 15% to 20% year-over-year in 2027, significantly lagging projected server CPU shipment growth, which points to a server DRAM shortage building through the year even before accounting for consumer demand.
There is a more encouraging thread inside the same evidence. TrendForce's July 30, 2026 outlook describes 2027 as a "diverging" market: DRAM supply remains tight, but NAND Flash supply conditions are expected to ease as new capacity comes online and consumer demand stays weak, with the NAND sufficiency ratio potentially turning positive. That is a meaningful signal, because it shows the same research firm expects capacity additions to start relieving pressure on at least one memory category in 2027, which is the first step toward the same eventually happening for DRAM.
On the supply side, the first new fab capacity that could matter for 2027 comes from projects already under construction rather than the newest announcements. SK hynix's M15X fab in Cheongju is slated for utilization by mid-2027, and the first phase of its Yongin Y1 cluster is targeted for completion around February 2027, with equipment installation beginning in Q2 2027 and initial output later in the year. Micron's first Idaho fab is scheduled to begin producing wafers in the second half of 2027. None of these dates guarantee lower retail prices in 2027; they represent the earliest points at which meaningfully more supply could start entering the market.
Could High RAM Prices Last Into 2028 or Beyond?
Yes, and multiple primary sources say so directly. Samsung's own guidance points to memory shortages persisting through 2028. SK hynix's CEO forecasts demand exceeding supply "even beyond 2030." UBS, cited by Reuters on July 10, 2026, expects the global DRAM industry to remain undersupplied until at least the second quarter of 2028.
The supply-side reason is concrete and dated: SK hynix's newly approved Y2 and M17 fabs, representing roughly $38 billion of investment, will not have their first cleanrooms ready until December 2028 and June 2029 respectively, and a cleanroom opening comes well before equipment installation and volume shipments. Samsung's P5 Fab 2 at Pyeongtaek is targeted for mass production in the latter part of 2028. Micron's Tongluo site, acquired from Powerchip, is expected to support meaningful shipments beginning in fiscal 2028, with a separate greenfield fab in New York not adding output until later still.
None of these dates are guarantees. Fab construction timelines slip regularly, and Korean officials have publicly noted that some of the largest announced investments, including an $800 billion Gwangju and South Jeolla megaproject, do not yet have firm construction start dates, which the government has explicitly left flexible "until the long-term memory chip cycle becomes clearer." Given that ambiguity, the most defensible statement is this: current primary-source guidance supports elevated prices persisting through at least 2027, plausibly into 2028, with real uncertainty about full normalization before 2030.
What Would Make RAM Prices Fall Faster?
A genuinely faster decline is possible, but it depends on several specific things happening together, not just the passage of time.
A pullback in AI infrastructure capital expenditure. If major hyperscalers signal slower AI server buildouts, HBM and server DRAM demand would ease, freeing capacity for conventional DDR5 and DDR4. Analysts covering the Roundhill Memory ETF explicitly flag hyperscaler capex guidance as a key variable to watch.
Faster-than-expected capacity ramps. If SK hynix's Yongin Y1 or Micron's Idaho fab reach volume production ahead of their current 2027 targets, incremental bit supply could arrive sooner than the base case assumes.
Continued, deepening consumer demand weakness. Gartner research cited by CNBC in June 2026 projected global PC shipments could fall 10.4% and smartphone shipments 8.4% in 2026 due to rising memory costs. If that weakness intensifies, module makers and OEMs may accept price concessions to move inventory.
Aggressive competition from Chinese suppliers. CXMT is reported to be targeting up to 30% of the DRAM market by 2030 and has already displaced some Western suppliers among Chinese memory brands, which could add non-Korean, non-Micron supply to the market over time, though quality and performance still lag established suppliers in early testing.
Two consecutive months of falling contract prices. Analysts tracking the Roundhill Memory ETF specifically identify this as a technical signal that has historically preceded corrections in memory pricing and memory-linked equities.
None of these conditions has been confirmed as of late August 2026. They represent the specific, evidence-based paths to an earlier-than-base-case decline, not a prediction that any of them will happen soon.
What Could Keep RAM Expensive for Longer?
Continued HBM and server DRAM growth. Samsung expects HBM4 sales to more than triple quarter-over-quarter in Q3 2026 and account for over 60% of total HBM revenue in the second half. That growth keeps pulling capacity away from conventional DRAM.
Long-term supply agreements that lock in allocation years in advance. Micron has completed pricing and volume agreements covering its entire 2026 HBM supply, and multiple manufacturers are shifting toward multi-year Strategic Customer Agreements rather than annual contracts, which reduces the spot capacity available to smaller buyers and independent module makers.
DDR4 production cuts continuing on schedule, keeping legacy-platform supply tight even as DDR5 eventually eases.
Long fab construction lead times. Every major new fab discussed above, from SK hynix's Y2 to Samsung's P5 Fab 2 to Micron's New York campus, has a first-output date in 2027 or later, and cleanroom construction, equipment installation, and yield ramp typically take a year or more beyond a facility's official opening.
A widening HBM-to-DDR5 wafer trade ratio. Micron has stated this ratio increases with each new HBM generation, meaning the same unit of AI memory demand consumes progressively more of the shared wafer capacity that would otherwise produce consumer DRAM.
Weighing the two lists: the bear case (prices stay high longer) currently rests on statements from primary sources with direct visibility into their own order books and capacity plans, made as recently as August 2026. The bull case (an earlier decline) rests mostly on macro conditions that have not yet materialized. On the current weight of evidence, a sustained decline before late 2027 is possible but not the most likely outcome; further deterioration or persistence into 2028-2030 is plausible; and an outright crash within the next two quarters is unlikely.
How to Tell When RAM Prices Have Really Started Falling
Headlines about "cooling" prices can be misleading, because a slower rate of increase is not the same as a decline. Here is what to actually watch, roughly in the order signals tend to appear.
Signal | What it means | Typically leads or lags |
|---|---|---|
DRAM spot prices falling for several consecutive weeks | Traders expect looser near-term supply | Leading indicator |
Contract-price forecasts turning flat or negative (e.g., TrendForce's next quarterly guidance) | Suppliers themselves expect softer demand or more supply | Leading indicator |
Two consecutive months of falling DDR5 contract prices | Analysts flag this as historically preceding real corrections in memory pricing and memory stocks | Leading indicator |
Producer inventories rebuilding after being at multi-quarter lows | Supply is catching up with demand | Coincident indicator |
Module makers discounting broadly across brands, not just one clearance sale | Cost relief is reaching the retail supply chain, not just one seller's promotion | Lagging indicator |
Retail replacement prices falling across several retailers and brands at once | New inventory is being restocked at lower cost | Lagging indicator |
HBM or server-DRAM capacity pressure visibly easing in manufacturer earnings calls | Manufacturers are reallocating capacity back toward conventional DRAM | Coincident to lagging indicator |
One discounted kit on one retailer's flash sale is not evidence the cycle has turned. Retail prices lag upstream contract and spot pricing because retailers sell through existing inventory before repricing to reflect new costs, and a single promotional listing can reflect a retailer's inventory decision rather than a market-wide shift. Look for the pattern across several of the signals above, sustained over multiple weeks, before concluding that a real decline has begun.
Should You Buy RAM Now or Wait?
There is a real, specific answer here, not just "it depends." The right choice depends mainly on whether your need is immediate and whether the capacity you want is something you genuinely use, not something you are buying out of fear that prices will keep rising.
Buy now if:
The PC is needed immediately for work, school, or a broken system with no working alternative.
Your current RAM capacity is actively limiting what you can do today (frequent swapping, crashes, or inability to run required software).
A genuinely good relative deal appears (a price meaningfully below the recent trailing average for that exact kit, not just below last week's price).
A planned build or upgrade cannot reasonably be delayed for business or personal reasons.
You are buying a small, specific top-up (for example, going from 16GB to 32GB on an existing system) rather than a large speculative purchase.
Wait if:
Your existing system is adequate for your actual workload today.
The purchase is speculative — you are buying because you fear prices will rise further, not because you need the capacity now.
You are considering far more capacity than your workload justifies, purely as insurance against future price increases.
The platform upgrade itself (new CPU, motherboard, or full rebuild) can reasonably wait; buying RAM alone rarely justifies rushing the rest of a build.
You have a defensible, evidence-based reason to expect the market to normalize before you actually need the hardware, such as a build planned for late 2027 or later.
Situation | Recommended action | Why |
|---|---|---|
Need a PC this month | Buy now, sized to actual need | Waiting has no evidence-based upside on a one-month horizon |
Existing RAM causes crashes or slowdowns today | Buy now, minimum sufficient upgrade | The cost of lost productivity outweighs plausible near-term savings |
Current PC works fine for the next 6 months | Wait and monitor the signals above | No forecaster expects meaningful 2026 relief, but delaying an unnecessary purchase avoids paying elevated prices for capacity you do not yet need |
Building a new DDR5 gaming PC, no urgency | Buy the RAM component alongside the rest of the build, at the sweet-spot capacity | Splitting the purchase rarely saves money; price direction is unlikely to reverse enough in weeks to justify delay |
Existing DDR4 owner needs more capacity | Buy now if the platform will remain in service | DDR4 is not guaranteed to get cheaper, and a working platform upgrade is usually the most cost-effective option available |
Buying capacity mainly for future-proofing | Wait, and reassess when the need becomes concrete | Overbuying now locks in shortage-era pricing for capacity you may not use for years |
Workstation or local-AI user needing 64GB or more | Buy now if the workload is active today; otherwise phase the purchase | High-capacity kits carry the steepest shortage premium; buy only what current projects require |
Budget buyer prioritizing value | Consider a smaller DDR5 kit or a compatible DDR4 upgrade on an existing platform | Matching capacity to real need is the single biggest lever a budget buyer has in this market |
Buy-or-Wait Advice by Time Horizon
If you can wait 1 month
Not likely to help. TrendForce's own Q3 2026 forecast still points to rising contract prices this quarter, and retail prices have generally followed that direction through August 2026. A one-month wait mainly exposes you to further increases with essentially no offsetting benefit.
If you can wait 3 months
Still not likely to help materially, based on current guidance. TrendForce's Q4 2026 forecast, revised upward on June 30, 2026, calls for continued contract price growth of 3% to 8% quarter-over-quarter. A three-month wait sits entirely inside the window every major forecaster currently expects prices to keep climbing.
If you can wait 6 months
Borderline, and worth watching rather than assuming. Six months from late August 2026 lands in early 2027, a period SK hynix's own CEO has flagged as the run-up to the industry's "worst year." There is no strong evidence this window brings relief, but it is close enough to the point where new fab timelines (SK hynix's M15X, Yongin Y1 phase one) start to matter that it is worth monitoring the signals described earlier rather than buying purely reactively.
If you can wait 12 months
More plausible, but still not guaranteed. Twelve months out lands in the middle of 2027, squarely inside the period several primary sources (SK hynix, Samsung, UBS) describe as the peak or continuation of the shortage, not its resolution. Some early capacity, like SK hynix's M15X, is scheduled to reach utilization around mid-2027, which could begin easing pressure at the margins, but "worst year" guidance for 2027 argues against expecting a clear price drop within this horizon.
If you can wait until 2027 or later
This is the horizon with the most supporting evidence for eventual relief, though still not certainty. The first wave of genuinely new fab capacity, from SK hynix's Yongin cluster, Samsung's Pyeongtaek P5 facilities, and Micron's Idaho and Tongluo sites, is scheduled to begin contributing meaningful output from late 2027 into 2028. TrendForce's 2027 outlook already anticipates NAND Flash conditions easing in 2027, which is at least a sign that new capacity investments are starting to show up somewhere in the memory market. If your purchase can genuinely wait that long, the evidence supports a real chance of better pricing, though not a guarantee, since demand from agentic AI and long-term supply agreements could keep absorbing new capacity as fast as it arrives.
DDR4 vs DDR5: Which Makes More Sense Right Now?
This decision depends heavily on what platform you already have, not just on sticker price.
Question | Answer |
|---|---|
Is DDR4 still the budget choice? | Usually yes on a per-gigabyte basis if you already own a compatible platform, but the gap has narrowed sharply and DDR4 is not immune to shortage pricing. |
Can DDR4 prices rise even while DDR5 eventually falls? | Yes. DDR4 supply is shrinking as manufacturers wind down production, independent of what happens to DDR5 capacity. |
Is it smart to build a brand-new DDR4 platform just to save on RAM? | Generally no. A new build on a legacy platform locks you into shrinking parts availability for the life of the system, for savings that may not persist. |
Should existing DDR4 owners upgrade RAM now? | Often yes, if the workload genuinely needs it, since upgrading an existing platform is usually far cheaper than a full DDR5 rebuild. |
Should an existing DDR5 owner switch platforms to chase lower prices? | No. There is no current evidence that switching platforms would meaningfully reduce total cost; it would add a full new-build expense on top of already-high memory prices. |
When does buying an older platform become false economy? | When the platform itself is nearing end of life, when DDR4 module supply for your capacity target is scarce, or when the total build cost (CPU, motherboard) erases any RAM savings. |
The practical rule: match your memory purchase to the platform you already have, or the platform you would choose anyway for its own merits (performance, longevity, upgrade path). Do not let short-term RAM pricing dictate a platform decision that will affect the system for years.
How Much RAM Should You Buy?
16GB
Still workable in 2026 for basic web browsing, office work, and older or less demanding games, but it is now a floor rather than a comfortable target for most general-purpose use. Heavier multitasking or modern creative software will feel constrained.
32GB
The mainstream sweet spot for most gaming and general-purpose builds in 2026. It comfortably supports current AAA titles, everyday multitasking, and moderate creative work, without paying the steep capacity premium that starts above this tier in the current market.
64GB
Genuinely useful for serious content creators (video editing, large Photoshop or Premiere projects), software developers running multiple virtual machines or containers, and enthusiasts who keep dozens of browser tabs and applications open simultaneously. Not necessary for typical gaming.
96GB
A niche but growing tier, useful for creators working with very large media files, some local AI experimentation, and users who need headroom beyond 64GB without moving to full workstation territory. Price per gigabyte at this tier is currently steep, so it should be a deliberate choice tied to a specific workload.
128GB and above
Appropriate for workstation use, virtualization hosts, large-scale creative workloads, local large language model experimentation, and scientific computing. This is not a tier to buy speculatively; at current pricing, 128GB kits carry an enormous premium, and most buyers who think they need this much capacity should confirm the requirement against an actual workload first.
A general caution applies across every tier: do not massively overbuy RAM during a shortage purely out of fear that prices will keep rising. That strategy locks in today's inflated price for capacity you may not use for years, and it assumes prices will only go up, which the evidence in this article does not support as a certainty.
Best Ways to Save Money on RAM During a Shortage
Compare total platform cost, not RAM in isolation. Sometimes a slightly different CPU or motherboard choice offsets a RAM price spike elsewhere in the build.
Avoid paying a large premium for RGB lighting or cosmetic heat spreaders when a plain kit with the same timings and capacity is available for less.
Compare timings (CAS latency) as well as frequency; a lower-MHz kit with tighter timings can sometimes outperform a higher-MHz kit with loose timings, at a lower price.
Check your motherboard's QVL (Qualified Vendor List) before buying, to avoid paying for speed or capacity your board cannot actually run stably.
Compare the cost of one larger 2-DIMM kit against buying capacity piecemeal over time; piecemeal purchases risk future compatibility issues and rarely save money in a rising market.
Avoid mixing incompatible kits unless necessary, since mismatched RAM can cause stability issues that cost more in troubleshooting time than any savings.
Consider whether a complete, pre-built PC or laptop with RAM already installed offers relatively better value than buying components separately, since OEMs sometimes have supply agreements that individual consumers cannot access.
Consider used or open-box RAM only from sellers offering adequate buyer protection (verified return policy, warranty transfer, or a reputable marketplace with buyer guarantees).
Set price alerts on specific SKUs you actually intend to buy, rather than shopping reactively.
Watch major sales events, but do not assume a seasonal sale guarantees a genuinely lower DRAM-driven price; retailers sometimes discount other components while holding RAM prices firm.
Avoid buying excessive capacity "just in case"; unused RAM sitting in a drawer is not a hedge, it is a sunk cost at shortage-era prices.
Common RAM Price Myths
"DDR4 is old, so it must keep getting cheaper."
False in this cycle. DDR4 production is being wound down by major manufacturers, which has pushed DDR4 spot chip prices to record highs even as the standard ages.
"If AI demand uses HBM, it cannot affect desktop DDR5."
False. HBM and DDR5 compete for the same wafer fabrication capacity, cleanroom space, and engineering resources, even though they are physically different products built for different purposes.
"One discounted kit means RAM prices have peaked."
Not reliable evidence on its own. A single promotional listing reflects one retailer's inventory decision, not an industry-wide shift in contract or spot pricing. Look for the sustained, multi-source signals described earlier in this article.
"New fabs announced today will fix prices next quarter."
False. Every major new fab discussed in this article, from SK hynix's Y2 to Samsung's P5 Fab 2, has a first-output date of 2027 or later, and full ramp to meaningful volume typically takes additional time beyond that.
"Retail RAM prices move immediately when DRAM spot prices fall."
False. Retail prices lag upstream pricing because retailers sell through existing inventory before repricing new stock, and module makers hold their own inventory buffers as well.
"You should buy 128GB now before it gets even worse."
Not supported by evidence for most buyers. This assumes certainty about future price direction that no forecaster claims to have, and it locks in today's steep high-capacity premium for capacity most buyers do not currently use.
"Waiting always means saving money."
False in the near term. Every current forecast for the remainder of 2026 points to continued price increases, meaning waiting a few weeks or months is more likely to cost more, not less.
"Higher MHz automatically means better value."
False. Real-world performance depends on the combination of frequency and timings (latency), and a cheaper, lower-MHz kit with tight timings can be a better value than a more expensive, higher-MHz kit with loose timings.
Bottom Line
RAM prices are not likely to fall in 2026. The evidence, from TrendForce's contract-price forecasts to direct statements by Samsung, SK hynix, and Micron, consistently points to continued increases through the rest of this year, a possible peak or worsening in 2027, and the earliest plausible window for real relief sitting in late 2027 to 2028, with some forecasts extending uncertainty into 2030.
If you need a working PC today, buy now, sized to your actual workload, not to a fear of future price increases. If your need is genuinely speculative or your current system is adequate, waiting a few weeks or months is unlikely to help based on current evidence, but a purchase planned for late 2027 or later has a real, evidence-based chance of landing in a somewhat better market. Whichever path you choose, watch the specific signals in this article, not headlines about single discounted kits, to judge when the cycle has actually turned.
FAQ
When will RAM prices go down?
No credible forecaster expects RAM prices to fall in 2026. TrendForce's July 2026 guidance still shows rising contract prices through Q4 2026. The earliest plausible window for a real, sustained decline is late 2027 to 2028, tied to new fab capacity from SK hynix and Samsung.
Will RAM prices fall in 2026?
It is very unlikely. Every major forecast published through August 2026, including TrendForce's Q3 and Q4 2026 projections, points to continued contract price increases, just at a slower pace than earlier in the year.
Will RAM get cheaper in 2027?
Unlikely for most of the year. SK hynix's CEO has called 2027 the industry's "worst year" for the memory shortage. Some early new capacity begins reaching utilization around mid-2027, which could start easing pressure at the margins later in the year, but this is not the same as a confirmed price decline.
Should I buy DDR5 RAM now or wait?
If you need a working PC now, buy the capacity you actually need. If your purchase is speculative or your current system is adequate, waiting a few months is unlikely to help based on current forecasts, though waiting until late 2027 or later has more supporting evidence.
Why is DDR5 so expensive?
DDR5 production shares fabrication capacity, cleanroom space, and engineering resources with HBM and server DRAM, both of which are in extremely high demand due to AI infrastructure spending. Manufacturers are prioritizing those higher-margin products, leaving less capacity for consumer DDR5.
Is DDR4 cheaper than DDR5?
Usually yes on a per-gigabyte basis if you already own a compatible platform, but the gap has narrowed considerably, and DDR4 pricing has also risen sharply as manufacturers wind down production.
Why is DDR4 getting expensive?
Samsung, SK hynix, and China's CXMT are all reducing DDR4 output to redirect capacity toward DDR5 and HBM, shrinking supply of an aging standard at the same time some buyers are turning to DDR4 platforms to avoid DDR5 prices.
Will AI keep RAM prices high?
Yes, based on current evidence. AI accelerators use HBM, which competes with conventional DRAM for the same manufacturing capacity, and server DRAM for AI workloads adds a second layer of competing demand. Samsung and SK hynix both describe this demand as accelerating, not easing, through the rest of 2026.
How long do DRAM shortages last?
This cycle has already run longer than a typical boom-bust DRAM cycle, and primary sources expect it to continue. Samsung points to shortages through 2028, and SK hynix's CEO has forecast supply constraints continuing beyond 2030.
Is 32GB RAM enough?
For most gaming and general-purpose use in 2026, yes. It is the current mainstream sweet spot, comfortably running modern games and everyday multitasking without paying the steep premium that begins at higher capacities.
Should I buy 64GB now?
Only if your workload genuinely requires it today, such as serious video editing, running multiple virtual machines, or heavy multitasking with many applications open at once. It is not necessary for typical gaming.
Can RAM prices suddenly crash?
It is possible but not the current base case. A crash would most likely require a pullback in AI infrastructure spending, faster-than-expected fab ramps, or a sharp drop in consumer demand happening together, none of which has been confirmed as of late August 2026.
What signals should I watch before buying?
Watch for DRAM spot prices falling for several consecutive weeks, contract-price forecasts turning flat or negative, and broad discounting across multiple retailers and brands rather than a single sale. A single discounted listing is not reliable evidence that prices have turned.
Is now a good time to buy RAM?
If you need a working PC immediately, yes, buy what you need. If your purchase is speculative, the evidence does not support urgency; waiting a few months is unlikely to help, but waiting until late 2027 or later has a real, though not guaranteed, chance of landing in a better market.
Key Takeaways
The current DRAM cycle is driven by AI infrastructure competing with consumer PCs for shared manufacturing capacity, not a temporary demand blip.
Contract DRAM prices are still rising through Q4 2026 based on TrendForce's most recent guidance; the pace is slowing, but the direction has not reversed.
Samsung, SK hynix, and Micron all describe the market as undersupplied well into 2027, with SK hynix explicitly warning of continued tightness beyond 2030.
New fab capacity that could meaningfully ease prices does not begin arriving until late 2027 at the earliest, and full ramp typically takes longer.
DDR4 is not a reliable escape from the shortage; production cuts have pushed some DDR4 pricing to record highs.
Retail prices lag upstream contract and spot pricing, so isolated sales or discounts are not reliable evidence of a market turn.
The most defensible buying strategy is to match purchases to genuine, current need rather than speculative future-proofing.
Actionable Next Steps
Decide the RAM capacity your actual workload requires today, using the capacity guidance in this article rather than a worst-case guess.
Confirm the exact compatible RAM type (DDR4 or DDR5, speed, and form factor) for your specific motherboard or laptop using its QVL or manufacturer documentation.
Establish a current market baseline by checking two or three retailers for your exact target SKU, using the price snapshot in this article as a reference point.
Compare several sellers, including your motherboard or system manufacturer's own store, before buying from the first listing you find.
Set a target price based on the recent trailing range for that SKU, not last year's price, which is no longer a realistic anchor in this market.
Buy now if urgency or genuine need is present and a fair relative price is available; otherwise, hold off.
If you choose to wait, monitor the market-turning signals described in this article rather than checking prices at random, and revisit your decision on a monthly basis rather than daily.
Glossary
DRAM: Dynamic Random-Access Memory, the type of chip used for a computer's main working memory.
DDR4: The fourth generation of Double Data Rate DRAM, an older but still widely used standard, gradually being phased out by manufacturers.
DDR5: The fifth and current generation of Double Data Rate DRAM, used in most new PCs and laptops built since roughly 2022.
DIMM: Dual In-line Memory Module, the physical stick of RAM installed in a desktop or server.
SODIMM: Small Outline DIMM, the smaller memory module format used in most laptops.
HBM: High Bandwidth Memory, a stacked form of DRAM built for extreme bandwidth, used primarily in AI accelerators and data center GPUs.
RDIMM: Registered DIMM, a type of memory module used in servers that includes a register to improve stability at high capacities.
Contract price: The price memory manufacturers negotiate directly with large buyers, typically set quarterly and reported by research firms like TrendForce.
Spot price: The price of memory chips traded on the open market for immediate delivery, which can move daily and often signals shifts before contract prices catch up.
Wafer: The thin slice of semiconductor material, usually silicon, on which many individual memory chips are manufactured simultaneously.
Process node: The specific generation of manufacturing technology used to produce a chip, which affects how many chips (and how much capacity) a single wafer can yield.
Bit supply: The total volume of memory capacity, measured in bits, that an industry or company produces over a given period.
Memory module: A finished, assembled piece of RAM ready for installation, as opposed to a raw individual memory chip.
CAS latency: A measurement of the delay, in clock cycles, between a memory request and when the data becomes available; lower is generally faster.
XMP: Extreme Memory Profile, an Intel-associated standard that lets RAM run at its advertised speed and timings beyond default settings.
EXPO: Extended Profiles for Overclocking, AMD's equivalent to XMP for enabling rated RAM speeds on compatible motherboards.
Supply-demand balance: The relationship between how much memory the industry can produce and how much buyers want to purchase at a given price.
LTA / long-term agreement: A multi-year contract between a memory manufacturer and a large customer that locks in supply volume and, often, pricing terms well in advance.
Sources & References
TrendForce — AI Server Demand Continues to Support Memory Prices in 3Q26 (July 3, 2026)
TrendForce — Diverging Memory Market Outlook in 2027 (July 30, 2026)
TrendForce — Micron to End Crucial Consumer Memory by Feb 2026 (December 4, 2025)
Samsung Electronics — Announces Second Quarter 2026 Results (Samsung Global Newsroom, July 30, 2026)
Micron Technology — Fiscal Q3 2026 Earnings Call Prepared Remarks (June 24, 2026)
Micron Technology — Fiscal Q2 2026 Earnings Call Prepared Remarks (March 18, 2026)
Micron Technology — Announces Exit from Crucial Consumer Business (December 3, 2025)
The Korea Times — Samsung, SK chip investment timelines leave room for adjustment (June 30, 2026)
CNBC — Inside SK Hynix's $720 billion AI-fueled buildout (August 13, 2026)
Tom's Hardware — 32GB of DDR5 6000 RAM now costs $400 (August 24, 2026)
Tom's Hardware — RAM price tracking 2026 (accessed August 25, 2026)
Tom's Hardware — Micron outlines grim outlook for DRAM supply (December 17, 2025)
Counterpoint Research — AI Demand Reshapes DRAM Rankings in Q2 2026 (August 4, 2026)
