← All guides

Should You Buy RAM Now for Local AI, or Wait Out the 2026 DDR5 Shortage?

Do not wait. TrendForce put Q1 2026 PC DRAM contract prices at 105-110% higher quarter over quarter, the steepest single-quarter rise on record, and no major analyst forecasts a decline before 2028. A 64GB DDR5-6000 kit that cost around $180-220 in mid-2025 sold for $680-880 as of August 2026, and a 32GB kit that was $80-120 now runs $375-589. If you need memory for local AI in the next 12 months, buying today is the cheap option — and if you have not bought the machine yet, soldered unified memory is now cheaper than the DIMMs alone.

Bottom Line

Buy now. Not because prices always go up, but because every published forecast points the same way and none of them call a bottom before 2027.

  • Need memory in the next 12 months: buy today. The forecast risk is asymmetric — the downside of waiting is far larger than the upside.
  • Building a new machine: strongly consider soldered unified memory instead of DIMMs. It is the one part of the market that did not fully reprice.
  • Already have enough RAM: do nothing. This is not a reason to upgrade early; it is a reason not to delay a purchase you already needed.
  • Speculating on a price drop: you are betting against TrendForce, Gartner, and AMD’s own guidance at once.

What actually happened to memory prices

TrendForce originally forecast a 55-60% DRAM price increase for Q1 2026. It then revised that to 90-95%. PC DRAM — the DDR4 and DDR5 modules that go into desktops and laptops — was singled out as the hardest-hit category, with contract pricing projected to jump 105-110% in a single quarter. That is the steepest single-quarter rise on record.

The spot data is worse than the contract data. DDR5 chip pricing went from $6.84 per GB in September 2025 to $27.20 per GB by December 2025 — close to a fourfold increase in one quarter.

Two things drove it:

  1. AI data centers ate the supply. Demand for high bandwidth memory surged with accelerator deployments, pulling fabrication capacity away from commodity DDR4 and DDR5.
  2. PC shipments beat expectations. Q4 2025 PC shipments came in much higher than forecast, which escalated an already tight market.

What that means at the checkout page

PartMid-2025August 2026Change
32GB (2x16GB) DDR5-6000 kit~$80-120~$375-589~4-5x
64GB (2x32GB) DDR5-6000 kit~$180-220~$680-880~4x
DDR5 per GB (chip contract)$6.84/GB (Sep 2025)$27.20/GB (Dec 2025)~4x
1TB NVMe SSD~$45-60~$90-165~2x
Mac mini M4, 16GB unified$599$799+33%
RTX 5090 32GB$1,999 MSRP~$4,300-5,000 street~2.2x

Prices are street/contract figures as cited by the sources below, current as of August 2026.

One correction to an earlier version of this table: we previously compared a $875 64GB kit against an $80-100 mid-2025 baseline and called it a 9x rise. That baseline was the mid-2025 price of a 32GB kit, not a 64GB one. Corrected, both capacities rose about 4-5x — enormous, but not nine-fold, and we would rather run the smaller true number than the bigger wrong one. GPU and RAM figures come from different trackers and are not measured the same way — treat the table as direction and magnitude, not a purchase quote.

The thing nobody else is saying: soldered memory became the cheap option

Here is the line that changes the recommendation.

A 64GB DDR5 kit at the top of its $680-880 range costs more than an entire Mac mini M4 with 16GB of unified memory at $799.

That is not a capacity-for-capacity comparison — 64GB of DIMMs is four times the memory — and we are not claiming the Mac is the better local-AI machine at that tier. The point is narrower and stranger: the DIMMs alone now cost more than a complete computer.

For a decade, the standing advice in this niche was that Apple’s memory upgrade pricing was a rip-off and you should buy a PC where RAM is cheap and expandable. That advice was correct. It is now wrong, for a specific and probably temporary reason: Apple, AMD, and the mini-PC vendors priced their soldered unified-memory configurations before the DRAM spike, and have not fully repriced them. Apple did raise Mac mini pricing in June 2026, but by 33% — against DIMMs that moved several times that.

So the machines everyone criticized for non-upgradeable memory are the ones whose memory is currently underpriced relative to the market. If you are buying a whole system rather than adding sticks to an existing one, unified memory is the hedge.

The honest caveat: this inverts back the moment Apple and the mini-PC vendors reprice, and they will. This is a window, not a new law of the universe.

Buy or wait, by situation

You already own a machine and want more RAM

Buy now, and buy the 64GB tier. The 64GB capacity was still the best value per gigabyte at roughly $12-14/GB as of August 2026, against $12-18/GB for 32GB kits. Buying 2x16GB now and “adding more later” is the worst plan available: you pay the shortage premium twice, and you may find the matching kit out of stock.

If you are running local models, memory capacity is what sets your ceiling — see what local LLM fits my machine before you decide how much you actually need. Many people buy 128GB for a workload that never exceeds 32GB.

You are building a new machine

Price the unified-memory options honestly against the parts list before you commit to DIMMs. The best budget local AI PC under $1,000 breakdown runs those numbers — and the short version is that the classic budget build does not fit under $1,000 anymore.

The shortage hedge, in three machines. All three ship with soldered unified memory priced before the DRAM spike:

Amazon affiliate links — we earn a small commission at no cost to you. Prices move fast right now; check before you buy.

You want to speculate on a price drop

You would be betting against Gartner’s forecast of a large memory cost surge, TrendForce’s upward revisions, and AMD’s public statement that it expects DDR5 to stay expensive until around 2028 as new fabs ramp. Industry analysts broadly expect the shortage to persist to at least 2028.

We cannot tell you prices will not fall. We can tell you that no forecast on the record says they will, and that the people with supply-chain visibility are the ones saying 2028.

Why the GPU market is the same story

Do not treat “should I buy RAM” and “should I buy a GPU” as separate questions. Memory chips now account for more than 80% of a GPU’s bill of materials, and rising memory prices have become the dominant factor in GPU pricing.

That has a practical consequence most buy-now-or-wait articles miss: you cannot dodge the memory spike by shifting budget between the two. Waiting on RAM to buy a GPU instead just moves your money into a different product whose cost is also mostly memory. NVIDIA has been preparing GeForce RTX price increases of 20-30% — the third such move since January 2026.

If a GPU is what you are actually shopping for, what to buy through the RTX price hike has the per-tier version of this call.

What we would do

  • Under $1,000 total budget: buy a complete unified-memory machine, not parts. The DIMM market is the worst place to spend right now.
  • Have a working PC, need more headroom: buy one 64GB kit today, at the best $/GB tier, and stop.
  • Have a working PC with enough RAM: buy nothing. Shortage panic is not a use case.
  • Need 128GB+ for 70B-class models: unified memory is now the default answer, not the exotic one. See can 24GB of VRAM run a 70B? for why that tier needs unified memory or multiple cards.

Sources

See Also

Need OpenClaw fixed live?

Remote rescue sessions for gateway, auth, tunnel, VPS, and model access problems.

See Rescue Session

Read next

64GB Local AI Rig: The Complete 2026 Parts List (August 2026 Prices)
Every part for a 64GB-RAM, single-GPU local AI tower in August 2026: AM5 CPU, B650 board, 64GB DDR5 (2x32), GPU by budget, PSU sizing for 24/7 inference, NVMe for a model library, cooler and case. Real prices during the DRAM shortage — roughly $1,700 to $3,100 built.
32GB vs 64GB RAM for Local LLMs: Which Should You Buy?
Should you buy 32GB or 64GB RAM for local LLMs and OpenClaw? A direct comparison with model tiers, tradeoffs, calculator links, and upgrade rules.
Best LLM for 64GB VRAM (July 2026): Dual RTX 5090 Picks, Not Mac RAM
Best local LLM for 64GB VRAM, July 2026: Laguna S 2.1 UD-IQ4_XS (57.6GB), Laguna XS 2.1, gpt-oss 120B Q4. Dual RTX 5090 vs 2x A6000 vs 96GB Blackwell.
Best Local LLMs for 64GB RAM (July 2026): Llama 4 Scout, gpt-oss 120B, DeepSeek V4 Flash & Laguna XS 2.1
Best local LLMs for 64GB RAM in July 2026. Llama 4 Scout (10M context, ~58GB Q4), gpt-oss 120B at Q4, DeepSeek V4 Flash (284B MoE, Ollama cloud), Laguna XS 2.1 (agentic coding, 33B-A3B, ~36GB Q8). Also: Mistral Small 4, Qwen 3.6 35B Q8.