Why Your Next PC Upgrade Costs So Much: Inside the RAM and GPU Price Spiral

Away from the model launches and funding rounds, the AI boom has produced one very physical, very unwelcome side effect for ordinary consumers: memory and graphics card prices have gone through the roof, and the industry itself says the pain isn't over yet. This week, with back-to-school and pre-built PC pricing back in the news, it's worth laying out exactly what's happening — and when, if ever, it ends.

How bad is it, actually?

Bad, and getting worse through the year rather than better. A 32GB DDR4 kit that sold for $60–$90 in October 2025 was going for $150–$180 by January 2026. By June, Tom's Hardware's daily tracker found the cheapest in-stock 32GB DDR5 kit in the U.S. at $374.97 — a capacity that routinely sold for $80–$120 just a year earlier. TrendForce data reviewed by Tech-Insider put DRAM contract prices up as much as 89% in a single quarter of 2026, with some year-over-year comparisons running past 100%. A basic 16GB memory module that cost around $40 a year ago now runs $170–$180, with some spot transactions hitting $200.

Graphics cards are riding the same wave, because every modern GPU ships with several gigabytes of GDDR memory soldered directly onto the board. GDDR6 spot prices have roughly tripled since 2025. AMD has raised the GDDR memory kit prices it charges board partners by about 10% twice in roughly six months — most recently effective July 2026 — and MSI told investors 2026 is "the most severe year since the company was founded," raising gaming hardware prices 15–30% across its lineup as a direct result.

Why is this happening?

It isn't crypto mining or scalper bots this time — it's AI datacenters. Samsung, SK Hynix, and Micron control more than 95% of global DRAM output between them, and all three have been shifting fab capacity away from ordinary consumer DDR5 and NAND flash toward HBM (high-bandwidth memory) production for Nvidia and other AI accelerator makers. HBM sold by the rack to a datacenter buyer is simply far more profitable per wafer than gaming memory sold one module at a time — and that math doesn't change no matter how much new capacity comes online in the near term.

The scale of the demand shift is genuinely hard to overstate: industry estimates suggest OpenAI's Stargate infrastructure project alone could absorb close to 40% of global DRAM output at peak build-out, requiring roughly 900,000 wafers a month. IDC projects 2026 DRAM supply will grow just 16% year-over-year, and NAND only 17% — both well below what the market as a whole actually needs.

There's a second factor compounding the first: many GPU makers had fixed-price memory supply agreements that held costs steady through the end of 2025. Those contracts have now largely expired, and 2026's renewals are landing at dramatically higher prices — which is why the pain arrived so suddenly at the start of the year rather than building gradually.

Who's feeling it

Everyone who buys hardware, basically. IDC has warned average PC prices could rise by as much as 8% in 2026 because of the memory crunch, and some system builders have started selling pre-built desktops without RAM installed just to sidestep volatile pricing. Wccftech reported AI laptops equipped with an Nvidia RTX PRO 500 Blackwell GPU saw a $530 price increase tied directly to memory costs. Even gaming consoles aren't exempt — Nintendo raised the Switch 2's price by $50 to $499.99, effective September 1, 2026, citing market conditions widely attributed to the memory shortage. Counterpoint has separately warned PCs and phones could climb 10–20% by year's end, with the broader PC market potentially contracting more than 11%.

So when does this end?

Here's the uncomfortable answer: not this year, and probably not next year either. Analyst consensus compiled across multiple research firms points to no meaningful relief in 2026 — prices are more likely to keep climbing than to fall through the rest of the year. The steep weekly increases of late 2025 have mostly paused, but a pause at elevated prices isn't the same thing as a decline. The earliest credible window for normalization that most manufacturers and analysts point to is late 2027, when new fabrication capacity begins ramping up — and some projections stretch that timeline into 2028.

The practical reason the timeline is so long: building new DRAM and HBM fab capacity takes years, not months, and manufacturers have every financial incentive to keep prioritizing the higher-margin AI memory business over consumer DDR5 for as long as AI demand keeps growing. Until either AI datacenter demand plateaus or genuinely new fab capacity comes online at scale, consumer memory is likely to stay the residual claimant on a supply chain that's optimizing for someone else's order book.

The practical takeaway for buyers

If you don't urgently need more memory or a new GPU, waiting is reasonable — but "waiting for prices to come down" and "waiting for late 2027" may end up being the same plan. If you do need to buy, a 16–32GB machine remains far from obsolete for most everyday and even many AI-assisted workloads, since the smaller end of the model spectrum keeps getting more capable per gigabyte. Used and renewed graphics cards are also increasingly worth a look while this shortage runs its course.