Used graphics cards in 2026: the memory crisis changes everything

The ritual has turned into a chore. You settle on the model you want, you open three tabs… and the price sails well past the recommended figure announced at launch. Tight stocks, “good value” models nowhere to be found, increases arriving in waves. Buying new has never felt so thankless. To understand why — and above all what to do about it — you have to go back to the cause. And it has almost nothing to do with graphics cards themselves.

At the root: a memory crisis, not a simple GPU shortage

The real culprit is memory. Since the autumn of 2025, the industry has been living through a global DRAM shortage, accelerated in 2026 by the insatiable appetite of artificial intelligence and data centres. Memory chip makers are redirecting capacity towards the high-performance modules destined for AI, which are far more profitable — at the expense of everything else.

Here is the point genuinely worth remembering, because it is so often overlooked: the GDDR fitted to graphics cards and the DDR5 that populates your RAM sticks come out of the same factories, through the same bottleneck. When one becomes scarce, the other climbs too. This is not an isolated quirk of the graphics card market: the entire memory ecosystem is contracting at once. Your next card and your next memory kit are going up together, for the same reason.

The domino effect: reduced output, SUPER cards cancelled, no relief for two years

The consequences follow mechanically. Short of GDDR7 in sufficient quantity, NVIDIA is preparing to cut its RTX 50 gaming GPU output by something like 30 to 40 % over 2026, with priority going to higher-margin AI chips. First collateral victim: the RTX 50 SUPER cards, the revisions expected to offer more memory at the same price — cancelled or postponed indefinitely. As for the real successor, the next generation (NVIDIA's Rubin architecture, sometimes referred to as RTX 60), it is not expected before late 2027, possibly 2028.

On pricing, AMD already raised its prices in the first quarter of 2026, by roughly +10 to +15 %, with NVIDIA following suit. The message for buyers is unambiguous: the new market is frozen at a high plateau, and no relief is in sight for about two years.

Given the choice, new or used, AMD often keeps a better performance-per-euro ratio than NVIDIA, particularly in the mid-range. One nuance deserves stating, though: as soon as you step outside pure gaming — streaming, video editing, creative or professional applications (3D rendering, compute, AI) — NVIDIA retains an edge, thanks to its encoders (NVENC) and to a better-supported software ecosystem (CUDA). The right call therefore depends as much on your use as on the price tag.

The tipping point: why used has become the rational choice

In this context, used cards change status outright. As long as new remained affordable, a second-hand card was a small-budget plan B, a stopgap. In 2026, the arithmetic is reversed. The previous generations — RTX 30 and 40 at NVIDIA, RX 6000 and 7000 at AMD — have lost none of their usefulness: they run games and simulators perfectly well, often at 1440p without breaking a sweat. And since new prices are no longer falling, they have all but stopped depreciating too.

The figures from our index bear this out: a used RTX 3080 trades around a 350 € median, an RTX 4070 around 440 €, an RX 9070 XT (current generation) around 600 €. At the other end of the spectrum, the very high end moves faster — the RTX 4090 sits around 1,925 € but has fallen clearly over two months. In short, for anyone building or upgrading a machine — including a sim racing build where the GPU accounts for 40 to 50 % of the budget — used is no longer a spare wheel, it is a clear-headed choice. On one condition: knowing what you are buying, and at what price.

The real trap of buying used: what is that price actually worth?

And this is where it all gets complicated. On the classifieds, the same model appears at prices that go off in every direction. Take the RTX 3070, one of the best-selling cards on the used market: our index puts it around a 250 € median, yet you will find them at 170 € and at 230 €. So is a 3070 listed at 200 € a bargain or a trap?

Without a landmark, there is no deciding. You hesitate, you fear overpaying — or, when selling, giving away a card that was worth more. And that fog mainly benefits the listings that are too good to be true. One thing is missing, only one: a reliable reference price.

The GPU Price Index: a reference price at last

Answering that question is exactly why we built the GPU Price Index. The idea fits in a sentence: give every model the real used-market price, computed from real listings. Not a catalogue rate, not a finger-in-the-air estimate, but the median price collected from Leboncoin listings, over a rolling window, model by model. We keep the median (the price at the centre of the listings) rather than an average, precisely so as to withstand extremes and bundles, and cards advertised as faulty are excluded from the calculation so they cannot distort the landmark.

In practice this breaks down into three uses. Buying? The GPU Price Index gives you the reference price per model and helps you set your budget. Selling? The resale estimator gives you a realistic range for your card, rather than aiming at random. Hunting for a bargain? The deals page surfaces the listings that break away clearly below the index.

For each model, a “deal” threshold marks the price below which a listing genuinely stands out: around 200 € for the RTX 3070, around 300 € for the RTX 3080. Take our example again: a 3070 at 200 € is therefore at the bottom of the range, right under the threshold. Interesting — provided the condition matches.

Buying without getting caught out

Because a good price alone does not make a good deal. A few simple reflexes head off most nasty surprises:

• Ex-mining cards: run continuously for months, with tired fans and thermal paste. Not a deal-breaker if the price reflects it and the seller is upfront, but worth knowing.
• “For parts” or “no display”: that is a dead card in disguise. Avoid, unless you repair cards yourself and know what you are taking on.
• An abnormally low price on a sought-after model: a massive discount often hides a fault… or a scam (fake profile, payment off-platform). A 3070 at 120 € when the index sits at 249 € is not a miracle, it is a signal.
• Warranty and receipt: a card still under manufacturer warranty is worth more, and the receipt protects you. Failing that, negotiate accordingly.
• Proof, not promises: insist on real photographs of the card, ideally a screenshot of the GPU recognised and stable under load (a game or stress test).

The good news: the index already sets aside a share of those dubious listings — models flagged as faulty, aberrant prices — before computing its landmarks, which cuts the noise. But the last filter is you.

In 2026, used is no longer a compromise, it is a strategy

The situation has changed. The memory crisis has locked the new market down for a good while, and for as long as it lasts, used will not be a fallback: it will be the smartest choice. Buying a previous-generation card at the right price is currently the best enjoyment-per-euro for gaming as much as for sim racing. And for a sim racer, the money saved on the GPU is not just a figure: the few hundred euros kept by buying at the right price are that much budget to shift where it really counts for the feel — a better direct drive base, or a load-cell pedal set. The whole thing comes down to one word: information. With a reliable reference price in front of you, you stop gambling and you buy — or sell — at the right price. That is precisely what the GPU Price Index puts at your disposal.

In short

ℹ️ Index figures collected on 24 June 2026 (rolling medians) — they change over time; see the GPU Price Index for up-to-date values. Market facts: Tom's Hardware FR, Developpez.