What the AI Boom Did to Used GPU Prices

How ChatGPT sent the used GPU market haywire — and where prices have landed in 2026

If you've tried to buy a used GPU in the last three years, you've felt it: prices that should have been falling as hardware ages have instead bounced around unpredictably, spiked without warning, and occasionally crashed back down just as fast. The culprit is well known — the explosion of interest in running AI models locally, which turned a quiet corner of the eBay market into something resembling a commodity exchange.

This is what actually happened, backed by real price data.

Before the Boom: The Quiet Market (Pre-2022)

Before November 2022, the used GPU market followed a predictable pattern. A card launched at a given MSRP, then depreciated steadily as newer hardware arrived. An RTX 3080 launched at $699 in late 2020 and would have been expected to drift toward $300–400 on the used market by 2023. A Tesla P40 — a 2017 datacenter card with no gaming use — was a niche item fetching $100–150 on eBay, bought mainly by researchers and hobbyists who knew what they were doing.

Then ChatGPT launched.

The ChatGPT Shock (Late 2022 – 2023)

OpenAI's ChatGPT went public in November 2022 and reached 100 million users in two months — the fastest-growing consumer application in history at the time. Within weeks, a wave of technically curious people discovered that you could run similar (if smaller) AI models on your own hardware, using tools like llama.cpp and the newly open-sourced LLaMA model family.

The bottleneck for running these models is simple: VRAM. A 7B parameter model needs roughly 4–8GB of VRAM; a 13B model needs 8–16GB; a 30B+ model needs 24GB or more. Suddenly, every GPU with a large VRAM buffer had a new kind of buyer.

The used GPU market responded immediately. Listings for cards with 16GB+ VRAM started disappearing faster than they appeared. Prices climbed. Sellers who'd been sitting on old datacenter stock discovered they had something people wanted.

The Datacenter Refugee: Tesla P40 and P100

The most dramatic beneficiaries of the AI boom weren't gaming cards at all — they were enterprise datacenter GPUs flooding the second-hand market as cloud companies cycled out aging hardware.

Tesla P40 (24GB GDDR5X)

The P40 was released in 2017 for AI inference in datacenters. By 2022, it was being retired in bulk from hyperscaler fleets, hitting eBay at $100–150. Then the LLM wave hit. The P40 offered something remarkable: 24GB of VRAM for under $200, at a time when a used RTX 3090 with the same 24GB cost $700+.

Prices climbed to $300–500 at the peak of buying frenzy in 2023. Today, with the initial rush settled, the P40 has come back down but hasn't returned to pre-boom levels. Current used price: $269–$314 on eBay — about double its pre-AI price, but significantly off the peak.

It remains one of the best-value options for running large models locally. See our Tesla P40 review for a full breakdown.

Tesla P100 (16GB HBM2)

The P100 is a different beast — its HBM2 memory gives it 732 GB/s of bandwidth, more than double the P40, making it significantly faster for LLM token generation despite having less VRAM. It saw similar demand pressure.

Current used price: ~$177 for the PCIe version on Amazon. However, there's a significant trap in this market.

Consumer Cards: Still Volatile in 2026

Unlike the datacenter cards, consumer RTX cards are actively tracked on Amazon with visible price history. What that data shows is striking: these aren't settled markets. They're still swinging.

RTX 3080 10GB — A $300 Swing in Weeks

The RTX 3080 launched at $699 in 2020 and is a capable card for AI workloads at 320W and 760 GB/s bandwidth — though its 10GB VRAM is limiting for larger models. Amazon's price tracking shows it ranging from $688 to $985 within a single 90-day window in early 2026. It spiked to ~$985 in early March, then crashed back to $688 by mid-March.

That's a nearly $300 swing — 43% — in a matter of weeks. The lesson: if you're buying an RTX 3080 and you see it above $800, wait.

Current price: ~$689 (at its 30-day low as of March 2026).

RTX 3080 Ti 16GB — Slowly Climbing

The 3080 Ti is the more interesting option from a VRAM standpoint — 16GB instead of 10GB, at a higher price. UK Amazon price data (a reliable proxy for global trends) shows it ranging from £1,234 in late December 2025 to ~£1,440+ by March 2026, with a slow upward drift rather than dramatic spikes.

Current UK price: ~£1,440, above the 30-day average — not the ideal moment to buy.

Current US used price: ~$700 on eBay.

RTX 3090 24GB — Still Commanding a Premium

The 3090 is the gold standard for used AI GPUs: 24GB GDDR6X at 936 GB/s, with full display output and active cooling. It launched at $1,499 in 2020 and has held its value better than almost any other GPU on the market — largely because of AI demand.

Amazon US price data shows it ranging from $1,499 to $1,729 within 90 days, currently sitting at the higher end. UK data tells a similar story: a dramatic spike to ~£1,868 around early February 2026, pulling back to ~£1,599 by March — still above average.

The 3090's used price has barely fallen since the AI boom began. Anyone expecting it to drift back toward $700 has been waiting three years. It may eventually, but the 24GB VRAM floor is holding it up.

Current US price: ~$1,500–1,730 depending on listing and timing.

RTX 3060 12GB — The Volatile Starter Card

The 3060 12GB occupies an interesting position: it's cheap enough that new buyers often reach for it first, but volatile enough that timing your purchase matters. Amazon US data shows a 90-day range of $309 to $669 — with a sharp spike to ~$550+ around early February 2026, then a crash back below $360.

The spike likely coincided with a new model release or a period of increased AI interest. The takeaway: the 3060 12GB can look cheap at $310 and expensive at $500, and both prices are real.

Current US price: ~$355, below the 90-day average — reasonable timing if you need one.

Where Prices Are Now (May 2026)

GPU	VRAM	Pre-AI est.	Peak est.	Today (used)
Tesla P40	24GB	~$100–150	~$400–500	$285
Tesla P100 PCIe	16GB HBM2	~$200–300	~$400–600	$148
RTX 3060 12GB	12GB	~$200	~$350–400	$250
RTX 3080 10GB	10GB	~$350–400	~$700–900	$689
RTX 3080 Ti	16GB	~$400–500	~$800–900	$620
RTX 3090	24GB	~$700	~$1,200–1,500	$1450
Titan V	12GB HBM2	~$600–800	~$800–1,000	$298

A few patterns stand out. The 3090 has barely corrected — AI demand has kept its floor elevated. The P100 has actually fallen below pre-boom levels in some listings, partly because supply (retired datacenter stock) has kept pace with demand. The Titan V is a strange case: once worth $3,000 new, it has 12GB HBM2 and commands collector interest, but has drifted below $350 — arguably a sleeper deal for the bandwidth alone.

What This Means If You're Buying

How to Track GPU Prices

The best strategy for buying a used GPU today is patience plus tooling:

• eBay sold listings — filter by "Sold Items" to see what cards are actually clearing, not what sellers are hoping for. The asking price and the sold price are often very different.
• Amazon price history — Amazon's built-in 30/90-day chart (visible on product pages) shows whether today's price is a spike or a dip. Buy on dips, not spikes.
• CamelCamelCamel — tracks long-term Amazon price history going back years, useful for seeing the full AI boom arc on specific ASINs.
• GPUDojo — our live price table shows current eBay, Amazon, and Newegg listings ranked by $/GB so you can compare across the whole market at once.