AI and the Price of Things
I think I noticed it first when Nvidia's RTX 5090 video card dropped: absolute bedlam to buy what looked like a ridiculously overpriced and overspec'd video card. Large computer retail sites were crashing. People were reselling immediately for big markups. It was a Covid toilet-paper situation.
In the Netherlands, the RTX 5090 launched at around €3,100; in the US at a recommended retail price of $2,000 in January 2025. Now it's never under €4,000/$3,600, even second hand.
The completely pimped-out Mac Studio M3 Ultra with 512GB unified RAM seemed ridiculous when it came out - around $14,000/€14,600. I shoulda bought two - they now occasionally pop up on the Dutch second-hand website Marktplaats for around €20,000.
It's no secret that AI is driving up the cost. Data center demand is often blamed, and I'm sure that's a primary driver, but so is panic buying and old-fashioned greed. There's no reason second-hand computer equipment has doubled in price – it's not aged wine, people! There's also little reason that consumer-grade hardware like the RTX 5090 is selling at the price it's getting, it's not going into data centers, except that Nvidia and the resellers worked out they could increase their profits dramatically. (Yes, I understand there's demand in the fabs, but the RTX 5090 price crunch predates the data center madness.)
That said, the data-center shortages are now really starting to bite. Apple's recent increase across the board is testament to that. For a brief moment, Apple's hardware was the absolute best deal on the market, which just demonstrates how weird our current timeline is.
AI home enthusiasts are reaching back into their cupboards and digging deep into retired enterprise sales bins to find one- to two-gen-old GPUs, memory, and CPUs. My RTX 3090, which I bought at the start of Covid for about €300, would cost €1,000 to replace, but it's still the best bang-for-buck Nvidia GPU in my opinion. Old mobos with DDR4 are being dusted off as DDR5 memory prices go through the roof. There's suddenly a renewed interest in last-gen Threadripper setups, especially for multi-GPU rigs, because at least the memory is affordable. Weird enterprise GPUs are being tested for their AI chops, often not very successfully.
We're also seeing bigger gaps for other GPU makers, AMD and Intel in particular, as Nvidia's exorbitant pricing (and in my case a desire to not support the monopoly) opens gaps. My next new card is not going to be an Nvidia – I'm eyeing out the Intel Arc B70 Pros. I'd consider a good second-hand RTX 3090 (or even an RTX A6000 if I can afford it), but once again Nvidia won't be seeing any of that money.
The real fun begins when this pricing starts hitting everything else. Your next cellphone, TV, and car will have AI data center tax built in.
The whole situation leaves me with two questions: Will the market cool, perhaps with a data center crash? And should I buy now or wait?
If you've ever put off buying a house in Cape Town or the Netherlands, hoping that the prices will reverse, you'll know the answer: market cooldowns don't drop prices. At best they stabilise for a while. But the guy who bought at a premium wants to at least get what he paid, and will usually wait it out rather than taking a haircut; and the vendors have realised that us suckers will absorb the higher prices. Even if they don't ship the same quantity, the profit margins are great, so why give us peons a break?
So unless we finally storm the Californian mansions and eat Jensen Huang, prices aren't going to come down. And I admit to panic-buying an M5 Macbook Pro to beat the Applepocalypse a couple weeks ago (it went up €500 overnight), and managed to juuust get the best deal on computer hardware while the alternate timeline held.
