Why GPUs Are Aging Like Milk (The 'New God' Trap)

I think GPUs are aging like milk right now. Every year, there’s a “new god GPU” on the market that makes the previous one feel instantly outdated. We’ve already seen this: first the H100, then the H200, and now the Blackwell B200, each bringing insane performance jumps.

In my view, buying a new flagship GPU is often a waste of money. Let me tell you why: You buy a flagship card, and 6–12 months later, the resale value is trash because new software demands even more VRAM and TFLOPS again.

The situation in data centers is even worse. Companies are pouring hundreds of billions into GPUs that run at crazy power levels all day. A significant chunk of them either die or become “old” within just 1–2 years. That is a lot of silicon going straight to the graveyard.

For normal people like us, the lesson is simple: don’t treat your laptop or PC like a mini data center. If you try to run AI workloads 24/7, you will cook your hardware in 1–2 years. Instead, use cloud GPUs when you really need to do heavy training or intensive AI work. Plan for 2‑year GPU cycles or just rent the computing power instead of chasing every new card.