Okay, this is a mess, the convention floor is loud, and I’m trying to get this out before my flight. Palit just announced the RTX 3060 Infinity 2 OC, right? It’s a 2021 GPU, back on the market in 2026. This is wild. It has 12GB of GDDR6 VRAM, and that’s the whole point, that’s the big deal.

Nvidia is bringing back this five-year-old card, the RTX 3060, at its original launch price, like $329. That was the 2021 price, and now it’s 2026, and we’re seeing it again. MSI also had a Ventus RTX 3060 12GB pop up on Newegg for $329, matching that 2021 launch price.

Why? Because of the memory situation, the VRAM crisis. It’s a structural memory shortage, not just a temporary thing. AI demand is just eating up all the high-bandwidth memory, all the GDDR7, everything. Companies are prioritizing HBM production for AI accelerators, and that means less capacity for consumer-grade GDDR6 and GDDR7.

So, the RTX 3060, with its 12GB GDDR6, it’s becoming this weird stopgap. It’s not a new architecture, it’s Ampere, from 2021. But that 12GB VRAM is what people need for local AI processing, for video editing, for 3D work, and even for 1080p gaming. It’s still a popular card, actually, the second most common GPU among Steam users, according to Valve’s latest hardware survey.

This whole memory famine, it’s forcing GPU vendors to re-release older cards. It’s like, Nvidia’s newest Blackwell family uses GDDR7, and that memory is expensive, and manufacturers are busy making HBM for better margins. So, the solution is, uh, creative.

The HBM demand has outstripped supply by a huge margin. An Nvidia H100 uses 80GB of HBM3, the H200 uses 141GB of HBM3E, and the B200 needs 192GB of HBM3E. Each gigabyte of HBM takes like 3 to 4 times the wafer capacity of standard DRAM. Micron says their HBM capacity for 2025 and 2026 is fully booked. SK Hynix confirmed all their DRAM, NAND, and HBM production through 2026 is essentially sold out.

AI is expected to consume around 20% of total global DRAM production in 2026. That’s a lot. And memory costs have gone up by several hundred percent in recent months. VRAM now makes up more than 80% of the total bill of materials on some high-end cards. That’s a huge shift from a couple years ago.

It’s not just the high-end, either. Mid-range GPUs, the ones in the $300 to $800 range, are projected to see another 10-25% price increase in 2026. The 12GB and 16GB models are most exposed because those are the memory configurations AI buyers want.

And speaking of AI, the US Commerce Department confirmed on July 14, 2026, that Nvidia has started shipping H200 AI chips to China. They’re under restricted licenses, and it’s a very small number, a “trivial” quantity, according to Jeffrey Kessler, the undersecretary for industry and security. Roughly 10 Chinese companies got licenses to buy these processors. This is a big deal, considering the restrictions since 2022.

The RTX 3060 is built on the 8nm process, and Nvidia is using Samsung’s older 8nm process for it, which isn’t as heavily used for the newest AI chips. So it doesn’t compete for the same limited manufacturing bandwidth as the 4nm or 5nm nodes for the RTX 40 and 50 series.

I mean, can you believe this? A five-year-old GPU, back at its original price, and it’s considered a good deal because of the VRAM. The RTX 5060, for example, it’s newer, it’s faster, but it often has less VRAM, like 8GB GDDR7. And for AI, VRAM capacity is often the first limiting factor for running large language models locally. If the model doesn’t fit in VRAM, performance drops hard. Most users for local AI in 2026 need 12GB-16GB VRAM for comfortable use.

I bought AMD stock, ticker AMD, back on January 12, 2024, at $146.56 a share. I’m holding until it hits $600 or until the AI bubble bursts, whichever comes first. It’s currently trading around $534.39 as of July 13, 2026. (It’s a gamble, I know, but you gotta play the market you’re in, right?)

The memory shortage is expected to last through 2026 and into 2027, maybe even 2028. New fabs take years to build. So, yeah, the RTX 3060 is here to stay for a bit, it seems. It’s a sign of the times, a very strange time for hardware.