Today's protagonist is the NVIDIA RTX PRO 6000 Blackwell. It was launched at an official price of 54,500 yuan, but now it's being hyped up to 78,300 yuan. This is just such an absurd product.

Let's talk about the price first.

When this card was first released, the official price was 8,000 US dollars, equivalent to approximately 54,500 Chinese yuan.

Is it expensive? Yes. But at least there was a price.

As a result? The demand for AI has completely drained the inventory, and the official price on NVIDIA's website has now risen to $8,900 (approximately 60,600 yuan), with the entire product line out of stock.

Only the Max-Q low-power version is hanging there alone, like a model room.

What about third-party channels? It's even more shocking:

The only slightly "decent" Newegg offered a deal of $9,349 (about 63,700 yuan) and threw in a free mini PC worth $700 as part of the promotion.

It's already a steal without a discount, and they even have to give away free stuff to get people to buy it during the sale.

Graphics cards are truly the real estate of the cyber era.

What about the 5090? Do ordinary people still stand a chance?

To be honest, for us ordinary gamers, the RTX 5090 has actually become a "cost-effective choice" this time around.

The official starting price of the consumer flagship RTX 5090 is 4,000 US dollars (approximately 27,200 yuan). Although the current third-party asking price is also over 6,000 US dollars (approximately 40,800 yuan), there is at least stock available.

Why? Because the price is too high and directly dissuaded a large number of ordinary enthusiasts.

But for AI developers, the 5090, at half the price of a workstation graphics card, has instead become a hot commodity.

For those AI teams with actual needs, the RTX PRO 6000 Blackwell is their top choice.

The reason is simple. Let's look at the specifications first:

• Core count: 24,064, which is 10.5% more than the 5090

• Tensor cores: 752

• RT Cores: 188

• FP32 floating-point performance: 125 TFLOPS

• AI Computing Power: 4000 TOPS

• VRAM: 96GB GDDR7

• Memory bandwidth: 512-bit

• Bandwidth: 1.8TB/s

When they saw the 96GB VRAM, AI developers' eyes lit up.

But what does this mean?

A single card can run a 230-billion-parameter large model with performance comparable to that of four RTX 5090 cards.

Four cards are only equal to the workload of one card, and the power consumption is only one-fourth.

With a rated power of 600W, it can be powered by a single 12V-2x6 power cable, and NVIDIA has also specially redesigned the dual-fan dual-slot heatsink.

Simply put: One card does the work of four and is even more power-efficient.

Here's another piece of bad news

Don't think it will pass after a while.

Industry reports have already stated: Due to the continued tight supply chain, the prices of GPUs and other PC components will continue to rise in 2026.

So, if you're planning to buy a graphics card, grab one while you can—don't hold out for a price drop.

With this wave of AI, graphics cards will only get more and more expensive, not cheaper and cheaper.

Unless you don't use AI, don't do development, and don't touch large models.

Then congratulations, you only need to worry about when the 5090 will be snapped up by AI developers.