This morning, distributors MSI and Colorful raised wholesale prices on RTX 50 cards in China, and across the entire lineup, not just the flagships. According to the pricing data circulating today, the RTX 5090 is up to 16 percent more expensive, the RTX 5080 around 20 percent, and the RTX 5070 Ti up to 20 percent - with some models of that last card now starting 38 percent above Nvidia's original suggested price. The RTX 5070 wasn't spared either: up to 15 percent higher. China happens to be showing this the sharpest right now, but the trend itself has been building for months and reaches well beyond one country.
It's not just China - Nvidia and AMD are warning partners too
The prices showing up at distributors today are the last link in a longer chain. Nvidia and AMD first raise the price of "kits" (the GPU plus its allotted memory) sold to board partners - companies like MSI, Colorful, Asus, or Sapphire. A few days earlier, Nvidia notified those partners of a price increase covering the entire GeForce lineup - both Blackwell RTX 50 cards on GDDR7 and older GDDR6 models. AMD did the same to its own partners, including Asus, Sapphire, XFX, and ASRock, with some reports naming Gigabyte and PowerColor too, raising GPU-plus-GDDR kit prices by roughly 10 percent, effective July. That's the second such move in under six months - the previous one landed over the winter. Only at the end of that chain do board partners decide how much of the increase to pass on to distributors and retailers - and that's exactly the decision showing up in MSI's and Colorful's price lists today.
Where this is coming from
Memory is the culprit here, not the GPU chip itself. AI data centers need ever more DRAM and high-bandwidth memory, and their operators pay far more for it than gamers pay for ordinary GDDR. So factories keep shifting production capacity toward server-grade memory, since the margin is simply better there - and what's left over for graphics cards keeps getting more expensive.
Global graphics card prices have risen 15 percent over the past three months at that pace, and Nvidia's pricier models have taken the brunt of it - the higher the tier, the larger memory's share of the card's total cost, and the more painful the increase. The same mechanism - data centers paying more for memory than the consumer market does - is also pushing the AI industry itself to hunt for cheaper memory standards, so pressure on GDDR pricing won't ease until HBM demand does.
On top of that, a self-reinforcing pattern has appeared: some shops and distributors in China have started deliberately holding back stock, betting on even higher prices a few weeks out. Industry analysts warn that DRAM prices alone could climb another 50 percent in the third quarter of this year, then another 40 percent in the fourth - and if retailers believe that, hoarding cards instead of selling them becomes the rational move for them, even if it means even less availability for buyers.
A broken pricing ladder
Nothing illustrates this better than the RTX 5070 Ti, a card that launched at $749. Today the cheapest units in the US start around $1,000, and real street prices have spent months bouncing between $900 and over $1,250. The catch: the RTX 5080 carries a suggested price of $999 - meaning the supposedly cheaper card now costs the same as, or more than, the official price of the tier above it. The whole "pay more, get more" logic breaks down once the lower-tier model can cost more than the higher tier's own MSRP.
This isn't the first time graphics cards have slipped their manufacturers' price lists - during the crypto mining boom of 2021, shops sold RTX cards at twice MSRP or more. The difference is in the cause: back then it was demand for the cards themselves for one specific use case. Today it's the cost of one specific component - memory - that happens to be made on the same production lines as a far more profitable product for data centers.
Availability itself is starting to feel the strain too, not just price. One of Japan's larger retailers warned customers outright that shipments of high-VRAM cards might simply not return to shelves at the previous pace - some manufacturers are starting to treat memory as a rationed component rather than something you can just order more of on demand.
What this looks like from Europe
It's worth tempering the most alarmist read of this - Europe isn't seeing panic on China's scale, at least not yet. The Radeon RX 9070 GRE was losing value just weeks ago, and concrete European pricing paints a more mixed picture than a one-way climb: the Gigabyte RX 9070 XT Gaming OC now runs about €655, 3 percent above its historical low but still nearly 10 percent below its own recent average. The RX 9060 XT looks worse - Gigabyte's and Asus's models sit 14-17 percent above their lowest recorded prices, closer to the top of what the card has ever cost.
What to do about it
A new generation of cards isn't expected before mid-2027, and with DRAM price hikes already forecast for Q3 and Q4 of this year, relief probably won't arrive before late 2027 either. That changes the math for anyone planning a purchase in the coming months: waiting for a "better moment" in this particular cycle more often means paying more, not less. If the DRAM forecasts hold, a card bought today could genuinely end up cheaper than the same card bought a quarter from now.





Comments
Discussion
Join the conversation around this story.
Join the discussion
Sign in to comment and reply to other readers.
Sign inNo comments yet
Start the discussion first.