Nvidia's top gaming card is now priced like a workstation part, and the AI memory rush is the reason buyers are seeing it at nearly five grand.
You wanted a new graphics card this year. So did everyone else. But the real fight isn't between gamers lining up for the same box. Tom's Hardware reported on August 5, 2026 that an Asus ROG Astral RTX 5090 BTF order placed through Nvidia's marketplace at $4,429 was canceled after a late $500 price increase, with the same card later listed on Nvidia's website at $4,929.99 before tax. Nvidia's own RTX 5090 Founders Edition still carries a $1,999 MSRP. That is not normal.
The broader market now looks almost as bad. In Tom's Hardware's August 10 check of Newegg RTX 50-series listings, the median RTX 5090 price was $4,699.99, up from $4,299.99 in June. The RTX 5070 was up 36% over the same period, and the RTX 5060 Ti 16GB was up 39%. Those are supposed to be consumer graphics cards. At these prices, they are starting to behave like rationed industrial equipment with RGB lighting attached.
Memory Is Setting The Price #
Here's the part that should worry you if you're waiting for a quick return to normal: this isn't just a launch shortage. It's memory. Samsung, SK Hynix, and Micron have been shifting production toward high-bandwidth memory used in AI accelerators, because that is where the biggest orders and margins are. TechSpot, citing IDC, reported in June that AI data centers are on track to consume 70% of global high-end DRAM output in 2026. That leaves PC makers, phone makers, console makers and graphics card partners fighting over the rest.
Gamers lose that auction.
Nvidia's reported RTX 5090 price hike turns local AI into a costlier bet Nvidia has reportedly told board partners that rising GDDR7 costs are pushing up RTX 5090 and RTX 5090D V2 pricing. The move matters beyond gaming because local AI developers and small startups rely on high-end consumer GPUs for inference and prototyping. - nvidia RTX 5090 price increase impact analysis - local AI workstation GPU cost justification challenges
The squeeze is physical. High-bandwidth memory takes more wafer area per gigabyte than ordinary consumer memory, so a move toward HBM doesn't just redirect supply, it reduces how much memory can be made for other uses from the same fab capacity. That is why a graphics card shortage can start inside an AI data center even when the card itself is sitting on a store page. The GPU isn't the only constraint. The memory around it has become the choke point.
Tom's Hardware's Korea pricing report put the pressure in plainer terms: GDDR7 modules were reportedly selling for about $20 per 2GB unit as RTX 50-series prices rose there by as much as 30%. The RTX 5090 carries 32GB of GDDR7. You don't need a spreadsheet to understand what happens when that input price moves. The bill of materials moves with it, then retailers test how much of the pain buyers will tolerate.
Hardware Makers Are Saying It Out Loud #
MSI has been blunt about how bad this is. Tom's Hardware reported in March that MSI general manager Huang Jinqing told investors the company expected gaming product prices to rise 15% to 30% in 2026 because of memory costs and Nvidia GPU supply constraints. He also described 2026 as the most severe year since MSI was founded, and pointed to a roughly 20% gap in Nvidia GPU supply. That is unusually blunt language from a company that would rather sell you a faster laptop than explain why the cheaper one may disappear.
The answer from manufacturers is already visible. MSI said it would focus more resources on mid-range and high-end products, while lower-end hardware gets squeezed. PCWorld also noted the same March comments and reported that MSI was protecting margins by leaning into more expensive hardware. If you're building a budget gaming PC, that sentence should land hard. The bottom of the market is where choice disappears first.
This is why the RTX 5090 story matters beyond one flagship card. Every AI infrastructure announcement sounds abstract when it is written as capital expenditure, data center demand or chip bookings. A $4,929.99 listing makes it concrete. The silicon that could have gone into consumer hardware is more valuable when it sits beside an AI accelerator in a server rack. So it goes there.
There is a recent precedent, but it only helps so much. Bitcoin mining strained GPU supply in 2021, and the market eased after mining economics changed and dedicated hardware took over more of the work. AI is different. AI servers and gaming PCs are pulling from overlapping memory supply chains, and the companies building model clusters are not casual buyers. They sign large contracts. They buy early. They can pay more.
You can still find cheaper cards, especially lower-VRAM models and some AMD alternatives, but the old assumption has been broken. Waiting a few months used to be a reasonable plan after a bad GPU launch. Now you are waiting on DRAM capacity, HBM allocation, Nvidia partner pricing and the appetite of AI data center buyers. That's a longer line. And right now, gamers are nowhere near the front of it.
Cadence's ChipStack AI Agent Can Now Design Chips Without Engineers Cadence Design Systems unveiled ChipStack, a Level 5 autonomous AI agent that runs chip design and verification loops with minimal human input, built with Nvidia and shown at Computex 2026. The company's Q2 FY26 revenue hit $1.584 billion, up 24.2%, with backlog at a record $8.1 billion and a raised full-year outlook of $6.26 to $6.34 billion. - AI chip design tool replaces traditional chip engineers - Cadence ChipStack autonomous agent Level 5 verification loop
Also read: CISA Gives Agencies Three Days to Patch a Critical Flaw in Ray • Beijing Opens Its Second World Humanoid Robot Games With 2,056 Machines Competing • Apple Rushes Out iOS 26.6.1 to Fix a Zero-Click Image Bug Meta Found First