Gaming

Leaks Say NVIDIA's Next Gamer GPU Waits Until 2028, and the Data Center Bought Every Wafer

By Omar aka Crypt0 · Sep 25, 2026

Two of the most accurate leakers in semiconductors say NVIDIA's next generation of gamer graphics cards is a 2028 product. Kopite7kimi posted that the GR20x family, the consumer Rubin stack expected to become the GeForce RTX 60 series, has been pushed back, and Kepler_L2 corroborated the claim, noting that board partners like ASUS, MSI, and Gigabyte have scrubbed late 2026 and 2027 consumer tape outs from their internal milestones. These are leaks, and official word still to come will settle the timeline.

The companion rumor is just as stark. The RTX 50 Super refresh is reportedly cancelled outright. The 24GB RTX 5080 Super and 18GB RTX 5070 Ti Super were designed to fix the base 5080's much criticized 16GB framebuffer, but yields on the 24Gb GDDR7 memory they needed fell below 65 percent through the summer, and Samsung and SK Hynix redirected their cleanroom lines to HBM for AI superclusters. If both reports hold, a single architecture stretches across three years, and calendar 2027 becomes a vacuum year for flagship desktop launches.

The economics explain the silence. One analysis of NVIDIA's latest quarter puts the Data Center division at 42.6 billion dollars in revenue with margins above 85 percent, against 2.84 billion for Gaming, barely 6 percent of the total. A single wafer of advanced silicon turned into AI accelerators can generate 17 to 23 times the revenue of the same wafer cut into gaming GPUs. With hyperscalers pre booking TSMC's advanced capacity through 2027 and AMD retreating from high end RDNA 5 halo chips, gamers face a seller's market where every seller is busy serving AI data centers.

This is the first time the silicon war has a public scoreboard, and gamers lost it. The same memory crunch already lifted the Steam Deck OLED by up to 46 percent earlier this year, with Valve tying the increase directly to DRAM and NAND costs. AI data centers are now the marginal buyer of everything, from wafers to memory, and the PC hardware story of 2027 may simply be a blank page.

The leaked engineering targets for the GR202 silicon describe a monolithic die over 620 square millimeters, more than 24,500 CUDA cores, fifth generation Tensor cores with native FP4, and a 384 or 512 bit memory interface on second generation GDDR7. Prototype runs on immature leading edge nodes reportedly yielded below 38 percent functional silicon, with mask sets costing over 120 million dollars per tape out. At those economics, every wafer becomes a choice between a 1,999 dollar graphics card and tens of thousands of dollars of server hardware. The spreadsheet has already voted.

For readers, the smart money is on the used market and on patience. If the leaks hold, today's cards are tomorrow's cards, and the upgrade cycle becomes a waiting game. The silver lining is a mature, well stocked generation with deep driver support, which is a fine place to sit out a drought.

Sources

• Tekin Analysis: RTX 6090 delayed and 50 Super cancelled
• Tech Insider: Steam Deck price increase hits 46% amid DRAM crisis

New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.