NVIDIA seems set to shake up the GPU market by adapting gaming graphics cards for artificial intelligence workloads. According to insider reports, prototype versions of the GeForce RTX 4090 with a staggering 96GB of VRAM are already being tested in the company’s labs, with mass production potentially starting in just three to four months.
Leaks from social media suggest that NVIDIA engineers are experimenting with quadrupling the VRAM capacity of the standard RTX 4090 (24GB). Achieving 96GB required the use of exclusive 4GB GDDR6X modules, which are not yet available on the market. This raises questions about the stability and cost of the upgraded model—preliminary estimates suggest its price could be twice that of the rumored 48GB variant.
Analysts speculate that NVIDIA’s move is driven by export restrictions. With China facing a shortage of specialized AI accelerators, the country could become the primary market for these upgraded RTX 4090s. Localized production of such cards would allow Chinese companies to train neural networks despite trade limitations.
Experts note that 96GB of VRAM is excessive for gaming, even in 4K resolution. However, for rendering, large language model (LLM) workloads, and big data processing, this capacity could provide a significant advantage. This effectively transforms the RTX 4090 into a hybrid solution, potentially serving as a more accessible alternative to professional GPUs like the A100 for startups and research institutions.
NVIDIA has yet to make an official statement, but the industry is already holding its breath — will the green giant once again redefine the rules of the game? The answer should become clear by the end of 2025.