NVIDIA has denied that the GeForce RTX 5090 is being discontinued. Jeff Fisher, the company's senior vice president overseeing consumer computing, said in an interview in Taipei on October 16 that NVIDIA will continue shipping the RTX 5090, according to Taiwan's Central News Agency. Reports had been circulating that new supply of GeForce GPUs had been halted, and the executive running the business has now stated outright that the product will continue. However, Fisher also acknowledged that supply capacity is constrained, so it remains to be seen how his remarks will affect actual restocking and retail prices.
NVIDIA executive rejects RTX 5090 end-of-life claims
What Fisher denied is the claim that the RTX 5090 has reached "end of life" (EOL), meaning production has ended. According to the Central News Agency, he described the RTX 5090 as an important product within the GeForce lineup and clearly rejected the discontinuation claims. He also stressed that the gaming business has supported the development of GeForce and CUDA, and that the technology built up there led to today's AI business.
In an interview published by 4Gamers the same day, Fisher likewise acknowledged supply constraints while saying that RTX 5090 production has not ended and that shipments will continue. He also mentioned that the GPU chip used in the RTX 5090 has several derivative products, but did not disclose how much production volume is allocated to each. His remarks add an official denial from an NVIDIA executive to discontinuation claims that had until now spread based on retailer activity and anonymous sources. Still, a policy of continuing shipments is separate from whether NVIDIA can secure enough supply to meet market demand.
Halting new GPU supply and continuing finished-card shipments can coexist
The discontinuation claims did not rest on social media rumors alone; some reporting was based on interviews with industry sources. Germany's PC Games Hardware reported on October 15, citing multiple sources around wholesalers, retailers, and OEM/ODM partners, that NVIDIA had stopped new allocations of GB202 GPU chips for GeForce. The sources are anonymous, however, and no materials directly confirming a supply halt, such as an internal NVIDIA notice, have been made public.
GB202 is the core GPU of the GeForce RTX 5090, but allocating chips to board makers is a different step from manufacturing and shipping graphics cards. Even if NVIDIA had stopped new GB202 allocations, finished cards could keep shipping as long as board makers hold GPUs they have already received or cards that have already been built. In other words, a halt in new GPU supply and continued RTX 5090 shipments are not necessarily contradictory.
At the same time, NVIDIA's statement that shipments will continue does not tell us whether GPU allocations will carry on as before or for how long supply will last. Because Fisher denied discontinuation itself, his remarks cannot simply be dismissed as an explanation of selling off remaining inventory either. Reconciling the PC Games Hardware report with NVIDIA's explanation would require clarity on which manufacturing or supply stage, and which period, each refers to.
At this point, it has not been confirmed whether new GPU allocations have actually been halted. Even if they had been, finished-card shipments could still be continuing, but it is also unclear whether that reflects NVIDIA's current operations. It is therefore inappropriate either to treat the anonymously sourced allocation-halt report as established policy or to conclude that the executive's denial resolves all supply problems.
What the RTX PRO comparison shows about enterprise GPUs
Product specifications offer some perspective on the view that NVIDIA may be steering GeForce-bound GPU supply toward enterprise products. For example, the RTX PRO 5500 Blackwell Workstation Edition carries 84GB of GDDR7 memory and is also intended for use cases where enterprises consolidate GPUs in racks. The standard GeForce RTX 5090, by contrast, comes with 32GB of GDDR7. Both belong to the same Blackwell generation but differ in intended use and memory configuration.
Based on the RTX 5090's official product page, Table 1 of NVIDIA's Blackwell architecture whitepaper, and the official RTX PRO 5500 specifications, the main differences are as follows.
| Item | GeForce RTX 5090 | RTX PRO 5500 Blackwell |
|---|---|---|
| GPU memory capacity | 32GB GDDR7 | 84GB GDDR7 |
| Rated memory bandwidth | 1,792GB/s | 1,398GB/s |
| Spec status | Standard specs | Provisional specs, subject to change |
The figures are based on the rated specifications that could be confirmed as of October 18, 2026. The RTX PRO 5500's official page says "coming soon," and its specifications may change, so they are not necessarily final. This table also compares only memory capacity and bandwidth; it does not indicate actual processing speed or how easy the products are to buy.
GPU memory capacity determines how much model and data can be held at once, while memory bandwidth indicates how much data can be transferred per unit of time. The RTX PRO 5500 has more memory than the RTX 5090, but the RTX 5090 has the higher rated memory bandwidth. For uses that keep large AI models or large volumes of data in GPU memory, 84GB is an advantage, but real-world speed also depends on compute performance, software optimization, and other factors. Larger memory alone therefore does not mean the RTX PRO 5500 is faster than the RTX 5090.
The RTX PRO 5500 also includes MIG (Multi-Instance GPU), which splits a GPU into multiple independent partitions and is described as able to create two 42GB partitions. This suits operations in which enterprises consolidate GPUs in servers and assign them to multiple users or workloads. NVIDIA is clearly giving enterprise products value different from gaming-oriented GeForce, but that alone is not evidence that it is redirecting GeForce GPU production wholesale to enterprise.
Nor can the profit NVIDIA earns per GPU chip be compared simply because the products sell at different prices. The two products differ in memory capacity and features, and details of component costs and sales terms are not public. The possibility that enterprise products are more profitable and the question of how much production capacity NVIDIA actually allocates to them need to be examined separately.
Supply capacity limits will shape RTX 5090 availability
Getting the RTX 5090 to market as a finished product requires not only the GB202 GPU but also 32GB of GDDR7 memory and other components, all assembled into a graphics card. Even if GPU production continues, constraints in procuring other parts or in manufacturing processes make it hard to increase the supply of finished products. How many cards users can actually buy also depends on which regions and sales channels the manufactured cards are allocated to.
Fisher's remarks, however, did not specify which components or processes are causing the capacity constraints. One might associate them with memory shortages amid expanding AI demand, but the HBM used in data centers and the GDDR7 in the RTX 5090 are different products. The statement that supply capacity is constrained therefore cannot by itself justify concluding that a particular memory shortage or manufacturing equipment problem is the cause. Identifying the supply bottleneck would require additional information from NVIDIA and related manufacturers.
For gamers and creators considering an RTX 5090, what matters is not only the policy of continued production but whether products are actually supplied reliably. NVIDIA's signal that it will keep prioritizing GeForce is one input for judging the product's continuity. Whether that policy translates into easier purchasing, however, will have to be judged from concrete developments such as board makers' shipment schedules, retailer stock arrivals, and delivery times.
What to watch: sustained restocking and price trends
To gauge the RTX 5090's supply situation, it is necessary to check not just whether a product page is listed but whether there is stock that can actually be ordered and how long delivery takes from order. Even if a product disappears from a retailer's page, that alone does not mean the manufacturer as a whole has ended production. Conversely, a page that remains up does not guarantee physical inventory or the next arrival.
If RTX 5090 shipments continue from multiple board makers and restocking at retailers stabilizes, it will be easier to judge that NVIDIA's policy of continued supply is being reflected in the actual market. Key indicators will be whether lead times from order to delivery shorten and whether selling prices settle. Even if restocking is confirmed, however, that would not directly prove that NVIDIA has resumed new GB202 allocations. Explanations from NVIDIA and board makers will still be needed to establish whether the reports about an allocation halt were accurate.
Fisher's remarks have clarified NVIDIA's position on the RTX 5090 discontinuation claims. But with supply capacity constrained, a product's continuation and users being able to buy it at a reasonable price are separate matters. Whether stable restocking and predictable delivery times follow will be the most important factor in judging whether RTX 5090 supply continues.