AI companies snap up gaming cards until prices rise
AI companies are snatching up gaming GPUs and driving prices through the roof—meanwhile, NVIDIA quietly launches the RTX PRO 5500 Just as gaming graphics cards have been snapped up by AI companies and prices have skyrocketed across the board, NVIDIA did something very "low-key": it quietly listed a new card on its official website—the RTX PRO 5500 Blackwell workstation edition. No launch event, no teaser—the page just went live. But this card's specs are worth a closer look. 8
📜 Table of Contents
AI Companies Snapping Up Gaming Cards Drive Prices Up, NVIDIA Quietly Lists RTX PRO 5500
Just as gaming GPUs are being snapped up by AI companies and prices are soaring across the board, NVIDIA did something very "low-key": it quietly listed a new card on its website — the RTX PRO 5500 Blackwell workstation edition. No launch event, no teaser, the page just went live. But this card's specs are worth a closer look.
84GB of VRAM, Core Count Matches the RTX 5090
The RTX PRO 5500 packs the full GB202 big core with 21,760 CUDA cores — exactly the same scale as the consumer-grade GeForce RTX 5090. For comparison, the flagship workstation card RTX PRO 6000 has 24,064 cores and 96GB of VRAM, while the RTX PRO 5000 in the same tier has only 14,080 cores and 72GB of VRAM.
In other words, this "5500" is actually closer to the 6000 in core count, exceeding the "5000" by more than half. NVIDIA's segmentation here is interesting: it has wedged yet another tier between the 5000 and 6000.
The most eye-catching part is the VRAM: 84GB GDDR7 with ECC support, bandwidth around 1398GB/s. There's a gap compared to the RTX PRO 6000's 1792GB/s (the bus width is listed as 416-bit, though many suspect it's actually 448-bit — the official page data is inconsistent), but for running local large models, 84GB means larger parameter models, longer context, and more headroom for runtime memory.
600W Power Draw, Directly Aligned with the Flagship
Maximum board power is 600W, on par with the RTX PRO 6000, while the RTX PRO 5000 is only 300W. Power requirements match the flagship — this is no longer the behavior of a "mid-range workstation card."
The deployment form factor is also different from the past: officially positioned as a rack-mounted workstation product, offered in both dual-slot full-height active air cooling and RXM liquid cooling configurations. The thinking is clear — it's not meant for you to slot under your desk, but for IT to pack GPUs into racks for team sharing.
MIG Splits It in Two, One Card Does the Work of Two
The RTX PRO 5500 supports Multi-Instance GPU (MIG): a single card can be partitioned into two fully isolated 42GB instances, each with its own VRAM, cache, and compute cores, plus quality-of-service guarantees. One card serves two users and two workloads, directly doubling resource utilization.
The I/O and multimedia section is also well-equipped: four DisplayPort 2.1b ports (up to 8K 240Hz, 16K 60Hz), three 9th-gen NVENC encoders plus three 6th-gen NVDEC decoders — one more decoder than the RTX 5090.
What Game Is NVIDIA Playing
Put the timeline together and it becomes clear: AI companies stockpiling have pushed gaming card prices through the roof, and complaints that "gamers can't afford consumer GPUs" are growing. At this moment, NVIDIA quietly lists an 84GB, 600W professional card aimed at rack deployment, with target customers stated plainly — Agent AI, physical AI, simulation, and graphics workloads, "enabling teams to centralize, share, and scale GPU compute across the enterprise."
Translated: gaming cards are for gamers. Need real compute? Get a rack, and pay professional card prices.
For teams wanting to run large models, this card sits between the PRO 5000's 72GB and the PRO 6000's 96GB, and the price will most likely land in between too — whether it becomes the "value pick" depends on pricing. For individual gamers, just look — this card isn't meant for you.
What do you think of NVIDIA's segmentation here? If you controlled a budget, would you pick the 84GB 5500 or the 96GB 6000?
Written by our editorial team; tools listed here are tested or verified against public sources. Links point to official sites or GitHub repos for reference only — no paid placements.