Can’t Get an RTX 5090? Here’s What to Build Instead

The RTX 5090 is becoming increasingly difficult (and expensive) to get hold of. But that doesn’t mean your next GPU build needs to be put on hold. From NVIDIA’s RTX PRO Blackwell range to AMD alternatives and multi-GPU setups, there are plenty of ways to keep your project moving.

13 November 2026

Your project can keep moving

UK prices for the RTX 5090 have passed £6,000 against £1,799 from a year ago, and now NVIDIA has stated it’s going end of life altogether. AI data centres are buying up the memory chips graphics cards use, and Nvidia is sending the 5090’s processor to its professional RTX PRO cards first (Tech Insider). Relief looks unlikely before mid 2027 (TechTimes). Strong alternatives are on sale today, and many suit professional work better.

Three questions that pick the right card

What will it run? AI, rendering, video and live graphics each stress a card differently.

How much memory does it need? Graphics memory (VRAM) is the card’s workbench. When the job overflows the bench, the card slows to a crawl or refuses to run it. The 5090 has 32GB, and for AI work memory size often matters more than speed.

Where will it live? A 5090 draws up to 575W. It sits happily in a tower under a desk and struggles in a warm rack.

Option 1: Step up to a pro card with more memory

Nvidia’s RTX PRO Blackwell cards are where its supply is going. They cost more up front and are built for the work most professional buyers want a 5090 for. The RTX PRO 6000 (96GB) uses the same top chip with three times the memory, so large AI models run whole on one card. The RTX PRO 5000 (48GB or 72GB) has less raw power and far more memory at around 300W, so it finishes jobs too big for a 5090 (SpeedyDrone). The RTX PRO 4500 (32GB) matches the 5090’s memory at 200W, ideal for editing, grading and design. All three add error-correcting memory, certified drivers and, on most models, blower coolers that suit racks.

Option 2: Buy a card you can get hold of today

The RTX 5080 (16GB) is very fast for editing and real-time graphics, and far easier to buy, though large AI models need more memory. The AMD Radeon AI PRO R9700 (32GB) matches the 5090’s memory at 300W for well under half its street price (AI Indigo). Check your software supports AMD first, as most AI tools are built for Nvidia. For business systems that need a full warranty, we steer customers towards current cards over second-hand 4090s.

Should you wait for the RTX 5080 Super?

It is still a rumour, unconfirmed by Nvidia (Shattered). Leaks point to 24GB of memory, the same processor as today’s 5080, around 415W and an early 2027 launch (Club386). It has been tipped and pulled at least three times since 2025, held back by the same memory shortage. If it arrives, it makes a useful middle ground. Plan around it once it is on sale.

Option 3: Split the work across two smaller cards

I like to think of it as a render farm in a single box. Each card takes its share of the frames, and the job finishes together. Rendering, encoding and running several AI models all split neatly across cards. Two RTX PRO 4500s give 64GB at 400W combined, less power than one 5090. More cards need the right lanes, power and airflow, and in a rack, cooling decides everything, so the card and the chassis get chosen together.

The options side by side

Card	Memory	Max power	Best for	Watch out for
RTX 5090	32GB	575W	Peak speed in a desktop tower	Supply, price, heat in a rack
RTX PRO 6000	96GB	600W (300W Max-Q)	Large AI models, huge scenes	Highest cost
RTX PRO 5000	48GB or 72GB	300W	AI and rendering that need memory	Slower than a 5090 on raw speed
RTX PRO 4500	32GB	200W	Editing, grading, design, multi-card builds	Low peak performance
RTX 5080	16GB	360W	Editing and real-time graphics	Memory too small for big AI work
Radeon AI PRO R9700	32GB	300W	Budget 32GB builds	Check software supports AMD

Power figures are the manufacturers’ maximum board power.

Read our NVIDIA professional GPU comparison for more information on the PRO card line-up.

How G2 builds a system you can trust

A GPU purchase usually needs sign-off from the technical lead, finance and whoever runs the rack. A 5090 picked on headline speed can keep one of those people happy and hand the other two a problem. We start with the workload and explain the reasoning behind every part.

We design and build rackmount and workstation systems in Hampshire, and you talk directly to the engineers who build yours. Our 2U GPU server fits two RTX 5090s or four RTX PRO 6000s, so it can follow whichever cards supply allows. Every system carries a G2 warranty, and we will tell you plainly when the best answer is to wait.

Front view of a 2U computer with no lid, displaying internals

Prove it on one system before you scale

Tell us what the system needs to do and we will spec one build around it. Run your real jobs on it, share the results, and roll out once the numbers back the choice.

Spec a single system first. Let the results make the case for the rest.