The NVIDIA RTX PRO 5000 72GB Blackwell is Here!
I think I’ve just found the new centrepiece for my workstation which I have promptly added to my purchase wish list. NVIDIA has officially made the RTX PRO 5000 72GB Blackwell GPU generally available, and the specs are a dream for anyone building in the AI and simulation space.
Why is this such a big deal?
In the world of Agentic AI and Local LLMs, VRAM is king. 72GB of ultra-fast GDDR7.
What this would mean for my projects:
- Bigger Local Models: I can now train, fine-tune, and prototype significantly larger models directly on my desktop without hitting a memory bottleneck.
- Agentic Workflows: Running multiple AI agents and data sources simultaneously requires a massive memory footprint – 72GB makes this a reality for local workstations.
- Insane Performance: We’re looking at 2,142 AI TOPS and up to 3.5x the performance of the previous generation for generative AI tasks.
- Data Privacy: Being able to serve these massive models locally means keeping sensitive client data off the public cloud while maintaining low latency.
As someone who builds enterprise-scale AI and complex cloud architectures, seeing this level of “supercomputer power” brought to the desktop is incredibly exciting.
Read the Nvidia Blog here.
Circa £8000 for serious local AI power… is it worth it? I would say so depending on your use case. I currently run 4090’s but model splitting is only available on certain models out of the box, something like this would really change the game entirely.
Who else is eyeing this for their 2026 builds? Look at the specs not the price.
Written by
