In a recent demo, Lisa Su showed off a compact AI PC powered by AMD’s Ryzen AI Max+ 395. The key feature is not just speed. It is memory. Ryzen AI Halo systems support up to 128GB of unified memory, which gives the CPU and GPU access to a much larger shared memory pool than most consumer PCs usually have for AI workloads. That matters because large AI models do not just need compute. They need enough memory to actually fit and run. For developers, researchers, founders, and power users, that is why cloud GPUs, hosted AI tools, API credits, and monthly subscriptions have become such a normal part of the workflow. If hardware like this makes local AI more practical, the economics start to change. Instead of renting compute every month, you own the machine. Instead of uploading sensitive files to external servers, more work stays on your desk. And instead of dealing with usage limits, you have a local system you can keep using. Cloud AI platforms like ChatGPT, Claude, Gemini, and Cursor are not going away. For most people, they are still easier and more capable. But for local agents, private documents, coding workflows, research, and anyone juggling multiple AI subscriptions, this feels like a serious step toward a different model. The future of AI will still include massive data centers. But more of it may also run on personal computers than people expect. Source: AMD #amd #ai #artificialintelligence #technology #localai #lisasu #founder #ceo #entrepreneur #businessmindset
ADVERTISEMENT
The fact they are cousins and even look alike is crazy
This is how AI should be, not in massive datacenters
"Locally" is music to my ears.
Amd : this thing uses CPU GPU and NPU Amd to the competition: FU 😅
Cool asf, also seems better for the environment
I'd say they clearly nailed it more than Nvidia
Finally i can have my own Jarvis
1 step closer to Jarvis
That woman made me a fortune. I bought AMD stock when she became CEO and I will sell when she leaves.
"Imagine what the military has."
See that’s how you sell AI to the masses. I can see something like this in a small cluster being useful for creative work.
So just a amd version of the Nvidia spark
THIS is the point. A data center for everyone.
It’s cool, but it’s four thousand dollars. For most people, just pay for the subscription. This is only for developers, like me.
Now it just needs a napster interface and about 200 million of them running in distributed loads.
I’m getting this. Taping it to my BADLY designed cardboard Halo Mark IV armor and naming the AI Cortana…
128 Gigs of Umem doesn't run a 200B model without quantizing it to 4bit. Which significantly reduces accuracy.
Fck there goes Peta bytes of ram into the ai hole again😢
If it runs offline ima get it for sure that “locally” hit my dance move 💃
The product is: AMD Ryzen AI Halo. It's competitor is NVidia DGX Spark