By Hamza in Quantum Weekly — Feb 23, 2026

Quantum #100

Issue #100 of the weekly HPC newsletter by HMx Labs. AWS gives us two gifts in the form of a new HPC VM type and nested virtualisation. CPUs becoming heterogeneous becomes a baked in certainty with Meta buying ARM CPUs. Microsoft becomes an Nvidia Exemplar

It seems that AWS got the news that this week’s edition is number 100 and gave us all a new AMD Turin powered HPC optimised VM type in the form of the new HPC8a. They also gave us nested virtualisation which is great news but not on AMD CPUs it seems. It’s an Intel only feature for now.

Microsoft isn’t sitting on its laurels though and is showing off their new Exemplar status with Nvidia and the installation of their new GB300s

My predictions a year ago of the CPU market becoming increasingly heterogenous got a big confirmation with Meta inking a deal with Nvidia not for GPUs but CPUs. Much is being made of this in press as being a death knell for Intel or showing Meta’s leadership as the first hyperscaler to buy Nvidia CPUs. I don’t think either is true. Intel (despite its woes) doesn’t seem to be going anywhere just yet. It’s no surprise that none of the big three cloud providers haven’t bought Nvidia CPUs either. They already have Graviton, Axion and Cobalt. They don’t need Nvidia for ARM CPUs. Meta on the other hand made a play on RISC-V but I think we’re still a little too early in that game for serous adoption. Nvidia also makes sense, as much like Isambard AI (and probably others) if you have a Grace Hopper GPU side to your supercomputer it makes sense to adopt the same CPU for the CPU only side. This is a relatively common pattern in the HPC world.

We also saw the first LLM ASIC. I’ll be honest I’m surprised it’s come this early in the game. I didn’t expect something like this till LLMs became more stable and we seem to be a long way from that! It does provide an early glimpse though, into what future supercomputers might look like. If your only GPU workload is an LLM, do you buy GPUs anymore? Will we see ASICs for LLM eventually just integrated into SoCs? Probably. But I think we’re some ways off that.

In our own news we’re hiring, and in somewhat of a first for us, this time it’s not an engineering role! Also, if you’re looking for an update on HAL our vibe coded scheduler, that’s down below too.