By Hamza in Quantum Weekly — Aug 17, 2026

The SC #125

Issue #125 of The SC. Weekly supercomputing news. Microsoft goes all in on its AI chips, Nvidia gives us an LLM router, Google extends AlphaFold as an HPC product and AWS gives us K8 inspired EC2 extensions.

You’d be forgiven for thinking that if you exclude all the mainstream AI news, especially everything that’s more about financing and power grids that nothing much of note happened last week. Even my AI research bots turned up little to write about. A little old fashioned manual digging in release notes did turn up a couple of interesting stories though.

We have word from Microsoft that that they will not only be releasing an update to their own AI chips (Maia) but also rolling out a surprisingly aggressive number of them. Treat that statement with a pinch of salt but it does rather confirm the very heterogenous nature of compute. Yes, present tense. Now.

Nvidia gave us an open source LLM router which I suppose could be interesting in some domains of HPC to try and control cost where we have converged LLM inference and classical compute. I’m not sure what the workflow for that looks like though. I guess you need the routing decision before the scheduling decision else you can’t allocate resources correctly. I wonder how long it is before we see an LLM router integrated into a HPC scheduler. For LLM inference only workloads this seems like a bit of a no brainer surely?

How is it possible that even with AI genies writing code at a rate that eclipses my fat fingers on a keyboard that I still have more possible ideas than engineering capacity.

Riddle me that batman.

Talking of converged AI and HPC compute, Google gives us AlphaFold for HPC as an addition to their catalogue of offerings. Looks like an out of the box way to run AlphaFold on Google Batch to reduce your compute wall time.

AWS also gave us Kubernetes style health checks for the application layer in EC2 (see the release notes in Noteworthy). Please let this be another nail in the coffin of HPC on K8.

Lastly, we welcomed Eduardo to the team last week and I had a bit of a rant about synthetic and natural intelligence.