By Hamza in Quantum Weekly — Jul 6, 2026

The SC #119

Issue #119 of The SC. Weekly supercomputing news. Meta is going to start renting its GPUs. And possibly we don’t need GPUs anymore too. Either way memory prices are so crazy even Apple is putting prices up but maybe South Korea will fix that?

You’d have to have your head buried pretty deep in the sand to have not heard the news that Meta is contemplating becoming the newest neocloud. Are they young enough to be a neocloud? Or do they just get to be a regular cloud selling a higher proportion of GPU compute? 🤔

That’s not the interesting bit. The interesting bit is watching for what happens next. We’ve already seen SpaceXAI (whatever it’s called this week) decided to resell their compute capacity to Anthropic and Google. Those deals happened pretty quickly, holding up the narrative that since approximately the end of 2025 there has been a strong demand (and increasing prices) for GPU compute. How quickly, and to whom, any Meta’s spare capacity gets sold may well prove to be a bell weather for how deep the demand for GPU compute (and by association, AI) really is. All against a backdrop of reduced tokenmaxxing and increased adoption of open weight models.

It goes without saying, I really don’t envy anyone having to make the decision on new hardware (especially GPU) purchases right now. At least in terms of trying to guess which way prices are going to go (even if the short-term trajectory is still upwards). Just to make it even harder though, there’s a very valid and interesting question around how much longer we will even need GPUs.

We have a few HPC heavyweights talking about a pattern that plays out over and over again in computing and specifically with CPUs. As a type of compute becomes more prevalent, the ability to perform it (in hardware) on the CPU follows. We saw this with floating point math, media encoding and cryptography, it’s only natural that mixed precision vector math follows right? We can hope. And it’s not like there isn’t prior art in the supercomputing world. I’m looking at you Fugaku and LineShine.

Right now, the demand for silicon seems so insatiable that even the master of supply chain economics, Apple, has put up prices. South Korea is doubling down with large investments into new plants so we can hope that eases supply issues. In about 5 years.

The real constraint to datacentre growth though has always been the utilities it is plumbed in and hooked up to. Interesting then, to see the impact its having on consolidation in the utility markets. You think something looks mature and highly optimized and then the demand increases tenfold overnight, I guess that wakes people up.

Lastly (and most importantly 😉), its HPC Club on Wednesday 🎉. I think we have 2 places left which is pretty unusual so close to the event. See you there!

All the News in Depth

Cloud and vendor releases for Supercomputing & HPC is a bit delayed this week. Our application for tracking seems to have run into a bug. That’ll teach me to vibe code it. Back soon!

Meta to become the latest neo cloud or the in depth Next Platform take

Of course we may not need GPUs too much longer

CPUs or GPUs, assuming we keep demanding the compute perhaps South Korea will be able to produce it or at least the memory that goes with it

Its really not secret that datacentre growth is constrained by (and has been forever) utilities. Interesting to see them starting to consolidate even further though