The Register Kettle

When it comes to AMD's latest AI chips it's follow the money, and the memory

Episode Summary

Don't expect GPU supply, or prices, to ease until 2025 at the earliest

Episode Notes

Over the last week AMD has been extolling the virtues of its latest kit, including the MI300 which Su's crew claim is the fastest AI processing package on the market.

The market for AI accelerators is projected to reach $400 billion by 2027 and Nvidia and AMD are the two biggest ponies in the race. With a forecast $400 billion in sales up for grabs in the next four years, this could be a very lucrative duopoly unless someone else shakes up the market.

Which is going to be difficult, since it's not just a matter of building a faster processor. Memory is much more important for AI workloads and packaging the units is also a major bottleneck that suppliers are struggling to deal with. As it is almost all of 2024's chip runs from both partners have probably been reserved by customers.

The Next Platform's co-editor Timothy Prickett Morgan and The Register's own Tobias Mann were both on the ground in Santa Clara for the launches and share their take on the new technology, and how the market is going to develop. It's clear the supply shortage isn't going to clear in the next year or so, and are AMD and Nvidia in a rush to see things change? Warning: May contain sarcasm.

Moderation is provided by The Register's US editor Iain Thomson and the show, and its accompanying podcast, is by our Deputy Editor in Chief (and The Next Platform's co-founder) Nicole Hemsoth Prickett. If you'd like to catch up on earlier editions there's an archive here.