AMD
AI demand: still drinking from the firehose
AMD CEO Lisa Su basically told Taiwan, “Yeah, the AI boom isn’t cooling off anytime soon.” She said demand for AI and computing capacity will stay “very, very high” for years, and AMD expects to substantially increase supply again in 2027 after already expanding through 2026.
For investors, that’s the kind of message that says two things at once: demand is real, and supply is still the annoying chaperone at the party.
The bottleneck buffet
Su didn’t stop at CPUs. She said the pain points run across the whole AI stack:
- advanced wafer capacity is still tight
- high-bandwidth memory remains constrained
- advanced packaging capacity is also in short supply
Translation: even if customers are lined up and ready to buy, the industry still has to find enough pieces to assemble the AI server Lego set.
Taiwan, suppliers, and a longer planning clock
AMD is leaning harder into its Taiwan supply chain, including plans to increase the more than $10 billion investment it announced in May. Su met with Foxconn, Acer, Asus, Quanta, and later TSMC executives as AMD keeps coordinating capacity three to five years out instead of the old one- or two-year playbook.
That matters because it signals AMD isn’t treating AI demand like a sugar high. It’s treating it like a multi-year meal plan.
Big picture
AMD shares rose more than 2% in premarket trading on the comments, which makes sense: the company is basically saying the AI customer backlog still looks chunky. The catch? The same boom that’s helping AMD also keeps exposing how hard it is to make enough chips, memory, and packaging to satisfy it all.