IMEC Says Today’s AI Will Look Ancient in 10 Years

Memory, power and cooling — not GPU compute — are becoming AI's binding constraints, with photonics emerging as the next infrastructure bottleneck to solve.

• Memory is the acute bottleneck: Prices for AI memory chips have risen 700% this year as hyperscalers buy out inventory from Micron, SK Hynix and Samsung, who control 95% of the market.
• Moore's Law is plateauing: imec's Steven Latré says transistor scaling is nearing physical limits (~1nm, 100,000x smaller than a human hair), forcing more inventive hardware architectures.

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.