Supplies figures on compute supply constraints, inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → cost declines, and the gap between AI spend and measured enterprise impact. These shape cost and availability assumptions for teams building on frontier models.
Key takeaways · AI-distilled
The a16z team says rising markets have so far been supported by earnings rather than multiple expansion.
One segment finds only 2% of enterprise AI is truly tracked, part of the gap the team describes between AI deployment and measurable enterprise impact.
Another chart shows AI power users spending about 20 times the median user.
The team traces the buildout's downstream effects beyond chips into power, construction and physical infrastructure, a theme it calls a new age of atoms.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
