🤠 LIVE from TX with TK, Jon Gray & Scott Wu
Live from Tomorrow X in Austin, guests argue the AI buildout is demand-led and supply-constrained, enterprises adopt agents in curated stages, and persistent agents are the next step for software work.
Watch on YouTube ↗Charlie's summary
- ▶ 4:34
The AI buildout differs from the railroad and fiber booms because demand is outrunning supply and is backed by long-term contracts
Jon says 19th-century railroads and 1990s fiber were built speculatively and with leverage, in the hope that demand would follow. Today he sees the reverse: hyperscalers with average market caps of $2.5 trillion and little net debt sign long-term contracts first. His firm's companies have raised model spending 20-fold in a year. Chips, power and capital are scarce, which he thinks may keep the buildout healthier.
- ▶ 10:10
Companies are still underusing AI, so Jon encourages spending and expects cost discipline later
Asked about huge token bills, Jon says adoption is still far closer to underuse than excess. Rental applications process 90% faster, and Ancestry processes family history 30 times faster. Others compared token spend to marketing budgets, where some waste is accepted while you reallocate. Jon grants that there may come a point of “driving a Ferrari to pick up the groceries”, with more attention to which model is used.
- ▶ 13:45
AI will multiply therapies, but the physical trial funnel stays the bottleneck, and that is where capital matters
Jon expects AI to produce far more candidate therapies through synthetic chemistry and biology. But trials involve people, the FDA and global rollout, and phase three trials cost billions. His firm sees capital as the constraint there and invests in it, which he calls a way to be a “real force for good”.
- ▶ 25:57
AI is halving chip design time at Micron, with memory now co-designed with customers
The Micron executive says AI use is most developed in engineering, with a target of finishing designs in half the time and doing twice as many designs with the same people. It also feeds design verification and simulation of process, thermals and chip physics. Memory used to be a commodity, but Micron now co-designs with customers years ahead and has raised R&D spending to say yes more often.
Give your email to unlock the rest of the key points — for every episode.
Do more with Charlie
Free — sign in with your email, no password.