Elon Enters the Chip Race, the S&P 500 Repricing, and Human Drivers Will Become Illegal | EP #242

Dave and Peter discuss Jensen Huang's framing: if a $500K engineer isn't spending $250K on tokens, they aren't using AI enough. Dave's own companies have implemented targets aiming for an 80/20 token-to-salary spend. Alex notes that tokens are the first measurable unit of cognitive productivity — unlike hours, they are introspectable and analyzable by other AI. Dave points out that using Amazon Bedrock automatically captures prompt history to S3, making it easy to start. The advice to CEOs: gather prompt strings, feed them to an AI, and ask it to evaluate which reports are least efficient.

Alex has been informally advising Nectome and urged them to publish their whole pig brain results. He compares Nectome's chemical method to 21CM's vitrification, both validated to preserve neural ultrastructure. With this evidence, he asks why adoption remains low and makes a direct plea: 'sign up for cryonics.' Peter echoes support, noting that after his previous call to action, Alcor saw a flood of inquiries. The segment positions cryonics as a rational hedge, given the accelerating progress in brain preservation.

Dave argues that with AVs and eVTOLs, location desirability will invert: currently congested urban centers lose appeal while remote beauty spots gain. He cites that 60% of land in LA is parking, which will become gardens and parks. He advises young listeners to keep renting and instead buy a 'second home' in a beautiful destination that will become more accessible; he tells his own wife to close a real estate transaction now. Peter shares that he has already converted his garage to a bedroom, anticipating not owning cars. The overall shift is from scarcity of accessible land to abundance.

Alex explained that labs repeatedly distill models, compressing giant models into tiny ones that retain most capabilities — a process he finds 'borderline magic.' Dave seizes on this to advise aspiring entrepreneurs: replay what Alex said, ask an AI to generalize, gather all related material, and craft a distilled model that outperforms general models on a narrow task. He guarantees 'instant business, instant value add, instant success.' The underlying principle is that the frontier of AI is a commodity, but niche data and application-specific tuning create defensibility.

Peter and Salim have previously advised companies to invite young entrepreneurs to pitch how they'd disrupt the firm, then fund the best ones to build an adjacent company that cannibalizes the parent. They cited IDEO as a success story. This protocol is about institutionalizing a 'burn the ships' mindset from the top.

The Terafab is a joint Tesla–xAI–SpaceX venture to produce both edge inference chips for robots and cars and radiation-hardened chips for space. The fab will be in Austin with potentially 100 million square feet of capacity. Musk's target of 1 TW/year implies 274 Starship launches per day (a launch every 5.3 minutes) to put 10 million tons of compute into orbit annually. This scale would eventually require mining the moon for materials — a pedawatt of compute would consume 3/100,000 of lunar mass, and an exawatt 3%, implying the moon is 'slated for disassembly' to build a Dyson swarm. The announcement is seen as both a response to chip supply constraints and a strategic move ahead of a potential SpaceX IPO, possibly merging all of Musk's industrial ventures into a single capital-raising entity. Panelists debated whether this creates a monopoly, but argued that such a massive mission (MTP) invites competition and that the compute abundance will benefit all.

The conversation splits into two camps. Dave B. argues that the transition to illegal human driving will happen quickly, mirroring indoor smoking bans, once sufficient data proves self-driving is e.g., 95% safer and a public ad campaign vilifies human error. He notes the number-one cause of death for children under 5 is car accidents. However, he believes the real delay will come from the shortage of chips — the technology and demand will be ready long before the supply of compute to run it. Salim adds that city centers will go first, then gradually expand. Peter imagines a car that lets you drive 99% of the time but overrides the 1% when you're about to crash. Alex pushes back hardest: he believes driving will never be banned but instead redefined as increasingly abstract 'driving' where the human gives high-level directions and the AI executes safely, preserving the feeling of control while eliminating risk. The real estate implications dominate the latter half: garages disappear (Peter already converted one), 60% of LA's land used for parking becomes parks, remote islands become accessible and skyrocket in value, and the entire car market collapses as vehicles sit idle 94% of the time and last a million miles.

Dave argues that the S&P won't collapse as a whole, but will see a massive reshuffling: companies that rest on legacy moats (insurance, oil) will die, while those that can continuously reinvent themselves will thrive. He notes that mid-cap companies outside the S&P are already at ~7x free cash flow, and AI automation will triple their free cash flow as multiples compress, creating bargains. Alex counters that free cash flow doesn't disappear — it flows to infrastructure, lunar mining, and other abundant sectors; thus, earnings multiples may simply reflect a post-moat world where capital values agility over stability. Salim adds that the EXO model needs to be rewritten because the only remaining moat is a living system that learns faster than competitors. Peter highlights that private equity is poised to use AI agents to rapidly retool cash-cow companies, turbocharging their transformation. The consensus: terminal value as a concept collapses, but capital finds new destinations, and investors must look at management's ability to iterate, not long-term cash flow visibility.

The discussion expands into distillation: Alex explains how large models generate synthetic data to train smaller, faster models that retain most capabilities — a process he calls 'borderline magic.' He envisions an ultimate distilled model of just a few million parameters. Peter asks about defensibility, and Dave asserts that data, not the model, is the moat. Alex adds that even Sam Altman's potential moats — data centers, best researchers, billions of users — are all tenuous because models walk out the door and cheap open-weight models erode distribution advantages. Therefore, value is migrating up the stack to higher-level frameworks like OpenClaw. Dave advises entrepreneurs to absorb Alex's distillation insights to build a focused model that solves a niche problem better than anyone, creating instant business.

Alex notes that while skeptics may dismiss this as easy because RISC-V has unit tests, the broader trend is undeniable: AI is beginning to redesign its own substrate. Dave emphasizes the economic unlock: because chip engineering costs are collapsing, companies can now design custom chips for every use case, achieving potentially 10x performance improvements. He points out that fabs like TSMC can handle tens of thousands of different designs without throughput loss, so the industry will shift from a few generic chips to a Cambrian explosion of optimized ones. This has hundreds of billions of dollars of implications, given the trillions being invested in data centers. The 50,000 hardware engineers will not be jobless; they will use AI to design countless variants. This is a huge abundance story.

Alex Ismail, who informally advises Nectome, nudged them to publish their results. He contrasts Nectome's chemical approach with 21CM's vitrification, both aiming to prevent ice crystal damage. The breakthrough is not just the technique but the public demonstration that an entire mammalian brain's neuronal structure can be preserved intact. This immediately raises the question: why aren't billions of people signing up for cryonics? Alex repeats his call to action: get a cryonics plan (Alcor, 21CM, or Nectome) as part of a longevity portfolio; it's the way to see the 23rd century. He discloses no financial interest. Peter echoes the sentiment.