On a recent All-In Podcast episode, Cerebras co-founder and CEO Andrew Feldman discussed how he believed investors, workers, and policymakers should weigh the implications of the arrival of advanced AI. Feldman believes that by the yardsticks the field itself set a generation ago, AGI is already here.
If Feldman and Palihapitiya are directionally right, capital is pouring into compute, energy, robotics, healthcare, and education while the ceiling on model capability remains unknown. The next major bottleneck may be humanity’s ability to deploy and coordinate the technology.
AI May Be Heavily Weighted Toward Abundance
Feldman’s central claim is that critics of AI are not doing a fair job at looking at both the Pros and the Cons. He points to medicine as the clearest line item on the positive side of the ledger, arguing, “There’s a shot that our children, none of them nor the people they love, will die of cancer. And that’s one thing that we can work on with this technology.”
Education is his second key point. Feldman notes that personalized tutoring has been proven effective for over 2,000 years, citing Aristotle tutoring Alexander the Great, yet modern schooling still teaches in a generalized way. He describes today’s classrooms as “factory farming” of education, run “the same way for 1,000 years,” and argues AI agents can finally deliver individualized instruction at population scale.
He didn’t dismiss job displacement. Feldman compares it to the end of horse-shoeing and carriage-building when cars arrived, while insisting the offsetting gains dwarf the losses. Chamath’s shorthand for the upside was blunt: “Unlimited energy, unlimited calories, unlimited knowledge, unlimited education, unlimited housing.”
By the Standards of 20 Years Ago, “AGI Is Already Here”
The second segment moved from economics to definitions. Chamath argued that AGI and superintelligence are best understood as “waypoints,” saying “we’ve hit it. We just haven’t exactly deployed it fully.” Feldman agreed on the historical yardstick: “by any definition we had 20 years ago, we’ve hit it. I mean, if you think about, oh, there’s a Turing test, blew it away.”
On recursive self-improvement, which Chamath calls “loop maxing,” Feldman said “these sort of loops are producing sort of not a little bit better answers, but vastly better answers,” and both acknowledged the ceiling is unknown. They credited Ilya Sutskever and Elon Musk as early voices who were dismissed as fringe and later vindicated.
The Next AI Bottleneck May Be People, Not Computing Power
If raw model capability is no longer the binding constraint, then what is? Feldman reframed it as an organizational question: “When are the problems no longer sort of intellectual problems and they’re now people problems?” As an example, he explained that great visions and projects today still require teams to execute them: “You spend a lot of time as a leader spraying WD-40 on your team… so friction is reduced.”
Chamath extended the point into the physical world: “I’d like a Palace of Versailles. I’ve got 100 acres somewhere out in Texas or Nevada. I’ll just send 1,000 Optimuses out there. Make me the Palace of Versailles.” While that example suggests that robots will one day do all our thinking and building, it also points to how we’ll naturally take on bigger, bolder projects that were never possible before AI.
What Investors Should Watch Next
Cerebras’s CEO Feldman’s framework gives investors a useful way to evaluate what comes next. If AI has already surpassed the benchmarks that defined AGI 20 years ago, and recursive improvement continues to produce dramatically better answers, it’s likely that raw intelligence will no longer be the binding constraint.
The challenge becomes supplying enough compute and energy, building physical systems capable of acting on AI instructions, and coordinating people and institutions around those capabilities. That points investors toward three major areas: compute infrastructure and the energy required to power it, robotics platforms that can execute instructions in the physical world, and healthcare and education software capable of turning model improvements into better human outcomes. The ceiling remains unknown, but the economic bottleneck may already be shifting from what AI can do to how quickly humanity can deploy it.
Contact [email protected] for any questions or corrections.
Top Gaining Stocks
Top Losing Stocks