Moonshots EP 185: America’s AI Plan, the End of Google Search, and the Next ChatGPT

Summary

A WTF episode with Peter Diamandis, Dave Blundin, and Alex Wissner-Gross (MIT/Harvard polymath, accelerationist) covering America’s emerging AI industrial strategy and the GPU arms race. The episode opens with AT&T’s 1993 “You Will” ads as a framing device for how badly incumbents predict who delivers the future. The core of the episode maps the compute buildout: Elon’s Colossus 2 targeting 5.5 million H100-equivalents (with a 50M target in 5 years), OpenAI crossing 1 million GPUs by year-end, and Meta building “Prometheus,” a multi-gigawatt data center in hurricane-proof tents because construction latency matters more than permanence. Alex frames the 50M GPU target as effectively a trillion-dollar AI supercluster in today’s GPU dollars. Meta’s super-intelligence team composition reveals the talent war: 50% Chinese-origin, 40% poached from OpenAI, 20% from DeepMind, each earning $10-100M+. Dave recounts Mark Chen (OpenAI head of research) turning down a billion-dollar offer from Zuckerberg, then telling him to invest more in human capital instead of just capex. Anthropic is valued at $100B with revenue surging from $3B to $4B in a single month, Claude generating $200M at 60% margins on coding alone. Alex argues Anthropic’s valuation reflects software engineering being the first major labor category to succumb to AI automation. Dave notes Dario Amodie’s unique positioning as the only frontier lab CEO who is a research pioneer from day one, focused on coding dominance specifically because it enables the self-improvement loop. The panel discusses Poly Market predictions showing Google overtaking OpenAI by year-end, and Alex makes the provocative claim that we’ve already quietly passed the Turing test and are now debating post-Turing benchmarks.

Key Segments

• [00:00-08:00] AT&T 1993 “You Will” framing, GPU arms race overview (xAI Colossus, OpenAI 1M GPUs, Meta Prometheus)
• [08:00-18:00] Data center form factors (tents, ocean, space, Tesla fleet as distributed compute), Meta Hyperion Manhattan-sized facility
• [18:00-26:00] Meta’s superintelligence team composition, Mark Chen’s billion-dollar refusal, O1 visa pipeline for AI talent
• [26:00-36:00] Poly Market predictions, post-Turing test era, Anthropic $100B valuation, self-improving code loop thesis

Notable Claims

• 50 million H100 equivalents = effectively a trillion-dollar supercluster at current GPU prices
• Meta’s AI researchers earning $10M-$100M+ annually — a preview of post-scarcity individual economics
• Anthropic revenue jumped $3B to $4B in a single month; Claude coding at $200M revenue with 60% margins
• GB200 super chips contain two Blackwell GPUs each, so public GPU counts should be halved for chip counts
• Alex Wissner-Gross: we have already passed the Turing test and barely noticed