Moonshots EP 197: Job Loss, Chip Race, and the “AI Bubble” with Brian (Blitzy) and Emad Mostaque

Summary

Panel episode with Diamandis, Dave Blundin, Immad Mostaque, and Brian Elliott (CEO of Blitzy) — with Wissner-Gross and Ismail both absent. The episode opens with a devastating segment on higher education collapse: Americans saying college is “very important” dropped from 75% (2010) to 35%. Tuition is up 900% since 1983. Most striking: a chart shows college graduates now face longer unemployment than non-graduates — briefly becoming the most unemployable demographic. Mostaque suggests universities should invest endowments in GPU clusters since compute capacity will determine research quality. The AI wars segment covers Gemini overtaking ChatGPT in US iOS downloads (150M users), though ChatGPT leads globally. Polymarket data shows Google at 99% for best model by end of September, with Alibaba’s Qwen at 91% for second place. Grok 5 hits 15.9% on ARC-AGI v2 (highest known), with Mostaque noting benchmark saturation predicted within 3-4 years. The compute scarcity discussion is the episode’s strongest thread: XAI’s Colossus 2 targeting gigawatt-scale (110K GP200 GPUs), Elon pledging first to 1GW/10GW/100GW/1TW; Nvidia’s $100B OpenAI deal buying ~20-30% of Jensen’s annual 5M chip output; Greg Brockman saying 10B GPUs needed (agents for everyone working while they sleep); data center capacity forecast to 4x from 44GW to 156GW by 2030 (panel says this is lowball). Brian Elliott provides founder-level insight: compute costs will rise but high-value use cases (like Blitzy’s enterprise coding) can outpay consumer uses by 100-1000x, so access follows economic value per FLOP.

Key Segments

• [00:01-10:00] College education crisis — perception collapse, tuition 900% up, graduates more unemployed than non-grads
• [12:00-17:00] Gemini overtakes ChatGPT in US, Grok 5 ARC-AGI v2 record (15.9%), Qwen rising fast
• [18:00-26:00] Compute scarcity deep dive — Colossus 2, $100B Nvidia-OpenAI deal, 10B GPUs needed, data center 4x by 2030

Notable Claims

• College perceived as “very important”: dropped from 75% (2010) to 35% (2025)
• College tuition up 900% since 1983; briefly, college grads were most unemployable demographic
• Gemini overtook ChatGPT in US iOS downloads (150M users)
• Qwen at 91% on Polymarket for second-best model — six model releases in a single day
• Grok 4: 15.9% on ARC-AGI v2 (highest known score)
• Nvidia $100B OpenAI deal: ~20-30% of annual 5M chip output; Greg Brockman says need 10B GPUs
• Data center capacity: 44GW today, 156GW by 2030 (panel says lowball)
• Mostaque: AI economic labor capability went from 1-2% to approaching 50% within months

Bias/Framing Notes

Brian Elliott (Blitzy CEO) gets prominent placement — Blitzy is a sponsor and Blundin-connected company. The “AI is not a bubble” framing is unanimous without any genuine contrarian voice. College education critique is valid on data but lacks nuance around field-specific outcomes (STEM vs humanities employment rates differ dramatically). Compute scarcity framing serves the interests of anyone invested in chip/data-center companies, which the panel members are.