Moonshots EP 138: Elon vs. OpenAI — The Battle Over For-Profit AI w/ Salim Ismail

Summary

Diamandis and Ismail dissect the OpenAI/Elon Musk legal dispute, SoftBank’s $100B US AI investment pledge, and the broader AI capital arms race. The core narrative: OpenAI released internal correspondence showing Musk actually wanted the for-profit conversion from the beginning and sought to become CEO, positioning OpenAI as part of his Tesla/SpaceX/X empire. The breakup came when Sam Altman and the board wanted independence, not when the nonprofit structure was abandoned. Diamandis was present at xAI’s first investment pitch and provides firsthand context.

The xAI GPU cluster story is the episode’s most interesting technical content. In May 2024, Musk announced he would build the world’s largest collocated GPU cluster (100K H100s) by end of summer. Most experts said coherence across that many collocated GPUs was impossible. Musk completed it in 122 days, cornering the US helium supply for cooling. He then doubled it and raised additional tens of billions. Diamandis frames this as Musk’s “unique special power” — first-principles reasoning applied to domains where conventional wisdom says something is impossible.

The capital landscape is staggering: Altman’s quote “whether we burn $500M or $5B or $50B a year, I don’t care” sets the tone. By end of 2024, $200B+ invested across Meta, Google, Microsoft, OpenAI, and xAI combined. SoftBank’s Masayoshi Son committed $100B to US AI on stage with Trump. Oman’s investment authority took a stake in xAI. Saudi Arabia aligned with Google/a16z for hundreds of billions. Both hosts see this as companies (not governments) playing for “all the marbles” in a ~$50T addressable market (half of global GDP being labor/cognitive work).

Ismail pushes for a parallel philosophical conversation about what intelligence means and what AGI actually is, arguing billions are being spent toward a “fuzzy” goal. He advocates for a multi-stakeholder dialogue including philosophers and spiritual leaders alongside engineers. Diamandis is blunter: “you’re being naive… nobody is going to slow down.” On AI risk, Diamandis reports Musk shifted from 80/20 (good/disaster) to 90/10 over the past year. Ismail subscribes to Mo Gawdat’s thesis that truly superintelligent systems will be “abundance and life-loving” by nature.

Bias/Sponsor Notes

Standard triple ad reads (Fountain Life, Viome, OneSkin). Diamandis has personal financial relationships with Musk (attended xAI’s first pitch) and favorable access that colors the coverage. The “digital God” framing and $50T TAM figures are designed to generate excitement rather than sober analysis. No skeptical voices on AI spending sustainability, no discussion of the dot-com parallels where similar capital floods yielded massive losses. Meta’s request to block OpenAI’s for-profit conversion is mentioned but not explored as potentially self-interested. The Sam Altman quote about not caring about burn rate is presented approvingly rather than as a potential red flag.