E165: Vision Pro: use or lose? Meta vs Snap, SaaS recovery, AI investing, rolling real estate crisis

Friedberg’s most concrete Vision Pro case was not entertainment but replacing handheld tools in his company’s greenhouse workflow. After doing the work himself, he said cameras, QR scanning, task guidance and spatial training could be folded into the headset, and his engineering team was already evaluating software for it.

Snap’s core problem was spending and dilution, not collapsing revenue. From 2021 to 2023 revenue rose $4.1B to $4.5B, but 2023 produced only $35M free cash flow against $1.3B stock compensation; Snap’s share count rose 4%, while Meta’s fell 0.5% after buybacks.

SaaS demand appears to have bottomed, but the old pricing power has not returned. Sacks cited Q4 reacceleration after six or seven weak quarters, while Calacanis described million-dollar deals shrinking to $200,000 and Friedberg said in-house engineering now creates a credible build-versus-buy alternative.

Chamath’s AI thesis is that foundation models trained on the same public data will be commoditized toward zero economic value. He expects value to accrue instead to proprietary data and high-speed inference infrastructure, especially providers controlling their own hardware rather than merely reselling Nvidia capacity.

Sacks argued OpenAI can still defend itself through a consumer-and-developer flywheel even if models commoditize. Chamath supplied the counterweight from experience: he built a useful custom GPT for Hustle but said latency, workflow integration and production reliability still made it unusable as a core tool.

YouTube may be Google’s most important AI moat because the asset compounds every day. Friedberg estimated 2,000–3,000 petabytes of multimodal data versus roughly 10 petabytes in Common Crawl, growing another 1–2 petabytes daily, making Google’s advantage potentially much harder to copy than a model architecture.

The office-property problem is being delayed, not resolved. The discussion used a $3T book-value office market versus a $1.8T estimate and described banks avoiding foreclosure through “pretend and extend,” because taking buildings back would force loan write-downs and expose losses.

Even fully occupied multifamily buildings can fail financially when cheap debt rolls off. Sacks’ example was a property earning a 6% yield but refinancing near 10%, turning leverage negative; because loans mature at different times, he described real estate as a rolling crisis rather than a single crash.