AI Bubble Pops, Zuck Freezes Hiring, Newsom’s 2028 Surge, Russia/Ukraine Endgame

Watch on YouTube ↗ | August 22, 2025 at 19:16 | 1:11:45 | All-In Podcast

Speakers

David Sacks — General Partner, Craft Ventures

David Friedberg — CEO, The Production Board

Jason Calacanis — Angel Investor / Founder, LAUNCH

Chamath Palihapitiya — CEO, Social Capital

Summary

The besties debate whether the sharp pullback in AI stocks is a bubble popping or a healthy correction, using the viral MIT study showing 95% of enterprise generative AI pilots failing, Sam Altman's bubble comments and Meta's AI hiring freeze as the evidence. Sacks argues the deflating narratives are the AGI and rapid-takeoff ones, not the technology itself, and that value will accrue to vertical applications and specialized models; Friedberg frames the capex wave as dotcom fiber and expects small specialized models to transform the unit economics; Chamath expects a long sorting cycle, enterprise resistance and a bull case that still supports OpenAI's valuation. The second half turns to politics, with Newsom leading early 2028 Democratic polling and a debate over California's record, then to Trump's Alaska summit with Putin and what a Russia-Ukraine endgame looks like.

An MIT study found 95% of enterprise generative AI pilots never reach production, with back-office automation showing far better ROI than sales and marketing tools.
David Sacks calls the roughly 10% drawdown in public AI stocks a healthy sentiment correction and says the investment super cycle is still early to mid-stage.
Model performance is clustering and labs keep leapfrogging each other, which the group reads as evidence against recursive self-improvement and near-term AGI.
David Friedberg compares AI capex to dotcom fiber: ROIC is unproven today, but a shift to networks of small specialized models could cut cost per token 10x to 100x.
Chamath Palihapitiya expects a sorting and churn phase among AI application startups and heavy internal resistance to AI inside large enterprises.
Meta's AI hiring freeze is read as digestion after aggressive aqua-hires rather than as the end of the AI boom.
Chamath lays out a bull case for OpenAI at a 500 billion dollar valuation based on DAU growth and a fraction of Facebook-level ARPU.
Politics and geopolitics: Newsom leads early 2028 Democratic polling, Sacks attacks California's fiscal and crime record, and the group debates Trump's Alaska summit and the Russia-Ukraine endgame.

Ideas

David Sacks General Partner, Craft Ventures 13:40

Healthy AI correction, super cycle intact.

Sacks argues the week's sell-off, roughly a 10% correction in public AI stocks after the viral MIT pilot study and Sam Altman's bubble comments, is a healthy correction in sentiment rather than the start of a bust cycle. What is being deflated are the fantastical narratives: AGI in two to three years, recursive self-improvement, rapid takeoff and 50% of knowledge workers losing their jobs. His evidence against takeoff is that model performance is clustering rather than separating, the labs keep leapfrogging each other, models are specializing instead of one becoming all-powerful, and GPT-5 landed as incremental progress against lofty expectations. Because this is a normal technology race, he says normal investment and policy logic applies, and he repeats later that this is not the bust part of the cycle and that we are still early to the middle of an AI investment super cycle.

David Friedberg CEO, The Production Board 21:20

Cheaper model architectures rescue AI capex.

Friedberg compares the current AI capex wave to dotcom-era fiber: connectivity was deployed faster than end users created economic value, the bubble burst, but the ultimate promise was realized orders of magnitude beyond expectations, just delayed by a few years. Whether this particular capex cycle, mostly raised by model companies and deployed into chips and data centers, earns its ROIC is still TBD. He points to three trends reshaping the economics: pairing human engineering with generative tools, pairing generative models with deterministic systems such as existing rendering engines, and a shift from one large model to networks of small specialized models. That SLM re-architecture dramatically lowers energy and dollar cost per token, so if it works the systems become 10x to 100x more efficient, ROIC on the buildout improves from challenged to much higher, and total token production goes up rather than down.

David Sacks General Partner, Craft Ventures 36:07

Vertical AI apps capture the value.

Sacks reads the enterprise survey result as evidence that AI value accrues to vertical applications and smaller specialized models rather than to one general foundation model. Generalized model deployments failed about 95% of the time because LLMs need enterprise context, detailed prompting, hallucination validation and iteration, which he calls the last-mile problems, while vertical applications, vertical models and SLM approaches with a tighter problem and data set showed much greater success. He argues that going from 90% to 99% effectiveness requires industry-specific knowledge, so the outcome is lots of vertical applications and specialized models capturing value across many separate markets instead of a single foundation model eating all the value, which he considers healthy for the ecosystem.

Jason Calacanis Angel Investor / Founder, LAUNCH 38:14

Tesla robotaxi still trails Waymo's reliability.

Calacanis places self-driving just past the peak of inflated expectations and argues the edge cases are more serious than people assumed, so deployment has to be slow and thoughtful. He contrasts Waymo, which runs a more deterministic system with no safety drivers, against Elon Musk's robotaxi program, which still carries safety drivers and uses a probabilistic model that needs an intervention every few hundred miles, which for a daily user means an intervention roughly every week or two. He expects the probabilistic approach to iterate down to Waymo's level eventually, but says the timetable for that crossover is the unanswered question, making robotaxi timing a setup to monitor rather than a settled outcome.

Up Next

Jackson Hole Preview: Will Fed Tank Markets This Week? | Chance Finucane

The David Lin Report · Aug 21

Senator Eric Schmitt: Exposing the Biggest Censorship Scandal in US History

All-In Podcast · Aug 21

How Close Are Banks To Another 2008? Expert Reveals 'Pendulum Swing' | Christopher Wolfe

The David Lin Report · Aug 20

6 Million Jobs At Risk: The Biggest Threats To Labor Market Warns Economist | Heidi Shierholz

The David Lin Report · Aug 19

One More Push Higher Before the Crash?

1000x Podcast · Aug 18

This All-In Podcast video, published August 22, 2025, features David Sacks, David Friedberg, Jason Calacanis discussing AIQ, Vertical AI software, TSLA. 4 trade ideas extracted by AI with direction and confidence scoring.

Speakers: David Sacks, David Friedberg, Jason Calacanis · Tickers: AIQ, Vertical AI software, TSLA