Google's AI Brain Drain, SpaceX's Huge Quarter, Airtable's 90% Collapse, US Data Fuels China AI
Jason Calacanis, David Friedberg, Brad Gerstner, David Sacks
8 Aug 20267 min read1h 45m
TL;DR
Google's AI brain drain — Jeff Dean leaving after 27 years and Demis Hassabis being moved to chair — reflects a strategic pivot toward infrastructure over frontier model development, creating a de facto duopoly between Anthropic and OpenAI at the frontier. Meanwhile SpaceX posted $7.8B in Q2 revenue, up 92% YoY, with Starlink alone potentially worth $1 trillion within 18 months on 12 million subscribers and $2.6B in quarterly adjusted EBIT. The panel argues the AI market is bifurcating: frontier intelligence commands a premium like Apple, while open-weight models commoditize everything else.
Key Moments
David Friedberg
“Capex is high alpha low beta in data center infrastructure — that capital — and model development theoretically could be high alpha but it's very high beta, it's a very risky way to deploy capital.”
Friedberg explains why Google's board is rationally shifting capex toward infrastructure and away from frontier model development.
“The latest we heard is Enthropic is now over 80 billion of ARR. Started the year at 10. It had forecast 100 billion as exit ARR for the year. And most people said that would be impossible to achieve. Now it looks like they're going to do it with a couple of months to spare.”
Sacks cites Anthropic's explosive growth as proof that frontier model companies can charge a premium unlike commodity AI providers.
“He pulled forward the $1 trillion target in ARR by a year from 2031 to 2030. Morgan Stanley's 2030 revenue estimate is 325 billion — also extraordinary — and this company did 18 billion in revenue last year.”
Gerstner contextualizes Elon's SpaceX guidance against Wall Street estimates after the company's first earnings as a public company.
All-In is a weekly podcast hosted by Jason Calacanis, David Friedberg, Chamath Palihapitiya, and David Sacks — four friends who debate tech, business, politics, and markets. The show is known for candid takes from insiders with skin in the game. It has become one of the most influential podcasts in Silicon Valley.
Takeaways
1
Anthropic's ARR growth makes prior forecasts look conservative Anthropic started 2026 at $10B ARR and was forecast to exit at $100B — a target widely dismissed as impossible. Sacks reports they're now above $80B and on track to hit $100B with months to spare, with updated estimates ranging $110–120B. This is the clearest empirical evidence that frontier model companies can sustain premium pricing even as open-source closes the gap.
2
Frontier AI is now a two-player race With Jeff Dean leaving Google and Demis Hassabis sidelined to a chairman role, the panel argues the frontier model race has effectively narrowed to Anthropic and OpenAI. Google, Microsoft, and Meta are resolving their internal channel conflicts by pivoting to infrastructure, not model-building. This creates a durable duopoly that can charge Apple-like premiums for true frontier intelligence.
3
Starlink alone could justify SpaceX's trillion-dollar valuation Friedberg calculates Starlink at 12 million subscribers growing 20% QoQ, generating $2.6B in adjusted EBIT per quarter, on a path to $30B in annual free cash flow within a year. At a 30x multiple — justified by high renewal rates and low churn — Starlink alone reaches a $1 trillion market cap within 18 months. The AI data center and Grok/Cursor businesses are effectively free upside on top of that.
4
Channel conflict is the structural weakness in Big Tech AI Google, Microsoft, and SpaceX all face the same tension: their cloud/infrastructure arms want to rent compute to external AI labs, while their internal model teams need that same compute to compete. Gerstner argues this conflict is being resolved in favor of infrastructure at every major incumbent. Companies without this conflict — Anthropic and OpenAI — have a structural advantage in staying at the frontier.
5
Enterprise AI spending will blend frontier and open-weight models Friedberg argues enterprises won't pick one model tier — they'll run cheap open-weight models for routine workflows and pay for premium frontier or specialized models (e.g., Gemini for video, life sciences) for high-stakes tasks. This means cloud providers that can host the full spectrum, like GCP, gain a bundling advantage over pure-play frontier labs. The strategic move is to be the orchestration layer, not the single-model vendor.
6
Compute infrastructure beats model R&D on ROIC Friedberg and Gerstner both argue that deploying capital into AI data centers is high-alpha, low-beta — with 30%+ ROIC cited by Morgan Stanley — while betting on frontier model development is high-alpha but extremely high-beta. The capex depreciation tax advantage under current US law makes infrastructure investment even more attractive right now. This explains why Google is rationally defunding its model teams.