Google's AI Brain Drain, SpaceX's Huge Quarter, Airtable's 90% Collapse, US Data Fuels China AI
The All In podcast discusses Google's AI brain drain with Jeff Dean leaving to start Discovery Loop, SpaceX's spectacular earnings with $7.8B revenue and $100B ARR guidance, and concerns about US data being sold to Chinese AI companies through data labeling startups.
Summary
The episode opens with discussion of major personnel changes at Google's AI division. Demis Hassabis moved to chair of DeepMind while several top AI researchers, including Jeff Dean (a 27-year Google veteran and employee #30), are leaving to start Discovery Loop focused on deep scientific breakthroughs. Google's Gemini 3.5 Pro is reportedly months behind schedule due to low morale. The hosts analyze this as rational capital allocation: Google is investing heavily in AI compute infrastructure ($200B CapEx) which offers better returns (high alpha, low beta) than frontier model development (high beta, risky). This explains why talent is departing—scientists interested in frontier models face channel conflict with Google's infrastructure-focused strategy. The duopoly of Anthropic and OpenAI emerges as the clear frontier model winners, with Anthropic hitting $80B+ ARR and growing 10x year-over-year. SpaceX reports exceptional Q2 earnings with $7.8B revenue (92% YoY growth) and announces $100B ARR guidance by year-end, with AI revenue from Colossus compute rental to Anthropic and Google more than tripling to $2.6B. CapEx surged to $18.4B quarterly (6x YoY), putting SpaceX on a $75B annual run rate. The hosts debate SpaceX's 13% post-IPO decline despite strong results, attributing it to market concerns about compute financing and willingness-to-pay. Starlink emerges as a massive cash machine generating $2.6B adjusted EBITDA on 12M subscribers, with potential to become a trillion-dollar business alone. The discussion covers Starship's importance for deploying V3 satellites with 10x bandwidth, enabling direct-to-cellular services. Airtable's sale to Bending Spoons for $1.28B (90% off its $11.7B peak valuation) illustrates SaaS valuations normalizing. The hosts analyze how Airtable's forced sales-led motion (contrary to its natural PLG model) only achieved 30% attainment, creating opportunity for a PE buyer to return to profitability by eliminating 80-90% of costs. The final segment addresses US data labeling companies (Surge AI, Mercor) selling training datasets to Chinese AI labs alongside US frontier labs. The hosts debate whether this represents meaningful technology transfer or inevitable data commoditization, with disagreement on whether restrictions should apply given China's ability to generate similar datasets independently.
About this episode
<p>(0:00) Bestie intros! Brad Gerstner fills in for Chamath</p> <p>(2:16) Major shakeups at Google: AI brain drain or better strategy?</p> <p>(20:39) SpaceX's big quarter: Terafab, AI Capex, $1T revenue projection?</p> <p>(45:44) All-In Summit Speaker Announcements!</p> <p>(48:01) Airtable sells for a 90% discount: SaaSpocalypse?</p> <p>(1:05:56) Chinese AI labs are buying US training data to catch up</p> <p>Apply for Summit 2026:</p> <p><a href="https://allin.com/events">https://allin.com/events</a></p> <p>Follow Brad:</p> <p><a href="https://x.com/altcap">https://x.com/altcap</a></p> <p>Follow the besties:</p> <p><a href="https://x.com/chamath">https://x.com/chamath</a></p> <p><a href="https://x.com/Jason">https://x.com/Jason</a></p> <p><a href="https://x.com/DavidSacks">https://x.com/DavidSacks</a></p> <p><a href="https://x.com/friedberg">https://x.com/friedberg</a></p> <p>Follow on X:</p> <p><a href="https://x.com/theallinpod">https://x.com/theallinpod</a></p> <p>Follow on Instagram:</p> <p><a href="https://www.instagram.com/theallinpod">https://www.instagram.com/theallinpod</a></p> <p>Follow on TikTok:</p> <p><a href="https://www.tiktok.com/@theallinpod">https://www.tiktok.com/@theallinpod</a></p> <p>Follow on LinkedIn:</p> <p><a href="https://www.linkedin.com/company/allinpod">https://www.linkedin.com/company/allinpod</a></p> <p>Intro Music Credit:</p> <p><a href="https://rb.gy/tppkzl">https://rb.gy/tppkzl</a></p> <p><a href="https://x.com/yung_spielburg">https://x.com/yung_spielburg</a></p> <p>Intro Video Credit:</p> <p><a href="https://x.com/TheZachEffect">https://x.com/TheZachEffect</a></p> <p>Referenced in the show:</p> <p><a href="https://x.com/the_ai_investor/status/2084687703707361429">https://x.com/the_ai_investor/status/2084687703707361429</a></p> <p><a href="https://x.com/Tesla/status/2085365278276284803">https://x.com/Tesla/status/2085365278276284803</a></p>
Key Insights
- Google is prioritizing high-return AI compute infrastructure investment over frontier model development, creating channel conflict that incentivizes scientist departures since venture capital is readily available for AI startups.
- The frontier model market has consolidated to a duopoly of Anthropic and OpenAI, with Anthropic achieving $80B+ ARR by year-end (started at $10B) through premium pricing for true frontier intelligence while commodity models trade on compute costs alone.
- SpaceX's $100B ARR guidance relies primarily on compute rental at $30-50 per watt, with only 2 gigawatts of capacity needed to hit that target, suggesting the frontier labs are paying 3-5x market pricing due to acute shortage and competitive necessity.
- Starlink's $2.6B adjusted EBITDA on $4.3B revenue with 12M subscribers (doubled YoY) could grow to a $1T+ standalone business within 18-24 months at typical subscription business multiples, funding SpaceX's other ventures as cash generation.
- Bending Spoons can acquire challenged-but-profitable SaaS companies like Airtable and generate 80%+ EBITDA margins by returning them to their natural product-led growth model and eliminating 85-90% of the forced sales-led infrastructure.
- Airtable never achieved ubiquity like Excel or Google Sheets despite cult following, lacking clear use case definition and getting disrupted by AI coding tools (Claude) that eliminate learning curves of no-code platforms.
- Frontier models maintain premium pricing because businesses in competitive industries and dealing with immature use cases will pay premiums to access the best intelligence, unlike commodity middle-market applications.
- Elon's willingness to make heroic long-term bets across Starship, TerraFab semiconductor fabrication, and data centers—using Starlink's cash generation—is unusual among modern CEOs compared to share buyback and dividend strategies.
- US data labeling companies selling identical training datasets to both US frontier labs and Chinese AI companies (Tencent, ByteDance, Alibaba) at scale ($500M/year) represents either commoditized data or concerning technology transfer depending on dataset proprietary value.
- China graduating more math and science PhDs annually than the rest of world combined means they possess inherent capability to create equivalent training datasets without US data sales, questioning whether restrictions would meaningfully prevent catch-up.
- AI compute spot pricing could face downward pressure if memory production increases 20% while demand surges 200%+, potentially extending payback periods from one year to multi-year and undermining SpaceX's financing assumptions.
- Open source models have achieved sufficient capability for 95% of mainstream applications, with meaningful frontier model advantages only appearing in highly competitive industries or immature use cases requiring exploration.
Topics
Transcript
All right, everybody, welcome back to your favorite podcast. It's the all in podcast. It's the summer. It's August 6. Having a hard time getting a quorum here on the podcast. But David Friedberg is here. David Friedberg is back our Sultan science. How you doing, brother? Great to be with you. It's great to be with you. And everybody loves Brad Gerstner is here. He's your Bruce Wayne. If markets are your game, he brings that namaste to your payday. Yes. Buy his glasses at discount and he'll get you one of those fancy Trump accounts. Alright. Welcome back to the program, Brad. I love it. I love it. You bring the rhymes back. I bring a little intro…
Full transcript available for MurmurCast members
Sign Up to AccessMore from All-In with Chamath, Jason, Sacks & Friedberg
Saronic Founders: Autonomous Warships, China's 230X Advantage & Swarms of Robot Ships
Saronic, a defense tech company co-founded by former Navy SEAL Dino Mavroukis and Vibh Altukar, has developed autonomous naval vessels used in a recent rescue operation in the Strait of Hormuz. The founders discuss how autonomous ships address America's severe shipbuilding capacity deficit (230:1 compared to China) and their plans to build Port Alpha, a massive new shipyard in Brownsville, Texas that will transform U.S. naval manufacturing.
Chip Stocks Crash, $20B Fund Margin Called, Frontier Labs: SLOW DOWN AI, Mamdani's Grocery Stores
The pod discusses chip stock crashes and a $20B hedge fund margin call, examines AI frontier labs' calls for pacing development amid security concerns, and explores NYC's new socialist grocery store initiative as political spectacle.
The $1/Hour Worker: Four Robotics CEOs on Humanoids at Home, China's Threat, and the End of Dangerous Jobs
Four leading robotics CEOs discuss the current state of humanoid and quadrupedal robots, their real-world deployment across industries, competition from China, and the timeline for achieving general-purpose robotics at scale. They emphasize that recent breakthroughs in AI perception and language models have shifted robotics from research to practical applications, with significant implications for labor, manufacturing, and national security.
The Fight Over Open Source AI, Anthropic's $1.5B Payout, NYC Socialists: Evictions = Violence?
The All-In podcast discusses three major topics: the geopolitical and business implications of Chinese open-source AI models (Kimmy K3) and potential U.S. bans, Anthropic's $1.5B copyright settlement and regulatory capture concerns, and NYC's new tenant-friendly housing policies that hosts argue will backfire by reducing housing supply and raising rents.
Mark Cuban on the AI Bubble: Who Actually Gets Wiped Out?
Mark Cuban discusses the AI bubble as distinct from the dot-com era, arguing it will primarily devastate VCs and PE firms rather than retail investors, while highlighting massive opportunities for entrepreneurs using AI tools like Lovable despite AI's current limitations in enterprise deployment requiring specialized expertise.