Google's AI Brain Drain, SpaceX's Huge Quarter, Airtable's 90% Collapse, US Data Fuels China AI

What It Covers

Google's AI talent exodus accelerates as Jeff Dean departs after 27 years and Demis Hassabis moves to a ceremonial role, signaling a strategic pivot toward infrastructure over frontier model development. Simultaneously, SpaceX reports 92% revenue growth with $7.8B in Q2, Airtable sells for 90% below peak valuation, and Chinese labs purchase US training data.

Key Questions Answered

• •Google's Capital Reallocation: Google is shifting $200B in CapEx toward data center infrastructure rather than frontier model development, treating compute-as-a-service as high-alpha, low-beta returns versus the high-risk model development path. Investors and boards increasingly view renting compute to Anthropic and OpenAI as more predictable ROIC than competing directly on frontier models, explaining why top researchers are departing for startups.
• •Frontier Model Duopoly: The frontier AI market has consolidated to effectively two players — Anthropic and OpenAI. Anthropic grew from $10M to $80M+ ARR in 2025 and is tracking toward $100M-$120M exit ARR. Non-frontier models cannot charge for the model layer itself, only for compute and inference. Enterprises in competitive industries will pay premiums for frontier intelligence, mirroring Apple's premium pricing against free Android alternatives.
• •SpaceX Starlink Bull Case: Starlink generates $4.3B quarterly revenue with $2.6B adjusted EBITDA at 12M subscribers, doubling year-over-year at $66 ARPU. Applying a 30x multiple to projected $30B free cash flow suggests Starlink alone could reach a $1T market cap within 18 months. The Starship v3 satellite deployment adds 60 terabits per second per launch versus 2.6 terabits for Falcon 9, enabling direct-to-cellular expansion.
• •SaaS No-Code Disruption: No-code tools represent the most AI-disrupted SaaS category because Claude, Lovable, and similar tools now perform the same functions without requiring users to learn proprietary interfaces. Airtable's sale to Bending Spoons at $2.25B — roughly 10% of its $11.7B peak — reflects this structural shift. Bending Spoons can potentially generate $300M-$400M annual EBITDA by eliminating the sales-led motion and returning to product-led growth.
• •Private Equity SaaS Playbook: When a PLG-driven SaaS company forces a sales-led overlay to satisfy venture return expectations, attainment rates collapse — Airtable's sales team hit only 30% of quota. Acquirers like Bending Spoons can restore profitability by eliminating 85-90% of cost structure, reverting to organic growth, and leveraging AI to maintain codebases without institutional knowledge. The payback period on the $2.25B acquisition could be under three years.
• •AI Training Data and China: US data labeling companies Surge AI and Merkor, valued above $20B combined, sell identical PhD-curated reinforcement learning datasets to both US frontier labs and Chinese AI companies including Tencent, ByteDance, and Alibaba. Chinese labs spend approximately $500M annually on these datasets. The strategic risk assessment hinges on whether this constitutes dual-use military technology transfer or commodity labor arbitrage that China could replicate independently.