On episode 274 of the All In Podcast, hosts Jason Calacanis, Chamath Palihapitiya, and David Friedberg were joined by Gavin Baker of Atreides Management, with David Sacks absent. The episode was dominated by major AI and SpaceX news, led by the bombshell SpaceX S-1 filing revealing the company is targeting a $1.75 trillion valuation in the largest IPO ever. The S-1 disclosed that Anthropic is paying SpaceX $15 billion annually for access to Colossus compute clusters, effectively adding a Starlink-sized revenue stream overnight. The panel discussed Andrej Karpathy joining Anthropic to lead recursive self-improvement efforts, and Cursor's Composer 2.5 model achieving Pareto dominance after just weeks of training on Colossus-2, demonstrating the power of combining massive compute with proprietary data. The conversation shifted to growing concerns about AI's impact on employment, with Zuckerberg and Cloudflare CEO Matthew Prince drawing criticism for dystopian layoff messaging that fuels public fear of training their own replacements. Chamath argued tech CEOs are 'terrible' at messaging and should stop issuing tone-deaf memos. Baker provided insight into NVIDIA's blowout earnings, noting the company is gaining share despite ASIC competition narratives, and highlighted the $20 billion CPU business emerging within NVIDIA. The episode also covered inflation fears with projections hitting 6% in Q2, the US-China tech CEO summit producing limited concrete outcomes beyond symbolic cooperation, and Friedberg offering his characteristically pessimistic macro outlook on debt spirals and inevitable economic corrections. Baker argued selling chips to China reduces the risk of them developing competing ecosystems, while the closure of the Strait of Hormuz paradoxically benefits America's energy independence and AI dominance.
SpaceX filed for a mid-June IPO targeting a $1.75 trillion valuation and $75 billion raise, more than double the largest IPO in history. The S-1 revealed Starlink generated $11.4 billion in revenue with $4.4 billion in operating income, while the AI business did $3.2 billion in revenue but posted $6.4 billion in operating losses as xAI plays catch-up in the AI race.
Read this moment →According to the SpaceX S-1 filing, Anthropic has committed to a $45 billion three-year deal with SpaceX to rent compute infrastructure at Colossus data centers, adding massive recurring revenue to SpaceX's AI business. This represents SpaceX adding the equivalent of a Starlink-sized business unit overnight through compute rental alone, positioning them as a dominant player in AI infrastructure services.
Read this moment →Meta installed monitoring software on all employee computers to train AI models while simultaneously laying off 8,000 workers, many on H-1B visas, creating widespread fear that employees are training their own replacements. Cloudflare CEO Matthew Prince was condemned for reducing laid-off workers to a label called 'measurers' in his termination memo, with Chamath calling the messaging 'from the PR school of retards' and worsening the AI backlash.
Read this moment →Andrej Karpathy, founding member of OpenAI and former Tesla FSD lead, joined Anthropic to head a new pre-training team focused on recursive self-improvement, where AI models improve themselves autonomously. Chamath argued this could enable order-of-magnitude yearly improvements, creating a 'new form of Moore's Law' for AI capabilities that may render breathless model-by-model coverage obsolete.
Read this moment →Cursor's Composer 2.5 model achieved Pareto dominance over all competing coding models after only 3-4 weeks of reinforcement learning on SpaceX's Colossus-2 compute cluster. Gavin Baker emphasized that Cursor allegedly possesses more proprietary tokens of coding data than exist on the public internet, and the dramatic improvement demonstrates the power of combining massive compute with proprietary training data.
Read this moment →