All-In

Naveen Rao × Chamath Palihapitiya

Unconventional AI taped out a physical dynamical computer five months after starting in earnest, generating images at 500 nanojoules each versus millijoules on a GPU.Its 4D architecture co-locates compute and memory, targeting 1,000× power efficiency as AI demand approaches an energy wall in about 3 years.A rack-scale product is promised within two years, with adoption hinging on model-layer porting and a new Python-based software stack.

Jensen Huang × President Trump

Jensen Huang cites failed forecasts on radiology, AI-generated code and entry-level jobs to challenge doomsday narratives, while arguing that AI incidents so far call for root-cause engineering and independent evaluation rather than sweeping regulation.With 80% of AI-native companies receiving $400 billion in recent venture funding using open models, NVIDIA is building infrastructure bottlenecks as Jensen predicts China will reach advanced lithography by 2030, leaving commercialization, lab controls and data-center execution to monitor.

Jason Calacanis × Pat Gelsinger × Anton Osika

Pat Gelsinger attributes Intel’s decline to replacing technical leadership with spreadsheet-driven capital allocation, while Apple, Nvidia and TSMC compounded patient capability-building into durable platforms.The semiconductor recovery remains exposed to Taiwan’s energy dependence, even as AI’s long runway is constrained by electricity; Lovable meanwhile reports production-scale adoption, with defensibility shifting toward orchestration, security, operational data and user feedback.

Andrew Feldman × Robin Rombach

Cerebras reports a $25 billion backlog as faster inference turns token capacity and data-center execution into scarce product advantages, while open models take routine and sovereign workloads.Feldman expects its architecture to improve well over 2x in 18 months; staged releases, red-teaming and recursive model improvement remain catalysts—and risks—as multimodal AI reaches film and robotics.

Chamath Palihapitiya × Jason Calacanis × David Sacks × David Friedberg

Palantir and NVIDIA’s sovereign-AI pitch shifts enterprise value toward owning compute, data, model weights, and proprietary “alpha” rather than renting frontier intelligence. 8090 made Claude 1.4x to 4x cheaper and 1.5x faster than Opus 4.8 alone, while a 16.4x-cheaper open model and falling token prices could challenge an OpenAI-Anthropic duopoly, with AI’s employment impact still unresolved.

Gavin Baker × Travis Kalanick

Mamdani-backed challengers swept three safe Democratic New York districts despite Polymarket assigning only a 26% chance, demonstrating the DSA’s organizational leverage.The panel linked socialism’s momentum to housing, debt, inflation, and downward mobility, while GLM 5.2 brought a 744-billion-parameter, MIT-licensed model near the coding frontier at 85% lower cost than GPT 5.5.Open weights may shift value from frontier labs toward chips, memory, hosting, and routing, as Micron’s sold-out HBM supply makes DRAM and power the immediate constraints.

Chamath Palihapitiya × Jason Calacanis × David Sacks × David Friedberg

Anthropic’s Fable 5 pairs benchmark leadership with 30-day retention, profiling, and user-specific downgrades, making access an enterprise-governance risk.Restrictions are pushing Ohalo toward local open models, while power costs rose from roughly $4 billion–$5 billion per gigawatt to $100 billion.May CPI at 4.2% and PPI at 6.5%, plus Iran-driven energy risk and disputed Los Angeles ballot patterns, leave macro and institutional trust unresolved.

Bill Gurley

Frontier-model convergence is shifting value toward hot-swappable control planes, local hardware, and portable context rather than any single model winner.Enterprise token bills show AI savings remain unproven, while a possible open-model crackdown could intensify centralization and leave the world dependent on Chinese models.

Gavin Baker × Chamath Palihapitiya × Jason Calacanis × David Sacks × David Friedberg

SpaceX’s proposed $75 billion raise at a $1.75 trillion valuation pairs Starlink’s $11.4 billion revenue and $4.4 billion operating income with terrestrial data centers potentially reaching $100-$200 billion by 2030-32.Nvidia’s $81.6 billion quarter and Anthropic’s profitability strengthen the AI case, but rival ASICs still lack comparable MLPerf tests while rising yields, 310% global debt-to-GDP, and uncertain regulation remain risks.

Michael Intrator × Aravind Srinivas × Arthur Mensch × Daniel Roberts

AI infrastructure is decommoditizing at cluster scale: CoreWeave finances contracted GPU “boxes,” IREN monetizes 4.5 GW of power, and demand now spans memory, networking and construction.citeturn0search0 Perplexity is positioning itself as a neutral model conductor with positive gross margins but no overall profitability, while Mistral’s enterprise opportunity depends on open deployment, human signal and deterministic controls that prevent agent data leakage.