BG2

Brad Gerstner × Gavin Baker × Andrew Fox × Clark Tang

SpaceX’s $135 IPO and $1.77T valuation already have terrestrial AI economics behind them, with EWS reaching the #4 hyperscaler position in 30 days and Anthropic contracts monetizing at $22-23B per gigawatt per year.Orbital compute is an optional 5x capex reduction conditional on Starship reuse and satellite reliability, while Cursor’s coding data and Fable 5 could extend frontier-model revenue; the $1.5T capex case still faces valuation, macro, and execution risk.

Apoorv Agrawal × Ali Ghodsi × Arvind Jain

Ali Ghodsi argues that AGI already exists and LLMs are commodities, shifting durable value toward proprietary data, business processes, and applications rather than model providers.The 95% project-failure rate reflects healthy experimentation, but frozen models and computer use remain unresolved; enterprise adoption, agent revenue, and Glean’s move toward a proactive personal work companion are the catalysts to monitor amid a clear startup bubble.

Brad Gerstner × Clark Tang × Jensen Huang

Nvidia’s $100 billion OpenAI partnership could support a self-build hyperscaler, with 10 gigawatts implying roughly $400 billion of potential Nvidia revenue.Jensen Huang says AI demand is much larger than consensus, while Nvidia’s 30x Hopper-to-Blackwell gain and power efficiency strengthen its moat; China and H1B talent remain risks.

Apoorv Agrawal × Sherwin Wu × Olivier Godement

OpenAI’s enterprise business predates ChatGPT and now embeds Palantir-style forward deployed engineers at T-Mobile, Amgen, and Los Alamos.Across 200 deployments, top-down buy-in, a bottom-up tiger team, and evals first underpin the climb from 46% to 99%.Connectors and RFT could unlock agents and proprietary-data advantages, but GPT-5’s reasoning-token versus latency trade-off remains unresolved.

Bill Gurley × Brad Gerstner

China’s seven-plus open-source labs are compounding through distillation, offering 90% of the intelligence at a 90% price discount, while 200x Google token growth and Groq’s rack utilization show inference demand remains strong.OpenAI and Meta could restore US enterprise share by Q4 this year or Q1 next year, but sub-cost pricing, model commoditization and a possible year-end China deal leave margins and policy outcomes unresolved.

Bill Gurley × Brad Gerstner × Kevin Weil

OpenAI is framed at $300B enterprise value today and $1.6T in five years, while Kevin Weil reports usage deepening beyond Instagram’s growth and tracks weekly actives toward daily, multiple-times-per-day engagement.Its research-to-product loop supports shipping at 70% as models improve, with action-taking, connectors, memory, and personalization ahead, but execution against serious competitors and data-access barriers remain key risks.

Bill Gurley × Brad Gerstner

DeepSeek’s R1 was genuinely 30 to 50% more efficient, but reaching o3-class performance may require 10x compute as export controls and inference demand constrain its cluster.Open-source adoption is accelerating while tariffs and DOGE raise policy risk, leaving a super cycle intact but discount rates, multiples, and Nvidia’s sanctions exposure under pressure.