NVIDIA (NVDA) remains a top-tier conviction play as it expands into a $20 billion CPU business while maintaining 75% gross margins and returning capital through an $80 billion buyback. Investors should prepare for the SpaceX (SPCX) IPO at a $1.75 trillion valuation, driven by its evolution into a global utility via Starlink and a massive AI infrastructure provider for firms like Anthropic. Focus on the "recursive self-improvement" trend in AI, where specialized tools like Cursor are creating dominant moats in professional coding and verticalized software. To hedge against rising 10-year Treasury yields and potential 6% inflation, consider a concentrated portfolio of five or fewer high-conviction stocks held for a 10-year horizon. Look for secondary opportunities in Natural Gas and Nuclear energy providers, as they are the essential backbone for the gigawatt-scale data centers required by the next generation of AI.

Based on the transcript from the All-In Podcast, here are the investment insights and asset analyses extracted from the discussion.

SpaceX (Private / Ticker: SPCX)

• The company has filed its S1 for a massive IPO, aiming to raise $75 billion at a $1.75 trillion valuation.
• Starlink is the primary "money printer," generating $11.4 billion in revenue with $4.4 billion in operating income.
• Elon Web Services (EWS): A new high-growth segment where SpaceX rents out massive compute clusters (Colossus 1 & 2). Anthropic is reportedly paying $1.25 billion/month ($15 billion/year) for this infrastructure.
• Starship is the key to "rapid reusability," which could lower the cost of mass-to-orbit significantly compared to the current Falcon 9 workhorse.

Takeaways

• Infrastructure Play: SpaceX is evolving from a launch provider into a global utility (Starlink) and an AI infrastructure provider (EWS).
• Valuation Underwriting: Analysts suggest underwriting the $2T valuation by looking at the potential for terrestrial data centers to generate $100B–$200B in revenue by 2030.
• Orbital Compute: A long-term "moonshot" (3–5 year horizon) involves placing data centers in space to bypass terrestrial regulations and cooling costs.

NVIDIA (NVDA)

• Reported "mind-boggling" Q1 results: $81.6 billion in revenue (up 85% YoY) and $48 billion in free cash flow.
• Operating at 75% gross margins despite massive scale.
• Announced an additional $80 billion share buyback and a 25x dividend increase.
• CPU Growth: NVIDIA’s CPU business is projected to hit $20 billion this year, making them a major player in a market traditionally dominated by Intel and AMD.

Takeaways

• Sovereign AI & Enterprise: Growth is shifting from just "hyperscalers" (Microsoft/Google) to "sovereign AI" (nations) and industrial enterprises.
• Market Efficiency: Gavin Baker notes the market is "cross-sectionally inefficient." If NVIDIA’s current P/E multiple is correct, other AI-adjacent stocks (power, cooling) may be overvalued; if those are correct, NVIDIA is undervalued.
• Dominance: NVIDIA remains the only company co-designing with every major AI lab, giving them a "learning mode" advantage that is difficult for competitors to bridge.

Anthropic (Private)

• Reported to be EBIT positive (profitable) in the most recent quarter, a significant milestone for the AI sector.
• Hired Andre Karpathy (formerly of Tesla and OpenAI) to lead a "recursive self-improvement" team.
• Currently paying $15 billion/year to SpaceX for compute power, indicating massive scale and demand for their models.

Takeaways

• Recursive AI: The focus on AI improving AI could lead to a "Moore’s Law on overdrive," where model quality improves by an order of magnitude annually.
• Revenue Growth: Anthropic and OpenAI are nearing a combined $100 billion ARR (Annual Recurring Revenue) with high margins on inference.

AI Sector Themes & Opportunities

• Recursive Self-Improvement: This is identified as the "final frontier." If models can successfully train themselves, the cost per token will drop precipitously, increasing ROI for end-users.
• Verticalized/Small Language Models (SLMs): The future is moving toward smaller, specialized models (e.g., Google’s Gemini Nano in Chrome) rather than just massive general models.
• AI "Harnesses": Tools like Cursor (coding) and Grok Build are becoming as valuable as the models themselves. The "harness" manages state and memory, making the AI agentic and useful for professional work.

Takeaways

• Focus on End-Users: Investors should look past the "model wars" and focus on companies solving specific problems in medicine (protein design), law, or coding.
• Coding as the Lead Use Case: Cursor is highlighted as a "Pareto dominant" tool in AI coding, showing that proprietary data (coding tokens) creates a massive competitive moat.

Macroeconomic Risks

• Bond Crisis: Treasury yields are rising (10-year at 4.6%, 30-year at 5.2%), which typically puts downward pressure on tech valuations.
• Inflation: Projections suggest CPI could hit 6% in Q2, potentially forcing the Fed to consider rate increases rather than cuts.
• Geopolitical Energy Risk: The closure of the Strait of Hormuz is a "black swan" risk. While bad for the global economy, it may relatively benefit the U.S. due to energy self-sufficiency (natural gas and oil exports).

Takeaways

• Concentration: Chamath Palihapitiya suggests a strategy of extreme concentration—holding 5 or fewer high-conviction stocks for 10+ years to weather market volatility.
• Energy as an AI Input: As data centers scale to Gigawatt levels, companies with access to cheap, reliable power (Natural Gas/Nuclear) will be the primary beneficiaries.