GavinSBaker 5.0 17 ideas

A relative value opportunity exists in semiconductors, favoring HBM DRAM-exposed companies over NAND-exposed ones, as the market has not correctly priced in long-term supply deficits and geopolitical risks.

1. THE FACT: Claude in Excel uses Excel only as a presentation layer, while Microsoft's implementation uses Claude to run Excel functions. He states, "For Microsoft, the beatings will evidently continue until morale improves." 2. THE BRIDGE: This highlights a fundamental difference in AI integration, with Claude demonstrating a more "AI-native" approach (using Excel as a front-end for its own powerful AI capabilities) versus Microsoft's more traditional approach (using AI to enhance existing Excel functions). This suggests Microsoft may be struggling to adapt to truly AI-native paradigms, potentially leading to inferior products and continued competitive pressure. 3. THE VERDICT: Watch or potentially short MSFT due to its seemingly less "AI-native" approach to integrating AI into its products (e.g., Excel), which could lead to competitive disadvantages against more agile, AI-first solutions like Claude.

1. THE FACT: Claude Cowork was "evidently built in 10 days with Claude Code" and is what Copilot "should have been," while Microsoft has been working on Copilot for years. 2. THE BRIDGE: This implies that Microsoft's AI development (Copilot) is significantly slower and less effective than competitors like Claude (Anthropic), despite a longer development period. This could indicate a competitive disadvantage for Microsoft in the rapidly evolving AI software space. 3. THE VERDICT: Watch or potentially short MSFT due to its apparent lag in AI product development (Copilot) compared to competitors like Claude, suggesting a potential competitive weakness in the AI software market.

1. THE FACT: Gavin Baker states Nvidia is buying Groq (later clarified as a licensing agreement with engineers joining Nvidia) for two reasons: 1) Inference is disaggregating into prefill and decode, and SRAM architectures have unique advantages in decode for memory bandwidth-intensive workloads. 2) This relates to Rubin CPX, Rubin, and the "Rubin SRAM" architectures. 2. THE BRIDGE: This indicates Nvidia's strategic move to enhance its inference capabilities, particularly in the decode phase, by leveraging SRAM architectures. This strengthens Nvidia's competitive position in the evolving AI inference market, suggesting continued dominance and innovation. The mention of Rubin architectures further solidifies Nvidia's long-term roadmap. 3. THE VERDICT: Long NVDA due to its strategic move to integrate Groq's SRAM architecture expertise (via licensing and talent acquisition) to optimize for AI inference, particularly in decode workloads, reinforcing its leadership in next-gen GPU architect

1. THE FACT: Starlink v3 will be 20 kilowatts with no thermal issues. Elon's stated plan is 100 kilowatts for each AI satellite, slightly less than a full Blackwell rack. Starship should be able to lift 10 to 15 megawatts to sun-synchronous orbit per flight. 2. THE BRIDGE: This provides concrete technical details supporting the feasibility and scale of deploying significant AI computing power in space. It reinforces the idea that space-based AI infrastructure is a serious and scalable endeavor, creating opportunities for companies involved in its development and deployment. 3. THE VERDICT: Long AI and space infrastructure, specifically companies that could benefit from the deployment of high-power AI satellites and the associated launch capabilities, as detailed by Elon Musk's plans for Starlink and Starship.

1. THE FACT: Gavin Baker is "deeply amused" by commentary that datacenters in space don't work, highlighting Elon Musk's experience with large GPU clusters and SpaceX's capabilities in mass to orbit and operating Starlink. He then details Starlink v3's 20kW power, Elon's plan for 100kW AI satellites (nearly a Blackwell rack), and Starship's ability to lift 10-15 megawatts to orbit. 2. THE BRIDGE: This suggests a strong belief in the viability and future potential of space-based data centers for AI, driven by SpaceX's technological advancements. Companies involved in space infrastructure, satellite technology, and potentially AI hardware designed for space deployment could benefit. 3. THE VERDICT: Long AI and space infrastructure plays, particularly those related to SpaceX's vision for space-based data centers, given the technical feasibility and potential for massive power delivery to orbit.

1. THE FACT: Gavin Baker mentions "Google losing" in the context of a discussion about investing in tech, AI, and being the low-cost producer. He also states that xAI is likely to be first to market with a Blackwell model. 2. THE BRIDGE: The implication is that Google is falling behind in the competitive AI landscape, particularly in being a low-cost producer and potentially in bringing next-gen models to market compared to competitors like xAI. This could negatively impact Google's competitive position and future growth in AI. 3. THE VERDICT: Watch or potentially short GOOGL due to concerns about its competitive positioning in AI, specifically regarding being a low-cost producer and potentially lagging in bringing advanced models to market compared to peers like xAI.

1. THE FACT: The next 12 months will see the GB300 (Nvidia's next-gen GPU) ramp up, followed by Rubin. Models trained and inferenced on these GPUs are expected to show a dramatic leap in capability, leading to the "first significant AI" advancements. 2. THE BRIDGE: The ramp-up of advanced GPUs like GB300 and Rubin suggests a significant increase in demand for high-performance computing hardware and the companies developing and deploying advanced AI models. This implies continued strong performance for GPU manufacturers and AI infrastructure providers. 3. THE VERDICT: Long NVDA and other AI infrastructure/hardware plays due to the anticipated dramatic leap in AI capabilities driven by new GPU architectures (GB300, Rubin) over the next 12 months.