Shyam Sankar

"We think that the value is going to accrue at the chips layer and at the ontology layer... the model companies have to run up the stack away from just providing a pure model because the model is commoditized." Foundational AI models (like GPT-4 or Claude) are becoming interchangeable commodities. Enterprise value will not be captured by the models themselves, but by the software infrastructure that connects these models to proprietary corporate data and real-world business decisions. Palantir owns this "ontology" layer. LONG. Palantir is uniquely positioned to be the primary software beneficiary of enterprise AI adoption because it provides the critical, non-commoditized infrastructure that makes AI actionable for large organizations. High valuation multiples; enterprise sales cycles remain complex despite technological acceleration; potential for big tech (Microsoft/AWS) to build competing ontology frameworks.

"We think that the value is going to accrue at the chips layer and at the ontology layer." While the software layer of AI models faces a race to the bottom in pricing, the physical compute required to train and run these models remains a scarce, highly valuable bottleneck. The companies designing the silicon are capturing the foundational economic rent of the AI boom. LONG. Semiconductor designers are the structural winners of the AI arms race, insulated from the pricing wars happening among foundational model providers. Geopolitical tensions over Taiwan (TSMC); cyclical overbuilding of data centers leading to an eventual digestion period for GPU demand.

"Airbus was in the midst of ramping the A350... we just helped them automate all of that and the ontology at that point was really related to parts, sequencing of work, defect rates... Now you have the greatest data asset possible for this tail." By deeply integrating Palantir's software into its final assembly lines, Airbus has created a structural data advantage over its competitors. This allows them to identify design defects faster, optimize production planning, and maximize in-service uptime, directly translating to better margins and faster delivery ramps. LONG. Airbus possesses a superior, software-optimized manufacturing process compared to its primary rival (Boeing), allowing it to capture outsized market share in the commercial aerospace duopoly. Global supply chain bottlenecks (engines, titanium); macroeconomic slowdowns reducing airline capital expenditures.

"We've been staring at this rare earths problem for the better part of 15 years and we more or less have solved it for like two billion bucks. It's roughly a rounding error relative to everything that we're doing and now we're on a trajectory to do it." The US government has recognized that reliance on China for critical materials is a national security emergency. They are actively subsidizing and securing domestic supply chains. US-based rare earth miners and processors will be the direct beneficiaries of this government-backed reshoring mandate. LONG. Domestic rare earth producers have a guaranteed, strategically motivated buyer (the US government/DIB) willing to subsidize operations to break China's monopoly. Commodity price volatility; China could flood the market with cheap rare earths to bankrupt Western competitors before they reach scale.

"We think that the value is going to accrue at the chips layer and at the ontology layer." While the software layer of AI models faces a race to the bottom in pricing, the physical compute required to train and run these models remains a scarce, highly valuable bottleneck. The companies designing the silicon are capturing the foundational economic rent of the AI boom. LONG. Semiconductor designers are the structural winners of the AI arms race, insulated from the pricing wars happening among foundational model providers. Geopolitical tensions over Taiwan (TSMC); cyclical overbuilding of data centers leading to an eventual digestion period for GPU demand.