Alex Sacerdote

Nvidia was bought at 4x earnings in 2023 because the AI S-curve is just beginning, with the infrastructure layer only 10% penetrated. The demand for compute is exploding, and Nvidia is the dominant chip supplier with massive earnings power not yet appreciated by the market.

Google is one of our largest positions. It has a strong consumer franchise with Gemini, a huge cash cow to fund AI, and is one of three likely winners in the foundational model oligopoly. Google's enterprise strength and ability to compete in coding make it a key AI bet.

Amazon was bought for AWS at a time when it was effectively free – the market was valuing it as a retail company while AWS had massive underappreciated earnings power as a cloud platform with a 7-year lead, scale advantages, and a huge TAM. The S-curve for cloud was just getting started.

Celestica has transformed from a commoditized contract manufacturer to a critical AI infrastructure supplier – sole supplier of Google TPU servers, dominant in liquid cooling, and has 50-60% share of cloud Ethernet switch market. AI servers are 40x more expensive than old servers and require constant innovation, making Celestica a high-growth, high-margin business with rising competitive advantages.

Tesla was bought in 2019 at 5x earnings because the electric vehicle S-curve was just inflecting – barriers to adoption (price, range, supply chain) had been removed, leading to exponential unit growth. The underappreciated earnings power at that point was enormous.

Apple was bought at 4x earnings because the smartphone S-curve was inflecting – Jobs brought iPhone to $200, AT&T had 3G, touchscreen made it easy. Exponential unit growth and a powerful ecosystem (app store, switching costs) led to enormous earnings growth that the market didn't anticipate.

Delta Electronics and Advanced Energy supply power supplies for AI servers. Each new Nvidia chip/rack uses 50-125% more power, driving ASPs up 40% per year for the next four years with higher margins. This is a clear beneficiary of the AI infrastructure buildout where demand outstrips supply.

Corning is the dominant fiber supplier for data centers, with a very high share and a product that is thinner, more bendable, and custom-manufactured. AI needs massive fiber (e.g., one Microsoft data center uses enough to circle Earth 4.5x). Future scale-up connections (copper to fiber) will 2-3x Corning's opportunity. Its fiber business is the fastest-growing, highest-margin part of the company.

We sold almost all application software and were net short entering 2025. The thesis is that AI is disrupting the old software model: AI products from incumbents are not good, CIO budgets are shifting to AI tokens, pricing power is eroding, seats are decreasing, and new AI-native startups could wipe out incumbents. The valuations are high and the AI exposure of most software companies is tiny (1-2% of sales).

Delta Electronics and Advanced Energy supply power supplies for AI servers. Each new Nvidia chip/rack uses 50-125% more power, driving ASPs up 40% per year for the next four years with higher margins. This is a clear beneficiary of the AI infrastructure buildout where demand outstrips supply.