History is About to Be Made... It's Worse Than You Think
Summary
Tom provides a comprehensive rebuttal to the bear case surrounding AI, specifically addressing claims from prominent investors that AI is a capital misallocation. Tom identifies three main bear arguments: the lack of moats in AI models, the excessive capital expenditure (capex) spending, and high valuations. Tom agrees that large language models (LLMs) are becoming commodities with no moat, but emphasizes that the infrastructure layer—including hardware, specialized software like CUDA, and deep institutional integrations—possesses massive competitive advantages. Tom points out that unlike the dot-com bubble where companies had no earnings and high debt, current AI leaders like Nvidia generate more net income than the top ten dot-com companies combined did in 2000.
Tom highlights the 'Jevons Paradox,' explaining that as AI becomes cheaper and more efficient, demand for it will explode rather than decrease. However, Tom identifies one legitimate risk: the energy bottleneck. Data center power requirements are expected to quadruple by 2034, and grid interconnections can take up to a decade. Tom views this physical constraint not as a reason to sell, but as a 'generational wealth opportunity' to invest in energy and cooling solutions. Tom concludes that investors should focus on companies solving these hardware and power constraints to benefit from the ongoing AI expansion.
Regarding specific stocks and entries:
Mentioned Stocks
Reasoning: Tom argues that Nvidia is actually cheap with a forward P/E in the 20s and a PEG ratio of 0.5. Tom highlights their 75% hardware margins and the massive moat created by the CUDA software ecosystem, which has over 6 million developers.
Reasoning: Tom states that while Palantir is priced for perfection, its metrics are 'one of one' with a Rule of 40 of 145 and 85% revenue growth. Tom emphasizes their government contracts, $8 billion in cash, and zero debt as significant competitive advantages.
Reasoning: Tom recommends Vertiv because cooling infrastructure is critical as AI chip racks approach physical power limits. Tom notes the stock has performed exceptionally well since being added to his list due to its essential role in data center builds.
Reasoning: Tom highlights Bloom Energy for its ability to provide on-site power that is independent of the electrical grid. Tom argues this solves the 'interconnection' problem where data centers are built but cannot wait 4-10 years to be plugged in.
Reasoning: Tom mentions CrowdStrike as a company reporting a record annual recurring revenue (ARR) growth of 32%. Tom includes it as an example of AI-related companies showing fundamental strength that contradicts the bubble narrative.
Reasoning: Tom identifies Constellation Energy as a key beneficiary of the energy bottleneck. Tom believes their nuclear energy capabilities will be a massive staple in powering the next generation of AI data centers.