In our last letter, we made the case that index-like investing is yesterday’s solution for today’s market. This letter is the other half of that conversation. If we are not going to own the index, we have to commit to something better. Our answer is simple: we invest in themes.
 

A theme is a discipline, not a slogan. It narrows the investable universe from a few thousand tickers to the few dozen actually pointed at where the world is going. Themes run on long, secular runways and are measured in decades, not quarters, which is what lets compounding actually work. And the build out of artificial intelligence, the biggest investable theme of this decade, is anything but subtle. It is one of the largest infrastructure projects in human history and it is dominating nightly newscasts, earnings calls, industry conferences, podcasts, water cooler talk and beyond.

The Magnificent Seven’s capital spending is roughly $725 billion in 2026, on a credible path past $1 trillion annually by 2027. Global management firm McKinsey & Company calls the data center build-out a $7 trillion race through 2030. The theme is confirmed with the balance-sheet commitments by the seven biggest companies in America. And it is not just the Mag Seven. The governments of the United States and China have made AI infrastructure industrial policy. Jensen Huang, Elon Musk, Satya Nadella, Sundar Pichai, Andy Jassy, Mark Zuckerberg, Tim Cook, and Sam Altman are on podcasts regularly explaining what they are building and why. Are they all wrong? Let the facts talk because the main players are telling you what is happening.

In the Q1 2026 earnings cycle, the key companies sang a similar tune. The most impressive to us: Anthropic grew 80× year-over-year and CEO Dario Amodei called it “too hard to handle,” then rented every megawatt of Elon Musk’s Colossus 1 to keep up. Alphabet’s Sundar Pichai told shareholders, “We are compute constrained in the near term.” Amazon’s Andy Jassy disclosed two AWS customers asked to buy out the entire 2026 Graviton capacity (AWS said no). Musk is on a path to 555,000 GPUs at Colossus 2 and a stated five-year target of 50 million H100-equivalents, aka hardware options, which is “more than everyone combined.” Again, not enough compute.

Jensen Huang is the CEO of Nvidia, a GPU producer and the largest company in the U.S. Thus, Jensen has a credible take on the AI revolution. When he talks, we listen. He recently described the AI ecosystem with the analogy of a five-layer cake.
 

Read the cake bottom-up. Layer 1 — Energy: the ceiling on intelligence; nuclear fuel (CCJ) and grid utilities (ETR). Layer 2 — Chips: GPUs (NVDA), HBM memory (MU), chip-design software (CDNS), fab materials (ENTG). Layer 3 — the AI Factory: racks, networking, cooling (NVT). Layer 4 — the models: GPT, Claude, Gemini, Grok, Llama. Layer 5 — the applications: where the Mag 7 monetize (GOOG).

Investing is, at its core, the business of being right about earnings before everyone else is. The market prices the consensus view. When earnings come in higher than the consensus, the stock goes up. When they come in lower, the stock goes down.
 

An investment edge worth having is a different view of where earnings are going. Right now, we think Wall Street is structurally underestimating earnings for the suppliers on the cake’s first two layers, and there are two reasons why. First, analysts are still modeling AI demand off cloud growth, and they have not internalized that inference is now a larger compute load than training. Every chatbot reply, every coding session, every customer-service bot is GPU traffic, and that traffic is exploding. Second, AI agents are about to accelerate that traffic. An agent does not ask one question; it asks 50, in a loop, in the background, all day. A single user with an agent uses roughly 100× the compute of the same user without an agent, and this is only the beginning.

There is a strange asymmetry in financial media. The people building AI — Jensen Huang, Sam Altman, Dario Amodei, Sundar Pichai, Andy Jassy, Elon Musk, Satya Nadella, Demis Hassabis — are vocal and public with their comments on the progress. They explain what they are seeing, what they are buying, what they are constrained by, and what comes next.
 

Meanwhile, many Wall Street equity analysts build financial models like they did in the past which have proven inaccurate in the AI buildout. Their job is to be slightly less wrong than the next analyst. Why listen to an analyst when Jensen Huang is on YouTube? We listen to the builders. Then we go buy the suppliers they cannot run without.

An index is a list someone else made for someone else’s reasons, weighted by market cap. It is backward-looking by design. A thesis looks forward, shaped by a theme focused on what matters most. NVDA is in the index; the rest of our six — MU, CDNS, ENTG, CCJ, ETR — are either not in it or are tiny. Memory next to every GPU. Software every chip is drawn in. Materials every fab needs. Uranium and electrons every PPA buys. Six positions, sized with conviction. Each one earning its keep on the same theme.

The evidence is sitting in plain sight — the capex, the earnings calls, the builders telling you on camera exactly what they cannot get enough of. So why are advisors still parked in the closet-index? Why the steady drumbeat of bubble talk, often from the very people who, by their own admission, do not use the technology? It is a fair question, and the honest answer has more to do with human nature than valuation.

1. Start with what most “active” managers actually do. They hug the index. They own the same mega-caps in roughly the same weights, shade a point here and there, and call it a strategy. Closet indexing is the safest career move in our business: own what everyone owns and you can never be too wrong, never be the only one wrong, and never   be the one who gets fired.

    

2. And here is the part nobody says out loud. The hard thing is not learning the new world. It is unlearning the old one. Most of us in this business are north of 50. I am one of them and change is difficult. It is way easier to put your head in the sand, hope the old way works again, and pray clients don’t put their portfolio into ChatGPT.

    

3. Douglas Adams had a rule for this. Anything that exists when you are born is just normal. Anything invented before you turn 35 is exciting and new. Anything invented after that is against the natural order of things. He meant it as a joke. It stops being a joke when it is allocating other people’s capital. The advisor who learned the index in 1995 feels AI as an intrusion, a fad to be endured, rather than what it is: the largest capital cycle of our professional lives.
    

Ultimately, the simplest tell has nothing to do with age — it has to do with use. If you do not use the technology, the thesis remains abstract. The people using it every day are not debating whether it is real; they are asking how anyone could miss it.

So most of the bubble calls are not analysis. They are the sound of people defending a world they already understand — and that is human, and forgivable, but not a reason to hold a portfolio built for the last decade. The index can only own what has already won. A theme looks forward. In a market being rebuilt from the power plant up, that is the whole difference between narrating history and owning a piece of it.

Don’t take our word for it. Look at the numbers. Read the earnings calls. Listen to the builders. Then ask the simple question: do my holdings reflect the world we are actually living in? If you are an advisor, think about the families you represent. Five years from now, when the receipts are in, will your book look like a portfolio built for the next decade or the last? Challenge the norm. Do the work. Find a modern solution. Fighting for what is right for clients is, in fact, the right thing to do.