PMS fund manager Saurabh Mukherjea, who runs the multi-million dollar Marcellus Investment Managers, is neither buying Nvidia nor the large language model companies. While the S&P 500 bets half its weight on seven hyperscalers and their suppliers, Mukherjea is building a global portfolio around a simpler question: if AI is real, who gets paid to make it physically possible? The answers he has arrived at are turbine manufacturers in Germany, chip foundry equipment makers, industrial component distributors in Europe and America and, through a completely separate thesis, Airbus and Hermès.
Global data centre capital expenditure is running at $700–800 billion a year, based on what the hyperscalers are disclosing in their annual reports, and Mukherjea believes it will hold at that level for the next four to five years. The supply chain that feeds that buildout — chip foundry equipment, cooling liquids, connectors, switchboards, gensets — will generate between $100 and $300 billion in annual revenue. But Mukherjea is not buying the AI companies at the centre of it. He is buying the pipes and wires and turbines that make it physically possible to run them.
"We are moving into a world shaped by the rise of AI," he told ET Markets in an interview on the sidelines of the ET Alpha Wealth Summit. "The entire supply chain will generate $100 to $300 billion of annual revenue. We want a slice of that."
The S&P 500 carries roughly 50% weight in the hyperscalers plus Nvidia. Mukherjea's portfolio holds only 20% in those seven names. The fund maintained zero exposure to Nvidia through all of 2024, a year when Nvidia surged approximately 2.7 times and pulled the index with it. "The one area where we are underweight is the LLMs themselves — that's where we see bubble risk," he said. "In chip-making equipment and foundries, there is very little reason to believe there's a massive bubble. But in the LLMs, there might be."
Marcellus has now launched its Global Equities Fund, a retail scheme designed to give Indian investors direct access to world-class global businesses via the GIFT City route. The New Fund Offer (NFO) closes on June 19 and is built around 4 multi-trillion dollar megatrends of power, defence and aerospace, AI-linked capex and luxury consumption.
The Power Trade: The Biggest Infrastructure Bet in History
The most consequential downstream AI bet is on electricity. An AI search uses ten times the power of a conventional Google search. Mukherjea's arithmetic shows the planet must install as much generation capacity over the next fifteen years as it has built since 1880. Generation alone will cost $600–700 billion in capex. Transmission and distribution adds close to another trillion. Global power demand has been growing at 4% per annum since ChatGPT arrived, against 2% before it, requiring 400 GW of new capacity every year.
So he is buying the companies that build the infrastructure those AI companies need to run. That means turbine manufacturers in Germany and America, construction equipment providers in America, and industrial component distributors across Europe. The transmission and distribution piece is deliberately well represented alongside generation because he thinks that is where the spending bottleneck will be most acute.
Defence: A Separate Story, Driven by Geopolitics Not Gigawatts
The defence and aerospace thesis has nothing to do with AI. It runs through a different channel entirely due to a structural shift in global military spending that Mukherjea traces to the changing distribution of corporate power. Over the last decade, American-listed companies' share of global profits has surged from 30% to nearly 50%, creating technological leverage that is now playing out through military confrontation. Since Russia invaded Ukraine four years ago, demand for fighter jets, drones and civilian aircraft has accelerated in a way that has never been seen before.
Airbus has its order book sold out until 2041. GE Aerospace, its engine supplier, has orders covering the next ten to eleven years. Safran, listed in Paris, has a full book extending to the seat belt on the plane. The sector is seeing annual capex in excess of $1 trillion. He bought Airbus in its entirety. Beyond Airbus, he is buying components across the world and also in India through the domestic Little Champs portfolio, where some manufacturers feed directly into the global aerospace supply chain.
Also Read | Ghayal hoon isiliye ghatak hoon! Why a global tech crash could be the right medicine for wounded Nifty bulls
AI and Data Centre Play
On semiconductors, Mukherjea draws a hard line between what he will own and what he will not. He holds TSMC, which he describes as exceptional and very difficult to replicate. He holds nothing in Korea. The memory chip thesis does not interest him. "Buying the memory companies, whose cross-cycle operating margins are barely 5–10%, at these astronomical valuations is not a particularly clever idea," he said. "We appreciate there is a memory shortage today. It is not clear to us there will be a permanent memory shortage."
What concerns him more is the behaviour surrounding the trade. "Korea looks louder and more concerning precisely because of the retail behaviour. People liquidating life insurance policies to buy semiconductor stocks is a worrying sign."
Ultra-Luxury: AI Is Concentrating Wealth at the Individual Level
The fourth theme is centred around ultra-luxury companies whose products allow people to signal a rise in status. Most of them are listed in Paris and Milan. Mukherjea's argument is that it connects through the inequality that AI is generating. Just as AI is concentrating profits at the country level, it is concentrating income at the individual level, creating a new class of rich people that goes well beyond the Elon Musks and Zuckerbergs.
The bright coders, system architects and star technologists inside banks and tech firms are being paid disproportionately more because they are doing the work of multiple people. "The ordinary office worker is increasingly facing redundancy — repetitive office jobs are exactly what AI is good at eliminating," Mukherjea said. The world's billionaire count is growing at 7% per annum. He built this thesis four years ago. He says it has played out.
The Portfolio
The result is a 40-stock portfolio with roughly 25% annual churn, managed by a stable team for four years. By contribution to profit, the split is approximately 55% America and 45% rest of the world. By listing, it is 70–75% America, with most of the remainder in Europe. There is a small position in Taiwan. No Korea. Almost nothing in the LLMs.
Arindam Mandal, who runs global equities for Marcellus from the United States and previously worked at Principal Global Investors in New York, argues the physical presence matters for stock selection. "Our on-ground presence in the United States gives us a distinct edge in identifying and investing in businesses that compound capital across economic cycles," he said.
(Disclaimer: Recommendations, suggestions, views and opinions given by the experts are their own. These do not represent the views of The Economic Times)
