The return of the empire — and its new arithmetic

For a generation, being a conglomerate — a single company that owns many unrelated businesses — was considered a mistake. General Electric, once the largest company in the world, spent 2023 and 2024 breaking itself into three separate companies. Siemens did something similar with its energy, healthcare and transport arms. The theory, repeated in every business school, was simple: shareholders can build their own diversified portfolio; they don't need a corporate parent to do it for them, and they'll pay less for a company that tries.

The theory still holds for the old-style conglomerate whose only real link between businesses was a shared corporate jet. What has changed is that a very different kind of empire has been quietly assembled underneath the coverage.

India's Reliance Industries added a mobile network, a digital platform and a retail chain to what used to be an oil-and-chemicals business, and became the country's most valuable company. Saudi Arabia's Aramco is no longer just an oil producer; through the country's sovereign wealth fund, it is also a chemicals, hydrogen and — increasingly — an artificial-intelligence holding. Warren Buffett's Berkshire Hathaway crossed a trillion dollars in market value in 2024 without ever pretending to be anything other than a conglomerate.

What the new empires have in common is a bet on two things: energy and computing. Both are commodities. Both require the kind of capital spending that only a handful of balance sheets in the world can carry. And both are the real constraint on almost every other business that matters right now — AI, electric cars, defence, the reshoring of factories to the West. Owning both, inside the same corporate wrapper, gives a company the kind of internal trades the market cannot easily copy.

The clearest expression of this is what the American cloud giants have started doing. In 2023, Microsoft signed a twenty-year deal with a power company to restart a shuttered nuclear reactor at Three Mile Island — the same site of America's most famous nuclear accident — purely to feed one of its data centres. Amazon bought a data-centre campus attached to another nuclear plant. Google signed agreements with a startup building small modular reactors. None of these companies would call themselves conglomerates. But the instinct — control the electricity, control the chips, control the model, control the customer — is the same one.

In the Gulf, the logic has been pushed further. Saudi Arabia's Public Investment Fund launched a new AI holding company called Humain in 2024, sitting alongside a digital arm of Aramco and a technology group called Alat. The pitch to Nvidia or AMD is not just: we will buy your chips. It is: we have the gas, the power plants, the desert real estate and the political will to build the data centres too. Come build them here.

SoftBank's story is the messiest and, in some ways, the most instructive. Its earlier Vision Funds became a case study in what happens when you throw money at every promising startup at once. Since 2022, the group has quietly re-centred itself around Arm, the British chip-design company whose 2023 stock market listing now anchors the whole enterprise, and around new investment vehicles that treat computing as the through-line for everything. Its founder's public forecasts are, as always, difficult to underwrite. The portfolio underneath them has become recognisably industrial.

The Korean chaebol — the family-controlled groups like Samsung, LG, Hyundai and SK — offer the longest-running version of the experiment. Samsung's memory chips, its foundry business and its display panels quietly subsidise each other in ways an activist investor would find intolerable in a Western company. The same structure, in a downturn, is what lets the group keep building new chip factories when its more focused Taiwanese rival, TSMC, has to raise outside money.

China's version is state-adjacent. The country's largest battery maker, its largest EV maker, and its most important chip and materials companies operate with a level of coordination between energy, materials and vehicle assembly that no Western regulator would allow and no Western board would attempt. The results in electric cars and batteries are precisely why European carmakers spent 2024 asking their governments for tariffs.

The Western pushback is beginning. Antitrust regulators in the United States and Europe spent the last decade worrying about horizontal deals — one company buying its direct competitor. The antitrust question of the late 2020s will be a different one: is it acceptable for a single company to own the power plant, the fibre-optic cable, the chip and the AI model on top of it all? The answer will differ by country, and the differences will matter.

For investors, the practical problem is how to value any of this. A traditional analysis — add up the parts, subtract a diversification discount — treats a conglomerate as a bag of unrelated bets. That works fine for old-style groups. It does not really work for a company whose whole thesis is that owning the power and the compute together creates value that neither business could generate alone.

The people building these structures are candid, in private, about the trade. They are accepting a valuation discount today in exchange for a call option on being the only balance sheet in the world that can build a gigawatt-scale AI campus in 2028. If the option expires worthless, the market will punish them the way it punished GE. If it doesn't, the map of global capital in the 2030s will look less like the stock market of 2015 and more like the great trading houses of a much earlier century.

None of this is quite the return of the sprawling 1970s conglomerate. The new empires are narrower, more technical, and organised around two inputs — electricity and silicon — that a decade ago almost nobody thought of as the same business. That, more than any single deal, is the story.