Hyperscalers, Silicon and Scarcity

The most valuable companies in the world are a map of whatever the economy has decided is scarce. Read the list in any decade and it tells you what mattered.

In the 1980s the top of the market was oil and iron — Exxon, IBM, General Electric — the companies that moved energy and ran the mainframes. The 1990s belonged to software and the network: Microsoft, Intel, and at the peak of the dot-com mania, Cisco, which for a brief moment in 2000 was the most valuable company on earth because it sold the plumbing of an internet everyone was certain they needed. The 2000s quietly handed leadership back to oil; ExxonMobil sat at the top for most of the decade while the bubble it had replaced finished deflating. The 2010s were the platform decade — Apple, Amazon, Google, Microsoft, Facebook — when value accrued to whoever owned the customer and the network effect, and Apple became the first company worth a trillion dollars. The early 2020s gave us a short romance with the electric future and Tesla's vertical run, and then the ground moved again.

What's scarce now is compute. Not the products built on top of AI, but the raw ability to train and run it. And so, predictably, leadership has rotated to the companies that sell the means of production rather than the products themselves. Nvidia became the most valuable company in the world by selling the chips every model is trained on. Behind it sits a quieter re-rating that's easy to miss: the memory makers. A modern AI accelerator is useless without an enormous amount of very fast memory bolted next to it, and high-bandwidth memory turned out to be the bottleneck inside the bottleneck. That is why Micron — for decades treated as a boring, deeply cyclical commodity business whose stock rose and fell with the spot price of DRAM — is suddenly being valued like a growth company. When everyone is digging for gold, the people selling shovels do well. When the shovels themselves are scarce, the shovel-makers get repriced entirely.

It's worth being precise about who's who. The hyperscalers — Amazon, Microsoft, Google, Meta — are the ones pouring hundreds of billions into datacenters. The names surging fastest are their suppliers. The center of gravity has moved one layer down the stack, from the people building the future to the people selling everyone else the equipment to build it.

The honest question is whether this is the new norm or the top of a familiar arc. There's a serious case that it's different this time: the demand is real, the spending is enormous and still accelerating, and compute is the genuine constraint on nearly everything that comes next. There's an equally serious case that we have seen this movie. Cisco was also selling real plumbing into real demand in 2000, and its stock still fell roughly eighty percent and never reclaimed that high. Memory in particular has always been violently cyclical — supply races to meet the shortage, prices collapse, and yesterday's growth story reverts to a commodity.

I don't know which it is, and neither does anyone who sounds certain. The thing worth watching isn't the price; it's the scarcity. If compute stays the bottleneck, today's leaders keep their crown. If efficiency and competition turn this year's shortage into next year's glut, the leaderboard turns over the way it always has.

Because that is the one durable lesson in the list: it always changes. Oil, then software, then oil again, then platforms, now the silicon that thinks. The only question that has ever mattered is whether you're early or late.