That has and continues to play out. Especially as deterministic software blends with probabilistic ‘AI’ software in this fourth year post-ChatGPT, of the AI Tech Wave.
But as I’ve said a few times in these pages, ‘AI’ are just two non-contiguous letters in the alphabet without chips.
In fact, it’s the fundamental factor (besides power and a few other things), gating how fast billions of people can learn the joys of ‘tokenmaxxing’. What tens of millions of developers are all addicted to doing.
“Semiconductors, or chips, are again turning out to be the It Girl of the global economy.”
“Why it matters: Chips are essential to the AI build-out, and that’s driving a huge burst of demand, creating supply shortages, pushing up prices and creating an investment frenzy.”
“It also puts chips at the center of geopolitics.”
“The latest: Nvidia CEO Jensen Huang boarded Air Force One Tuesday night, joining the delegation to China, during a refueling stop in Anchorage.”
“A source tells Axios’ Mike Allen that President Trump rang Huang and invited him after media reports that the chip mogul wasn’t part of the big delegation of U.S. CEOs joining the summit in Beijing.”
Wall Street has not missed the semiconductor swing of late:
“Zoom in: The stock market is now largely a story about chips. Since the launch of ChatGPT in 2022, the PHLX semiconductor index, which tracks 30 of the largest companies in the industry, has grown to account for 16% of the S&P 500’s market cap, up from 4%, Bloomberg’s John Authers noted earlier this week.”
“Chip stocks, which had been powering a rally since March, dragged the market lower on Tuesday as the PHLX index fell 3%.”
And it’s just sinking in for many investors that old habits of treating semiconductors as cyclical commodities at the bottom of tech stacks, may need to be modified for the AI Tech Wave. It’s Box no. 1 below, and has been for most of the major tech waves to date.
And it may stay a bit different for longer than a few quarters.
“The big picture: It’s hard to overemphasize how weird the chip market is at the moment. Outside of the pandemic, when supply chain issues drove up costs, typically the price of computing power has trended down.”
“Now, the frenzied demand for “compute” to power AI has driven up prices throughout the chip supply chain: from the fanciest logic chips to memory chips that store data to older ones that power infrastructure like cars or industrial machinery.”
“The part no one really bet on for 2025 and 2026 was the acute shortages,” says Kelly Littlepage, founder of OneChronos, a company that designs new kinds of markets for investing. “Even older chips are now appreciating in value as an asset.”
“What to watch: Increasingly, companies are looking for ways to hedge the cost of compute.”
“Littlepage’s firm, which already offers so-called “smart markets” in which institutional investors trade complex stock and FX instruments, is developing one for compute, with input from Paul Milgrom, who won a Nobel Prize in 2020 for his work on auction theory.”
“Driving the news: It’s not alone. CME Group said Tuesday that it is launching a new futures market to trade compute.”
““As the backbone of the digital economy, compute is the new oil of the 21st century,” CME CEO Terry Duffy said in a statement.”
There he said the ‘Oil’ world. It’s just that chips have been the ‘oil’ of the tech world for decades now. It’s just that everyone needs A LOT MORE of them all of a sudden with this AI ‘inference token matrix math thing’ that AI Agents and AI Reasoning particularly want in large dollops.
And this time the chip supply chain globally is particularly gummed up with this geopolitical thing. Which is why there were at least four CEOs traveling with the US President to China, representing over $12 trillion in US market cap. And they forgot one of them and had to pick him up in Alaska.
“Friction point: Geopolitics hangs over all of this. Trump’s meeting with Chinese leader Xi Jinping this week swings a spotlight on the global trade in chips.”
“The U.S. dominates in advanced AI compute that sits at the top of the stack. But China owns the bottom — it has the critical minerals, and the more basic foundational chips.”
“In other words, they need each other. The prospect of either economy achieving “industrial sovereignty” isn’t likely anytime soon, as a report from Deutsche Bank notes. The authors call chips “the biggest AI bottleneck.”
“Between the lines: In the middle sits Taiwan. Most of the world’s most advanced chips are manufactured on the island, as Treasury Secretary Scott Bessent said at Davos this year.”
“If that island were blockaded, if that capacity were destroyed, it would be an economic apocalypse.”
“The bottom line: Chips are kind of a big deal.”
It’s particularly true since AI in my view is still in its ‘mainframe’ era, where massive AI Data Centers worth trillions are needed to generate the majority of the AI Tokens that truly make the compute go around. At least until we get to the PC phase of this AI era, where local compute can take off much of the load from the ‘mainframes. I’ve written and discussed this theme at length in these pages.
I’ve long said emiconductor chips are like potato chips. They can always make more. That’s generally been true through most tech waves.
This time, though, it’s like eight billion people are rushing into the McDonald’s trying to SuperSize their fries at the same time.
The PHLX semiconductor index moving from 4 to 16 percent of S&P 500 market cap since ChatGPT is the cleanest signal that investors are repricing chips out of the cyclical commodity bucket. The OneChronos compute-hedging mention is the corollary: when forward compute becomes a tradeable instrument, semis behave like an asset class with a yield curve, not a hardware spec sheet. Once that frame holds across two reporting cycles, the cyclical mean-reversion call gets a lot harder to make.
The PHLX semiconductor index moving from 4 to 16 percent of S&P 500 market cap since ChatGPT is the cleanest signal that investors are repricing chips out of the cyclical commodity bucket. The OneChronos compute-hedging mention is the corollary: when forward compute becomes a tradeable instrument, semis behave like an asset class with a yield curve, not a hardware spec sheet. Once that frame holds across two reporting cycles, the cyclical mean-reversion call gets a lot harder to make.