Markets & Strategy · Lesson 5 of 5
The AI Trade: Chips, Clouds, and How to Analyze a Theme
Key takeaways
- The AI trade is a supply chain, not a single stock: equipment makers, foundries, chip designers, memory suppliers, cloud hyperscalers, and application companies occupy different layers with very different economics.
- Index investors already own the theme: a handful of AI-linked mega-caps has recently made up roughly a third of the S&P 500, so a plain index fund is already a substantial AI position.
- A technology being transformative does not make its stocks profitable to buy at any price — the dot-com era transformed the world while many of its flagship stocks lost money for a decade or more.
- Thematic funds typically launch after a theme has already run, charge multiples of a broad index fund’s fee, and overlap heavily with what investors already hold.
No topic generates more questions at brokerages today than artificial intelligence stocks. This lesson maps the ecosystem factually and then does something more durable: shows how to analyze any investment theme with the same discipline. Companies are named below as factual illustrations of where they sit in the supply chain — SweatyOpossum does not recommend securities, and nothing here is a suggestion to own or avoid any of them.
The AI value chain, layer by layer
Modern AI systems are trained and run on specialized processors in vast data centers. Follow the hardware from the ground up:
Equipment. At the base sit the machines that make chipmaking possible — most famously the Dutch firm ASML, the world's only producer of the extreme-ultraviolet lithography machines required for cutting-edge chips. One company, one chokepoint.
Foundries. Most advanced chips are manufactured by Taiwan's TSMC, which fabricates the leading-edge processors for nearly every major designer. Its dominance is why Taiwan appears in every serious discussion of both emerging markets and geopolitical risk.
Chip designers. NVIDIA's graphics processors became the standard engine for training AI models, making it — for stretches of 2024–2025 — the most valuable public company on earth; AMD competes, and several cloud giants now design custom AI chips in-house. Memory suppliers such as SK Hynix and Micron provide the high-bandwidth memory each AI processor consumes.
Hyperscalers. Microsoft, Alphabet, Amazon, and Meta buy those chips by the hundreds of thousands, spending tens of billions of dollars a year building data centers — capital expenditure that has been the chip layer's revenue. Model builders (largely private companies) and application layers sit on top, where the eventual consumer value is supposed to emerge.
The early profits of the AI era concentrated in the lower layers — the classic “picks and shovels” pattern, where suppliers of tools profit before (and whether or not) the prospectors do. The open analytical question for the theme is whether the applications eventually earn enough to justify the infrastructure spending beneath them.
You probably already own it
Here is the fact most AI-curious investors miss: the theme is already in their portfolio. The largest AI-linked companies have recently constituted roughly a third of the S&P 500 by weight. A $10,000 broad U.S. index position therefore already carries several thousand dollars of AI-ecosystem exposure — without any decision, fee, or new fund. The question is never “how do I get AI exposure?” but “do I want more than the large dose I already hold?” — which is the tilt-sizing arithmetic from the previous lesson.
Worked example: the dot-com rhyme
In 2000, the internet's transformative future was correctly forecast — and its flagship infrastructure stock still devastated late buyers. Cisco Systems, the company selling the network hardware of the boom, briefly became the world's most valuable company in March 2000 at roughly $80 per share. The internet then succeeded beyond every prediction: traffic exploded, commerce moved online, Cisco kept growing and earning billions. Yet a buyer at the peak waited over two decades for the share price alone to return to $80, because the peak price had assumed even more than the spectacular reality that followed.
The lesson is not that AI equals the dot-com bubble — nobody knows that. It is that being right about a technology is not the same as earning money in its stocks; the entry price decides, exactly as the expectations arithmetic in growth versus value showed.
Risks specific to this theme
Four stand out. Cyclicality: semiconductors are historically boom-bust; today's shortage-driven pricing power has repeatedly become tomorrow's glut. Customer concentration: a chip designer whose revenue depends on four hyperscalers is exposed to their capex moods — and those same customers are building competing chips. Geopolitics: the leading-edge supply chain runs overwhelmingly through Taiwan, with export controls already reshaping who may sell what to whom. Embedded expectations: valuations across the theme price years of continued extraordinary growth; merely good results can produce poor returns.
Thematic funds: read the label
Dozens of AI-branded funds now exist. Three patterns from research on thematic funds generally: they tend to launch after a theme has performed (capturing enthusiasm, not the run), they charge fees several times a broad index fund's, and their holdings usually overlap heavily with plain large-cap indexes — the same giants, repackaged at a higher price. Any candidate fund deserves the checklist from evaluating investments: what does it hold, what does it cost, what does it add to what you already own, and what has to go right for it to work?
Common misconceptions
“If AI succeeds, AI stocks are certain to be great investments.”
Prices already embed enormous expectations. The internet succeeded spectacularly, and buyers of its flagship stocks at 2000 prices still waited decades to break even. Returns come from results relative to what the price assumed, not from the technology working.
“You need to buy a special fund or stock to get AI exposure.”
AI-linked mega-caps have recently been roughly a third of the S&P 500, so any broad index fund already carries a large AI position. The real decision is whether to hold more than that already-substantial dose.
“Today’s chip leader will stay the leader, so the choice is easy.”
Semiconductor history is a graveyard of “permanent” leaders — Intel dominated for decades before losing the manufacturing crown, and hyperscalers are designing rival chips today. Leadership in fast-moving hardware has repeatedly changed hands, which is an argument for breadth over conviction bets.