AI Is a Supply Chain Rather Than an Industry: Here's Why
AI is an interconnected six-layer supply chain that stretches from chip design to software monetization. So owning different "AI stocks" often just means you're placing the same bet on the capital spending of a few big tech players.
Editor's note: This is the second article in a four-part series on AI concentration risk in growth portfolios. The first, Your Diversified ETF Isn't as Diversified as You Think, examined how popular growth ETFs concentrate exposure to this single economic engine. The next two examine the risks facing the supply chain itself and the corporate adoption timeline that will determine which layers ultimately earn their valuations.
Every AI stock you own sits somewhere on an assembly line, whether the company describes itself that way or not.
Investors talk about AI the way they talked about "tech" in 1999, as if it were one industry with one business model. It's not. AI runs through a supply chain with as many distinct layers as an automobile, from raw material to finished product, and each layer carries different economics, different competitors and different risks.
From just $107.88 $24.99 for Kiplinger Personal Finance
Become a smarter, better informed investor. Subscribe from just $107.88 $24.99, plus get up to 4 Special Issues
Sign up for Kiplinger’s Free Newsletters
Profit and prosper with the best of expert advice on investing, taxes, retirement, personal finance and more - straight to your e-mail.
Profit and prosper with the best of expert advice - straight to your e-mail.
Where a company sits on that chain tells you more about its investment case than any AI narrative in its earnings call.
Layer No. 1: Designing the engine
Every AI system starts with a chip designed to run it. Nvidia remains the dominant name here, but AMD, Broadcom and Marvell all design chips or chip components that power AI workloads, either as general-purpose accelerators or as custom silicon built for a specific hyperscaler customer.
These are the companies that draw the most investor attention for a reason: They capture the highest margins in the chain because designing a chip that outperforms the field is genuinely difficult.
That attention creates a diversification illusion, though. If you own Nvidia, AMD, Broadcom and Marvell, you own four companies competing for the same customers, exposed to the same demand cycle and vulnerable to the same slowdown if hyperscaler spending decelerates.
Layer No. 2: Manufacturing the chip
A chip design is a blueprint. Someone still has to build it, and almost nobody can build the most advanced ones. Taiwan Semiconductor Manufacturing Company holds what amounts to a virtual monopoly on fabricating the most advanced logic chips used in AI accelerators.
TSMC can't do that without ASML, the Dutch company that holds more than 90% of the market for the extreme ultraviolet lithography machines required to print circuits at the smallest scale. There is no second supplier.
Applied Materials, Lam Research and KLA supply the deposition, etching, and inspection equipment that fills in the rest of the fabrication process, while Synopsys and Cadence supply the design software that makes the chip layouts possible.
None of these companies show up in most investors' mental picture of "AI stocks." All carry as much exposure to the AI capital spending cycle as Nvidia does.
Layer No. 3: Moving the data
A finished chip is useless sitting in a warehouse. AI workloads move enormous volumes of data between chips, servers and storage, and that movement has its own set of suppliers. Micron and Western Digital supply the memory and storage. Arista Networks, and Broadcom again, supply the networking equipment that connects thousands of chips into a single working cluster.
This layer is where bottlenecks tend to appear first. A chip shortage gets the headlines, but a memory shortage or a networking constraint can slow an AI buildout just as effectively, and it usually gets far less attention from investors watching the wrong part of the chain.
Layer No. 4: Building the factory
An AI data center is a construction project before it's a technology project. Vertiv and Eaton supply the power and cooling systems that keep tens of thousands of chips from overheating. Equinix and Digital Realty build and operate the physical data centers.
Quanta Services and Comfort Systems handle the electrical and mechanical construction work, and Trane Technologies supplies industrial cooling systems built for facilities that consume as much power as a small city.
This is the layer where AI stops looking like a technology story and starts looking like an infrastructure and utilities story.
These companies don't compete on chip performance. They compete on construction timelines, power availability and real estate. This makes their risks look nothing like Nvidia's risks even though their revenue depends on the same buildout.
Layer No. 5: Writing the checks
Microsoft, Amazon, Alphabet and Meta fund the entire chain. Combined, the four are projected to spend $700 billion to $725 billion on capital expenditures in 2026, an increase of about 60% to 77% over 2025. Most of that money flows to the layers above: Chip purchases, construction contracts, power agreements and networking equipment.
These four companies occupy a strange position in the chain. They're simultaneously the largest customers for every layer beneath them and the companies trying to sell AI services to justify that spending.
One company's cost of doing business is another company's entire revenue line. That circularity is not a scandal. It is how every capital-intensive industry works, and it's also why a slowdown at the top of this chain gets felt at every layer beneath it, almost immediately.
Layer No. 6: Monetizing the investment
The last layer has to prove the whole chain was worth building: the software companies selling AI capability to actual paying customers.
Salesforce, Adobe, ServiceNow, Palantir and Datadog all sell AI-enabled products to enterprises, and this is where the infrastructure spending described above has to eventually convert into revenue that justifies it.
This layer carries a different risk than the ones below it. The chip, fabrication and infrastructure layers get paid regardless of whether enterprise customers adopt AI tools at the pace hyperscalers are betting on.
The monetization layer only gets paid if that adoption happens on schedule. That distinction matters more than it sounds.
Why the map matters more than the ticker
Once you see AI as six layers instead of one sector, a different question arises. It's not, "Do I own AI stocks?" It's, "How many layers of the same buildout do I own, and what happens to all of them at once if the assumption underneath the buildout turns out to be wrong?"
A supply chain is only as strong as its most exposed layer, and in AI, every layer currently depends on the same four customers continuing to spend at a pace none of them has sustained before. Know which layer you own before you assume you know what you're exposed to.
Three considerations answer that for your own holdings.
1. Pull the top 10 holdings of every growth or tech fund you own and map them to a layer.
If Nvidia, Broadcom and a hyperscaler all show up in three different funds, you don't own three funds. You own one bet, sized three times over.
2. Check whether your exposure sits entirely in layers one and five.
Chip design and hyperscaler capital expenditures (capex) get the headlines and the fund flows. Layers two through four — the fabrication equipment, the memory and networking suppliers, the data center builders — often trade at lower multiples for the same underlying demand.
That's where the mismatch between attention and exposure tends to correct first.
3. Ask what happens to your specific holdings if hyperscaler capex growth merely slows, rather than reverses.
A deceleration from 70% growth to 20% growth still shows up as a down year for every layer beneath it. You don't need a bust scenario to feel this. A pause is enough.
Related Content
- Worried About an AI Bubble? 5 Ways to Ensure Your Portfolio is Prepared — Whether It Bursts or Not
- Is the 'AI Bubble' a Myth? Why Tech Experts Say AI's Boom Is Just the Beginning
- The AI Investment Nobody Is Talking About? The Infrastructure That Powers It
- Will AI Pay Dividends for Your Firm? To Find Out, Budget for the Whole Iceberg, Not Just the Tip
- Using Google AI Tools Can Give Your Advisory Firm the Edge — If You Do These 5 Things First
