Is AI CapEx the Next AWS or the Next Fiber Optic Bubble — Both Analogies Matter

TL;DR: Bears compare today's AI CapEx to the 1990s fiber optic glut. Bulls compare it to the early-2010s AWS build-out. Both analogies are honest, both are partial — and the price you're paying decides which one matters more.

Two Histories Are Being Used to Argue About AI CapEx

The argument over AI infrastructure spending isn't really about today's numbers. It's a fight between two historical analogies, and which one you reach for tells you a lot about what kind of investor you are.

The Bear Case: This Looks Like the Fiber Optic Bubble

The bears have a point I take seriously. Their core argument — let me restate it as cleanly as I can — is that the Big Four are all building at the same time, with no coordination, on the assumption that demand will keep doubling. When everyone scales supply at once on the same belief, you very often get a glut.

Their preferred analogy is the late-1990s telecom boom. Carriers laid enormous amounts of fiber optic cable across the country and across oceans, betting on internet traffic that would justify it. After the dotcom bust, much of that fiber sat dark for years.

Three concerns sit underneath the bear case:

1. Efficiency improvements in AI models — there are early examples of strong models built on far less compute than the industry assumed. If that becomes the norm, some data centers will sit idle.
2. Falling free cash flow at all four hyperscalers, while AI revenue is still a fraction of the AI spend.
3. Multiple compression. Costco and Walmart trading at richer multiples than Microsoft or Meta is, in itself, a market signal that investors prefer cash today over cash promised later.

These are legitimate questions. Asking them doesn't mean the answer is no. It means you should be asking the price.

The Bull Case: This Is the AWS Moment Again

The bulls reach for a different memory. In the early 2010s, Amazon was burning billions building Amazon Web Services. Investors hated it. Analysts questioned the strategy. The cash going out was visible; the returns were not.

Then AWS turned into one of the most profitable businesses ever built. Last year alone, AWS produced $128 billion in revenue and over $45 billion in operating income.

Bill Ackman has made this comparison directly — investors should be applauding ambitious CapEx, not punishing it. The companies that build the most AI infrastructure today, that lock in enterprise customers earliest, that accumulate the most data and compute, are the most likely to dominate the next decade. The cost of not building, in this view, is losing the race entirely.

Side-by-Side: Which Analogy Fits Better?

Dimension	Fiber Optic Bubble (Bear)	AWS Build-Out (Bull)
Builders	Many telcos, all chasing the same demand	Amazon largely alone for years
Demand validation	Speculative ("traffic will explode")	Already had clear early enterprise pull
Customer lock-in	Low (fiber is fungible)	High (AWS workloads are sticky)
Capital structure	Heavy debt, weak balance sheets	Internal cash flow, strong balance sheets
End state	Years of dark fiber	One of the most profitable businesses ever

When I lay it out like this, the AWS analogy fits better on the supplier side — these are cash-rich companies, not over-leveraged telcos. But the fiber analogy still has bite on the timing of demand. Multiple players building simultaneously is genuinely different from Amazon-alone-in-2012.

How I'm Reconciling Both

Honestly, I think both sides are partly right.

Some of this CapEx will look brilliantly timed in retrospect. Some of it will look excessive. The market will sort winners from losers over the next few years, which is exactly why working through individual companies — not just "the AI trade" as a category — matters so much right now.

The question that actually decides the outcome for any individual investor isn't "bull or bear." It's what price are you paying? A great business at a stretched multiple can still produce a bad return. A flawed strategy bought cheap can still produce a fine one.

FAQ

Q: Why does demand-side timing matter more than the supply-side argument? A: Because we already know these companies can build. The unknown is whether enterprise AI adoption arrives fast enough to absorb the capacity. Most companies have barely begun integrating AI into operations. The build-out is several years ahead of the curve, and that gap is where the risk sits.

Q: Is "AWS again" a strong analogy or a convenient one? A: Partly convenient. AWS faced one early skeptical phase; AI infrastructure is being built by four well-capitalized companies simultaneously. The structural similarity is "front-loaded CapEx for cloud-style infrastructure." The difference is competitive density.

Q: What single signal would change my view? A: Either AI revenue growth at the hyperscalers visibly closing the gap with CapEx — which would settle the bull case — or evidence that compute efficiency is improving fast enough to make a chunk of the build-out unnecessary, which would validate the bear case.