It Doesn't Matter If It's a Bubble.

The question isn't whether the valuations hold. It's what gets built regardless.

The Three Speeds of AI | Part 3 of 5

This is Part 3 of a five-part series on the three speeds of AI. So far: AI moves at three decoupled speeds — capability, infrastructure, and adoption — and most leaders calibrate primarily to adoption, the most visible and least predictive layer. Part 2 looked at capability: the two-year organizational learning gap that no subscription can close. This piece examines infrastructure — and why the signals that matter most are being read as stock market news rather than as the operating environment your business is about to inhabit. [Part 1] | [Part 2]

The signals are not subtle.

NVIDIA’s market position. Taiwan Semiconductor building a $165 billion gigafab cluster in Arizona — five fabrication facilities, two advanced packaging facilities, the most advanced chip manufacturing in the world being relocated onto US soil at scale. The Stargate project — SoftBank, OpenAI, Oracle, backed by the US government — committing $500 billion to AI data center construction, with $100 billion deploying immediately across sites in Texas, Ohio, and New Mexico. xAI building what its founders describe as “by far the most powerful AI system on Earth” in Memphis, now expanding with a $20 billion investment in Mississippi. The four largest hyperscalers guiding toward nearly $700 billion in infrastructure capital expenditure in 2026 alone.

You would have to work hard to miss this.

And yet when these signals come up in board conversations, the questions I hear most often are: Is this a good investment for my portfolio? Is this a bubble? And either way — what does any of this have to do with us?

All three questions are understandable. All three are aimed at the wrong thing.

I hear them most often not from AI companies — they’re building the infrastructure and already know what it means — but from the leaders of closely held firms, family enterprises, and independently owned businesses whose operating environment this infrastructure is about to reshape.

The bubble question is the most instructive one to examine, because it has a historical answer.

The Shift·Lift subscribers have full access to all past articles, including the work that led into this series. Be sure to order your copy of CRAFT Thinking™ available on Amazon; Free for Kindle Unlimited readers.

In the late 1990s, the fiber optic industry overbuilt spectacularly. Companies poured billions into laying cable across continents, under oceans, through cities — far in excess of what the internet of that moment required. Valuations soared and then collapsed. Companies went bankrupt. Investors were wiped out. By any financial measure, it was a bubble.

And yet: the fiber is still in the ground. It didn’t disappear when the stock prices did. The physical infrastructure that was overbuilt in 1999 is a critical part of what the internet runs on today. Nobody looks at the connected world we now inhabit and concludes that the fiber optic build-out was a mistake. The financial bubble was real. The infrastructure was also real. They were two entirely different things.

The same distinction applies now.

Whether AI valuations are justified, whether any particular company is overpriced, whether a correction is coming — these are financial questions. The infrastructure being built is a physical question. Data centers don’t evaporate if the Nasdaq corrects. Chip fabrication facilities in Arizona don’t stop manufacturing if a valuation gets revised downward. The $500 billion committed to Stargate is not a financial instrument — it is construction. Steel, concrete, power systems, cooling infrastructure, servers. It will be built whether or not the stock price holds.

What infrastructure spending at this scale actually signals is a direction.

When the world’s largest technology organizations — with full visibility into capability trajectories that most operators cannot see — commit capital at this magnitude, they are not speculating. They are expressing a conviction about where the system is heading. The scale of the conviction is itself the data point.

There is one dimension of that direction worth noting separately. AI systems are increasingly being used to develop better AI systems. The infrastructure being built isn’t only serving current applications — it is enabling the next generation of capability, which will require and enable further infrastructure in turn. We are at the beginning of that curve, not the middle of it. Infrastructure tells us that plainly.

This is what private company leaders are actually looking at when they see these numbers. And what most are not yet reading in them.

The question isn’t whether to buy NVIDIA stock. The question is: what does $700 billion in annual infrastructure investment — committed by organizations with better forward visibility than almost anyone else — imply about the operating environment your business will inhabit in three to five years?

That is not a portfolio question. It is a positioning question.

The fiber went into the ground in 1999. The internet that ran on it arrived later. The companies that understood the direction — even through the crash — were positioned for what came next. The ones who let the Nasdaq correction convince them that nothing fundamental was changing were not.

The capital is committed. The direction is set. What you do with that information is still yours to determine.

Next: Adoption — Why It Feels Slower Than It Is