The NTT Lesson for AI Stocks

Every new technology wave invites a favorite pastime: historical analogy. Railways, electrification, the internet—each era had its own version of the claim that “this time is different.” Today, the blockbuster tech stocks driving AI, cloud computing, and space infrastructure are being called the safest bets of a generation. But the most revealing comparison is the one most analysts ignore: Nippon Telegraph and Telephone’s IPO in February 1987.

NTT was the largest company on earth by market cap at its peak. It was a state-sanctioned monopoly with a guaranteed revenue stream, owning the entire Japanese telecom infrastructure. It was seen as utterly safe. Yet the stock lost 88% of its value over the next five years and did not regain its IPO price for nearly two decades. The reason was not a business failure—NTT remained profitable. The reason was a structural mismatch between investor expectations and the economic reality of capital-intensive infrastructure. Today’s AI infrastructure stocks—the hyperscalers, chip makers, and data center landlords—face the same trap.

The Overlooked Angle

The question of safety for today’s blockbuster tech stocks is usually framed around competitive moats, revenue growth, or regulatory risk. All of these miss the real danger: the brutal arithmetic of high fixed costs, long depreciation cycles, and diminishing returns on invested capital. The overlooked angle is that many of these companies are not true growth stocks; they are capital-intensive utilities dressed in technology clothing. The NTT story was not a one-off anomaly. It was a textbook example of what happens when a market misprices the risk of infrastructure assets during a hype cycle.

Why This Small Detail Matters

The detail that matters is the ratio of capital expenditure to revenue. For a software company, a 10% revenue growth can fall to 5% without destroying the business. For a company that must spend 40% of its revenue on equipment that becomes obsolete in five years, a growth slowdown is fatal. The cash flow that looked bountiful during expansion becomes a trickle when replacement capex eats into profits. NTT’s telephone network required constant reinvestment. Its monopoly status did not save it from the need to pour capital into copper lines just as fiber was emerging. Today’s data center operators and AI chip designers face a similar treadmill: every new generation of GPUs demands a new wave of spending, and the depreciation clock ticks faster than revenue can adjust.

The Economic Mechanism

Let’s break down the economic mechanism using a framework that strips away the narrative. The key measure is return on invested capital (ROIC). For a business to be a safe long-term investment, its ROIC must exceed its cost of capital consistently. A capital-intensive infrastructure business has a high denominator—lots of assets—and a numerator that depends on pricing power and utilization. When growth is high, utilization is high, and the business appears to generate attractive returns. But the moment demand growth slows, or when obsolescence forces premature replacement of assets, ROIC collapses.

Consider a typical AI data center: $1 billion of capital for a single facility. Expected life of equipment: five years. Straight-line depreciation: $200 million per year. If revenue from that data center is $300 million per year, operating profit is $100 million—a 10% return on capital. That is below the cost of capital for many of these firms, especially when you account for debt financing. The only way to boost returns is to lever up or to price aggressively. But pricing power is constrained by competition—cloud providers are waging a price war, and chip makers are under constant pressure from customers like Microsoft and Amazon who design their own silicon. The result is a race to the bottom where every player must invest more just to stay relevant, eroding the returns that attracted investors in the first place.

NTT faced the same dynamic. Its telephone network required billions in investment. Revenue grew as Japan’s economy boomed, but the growth was linear, not exponential. The stock price, however, had priced in exponential expectations. When the bubble burst, the market realized that NTT was not a tech miracle—it was a regulated utility with a finite growth ceiling. The same realization awaits today’s AI infrastructure behemoths once the narrative shifts from “AI will change everything” to “AI will require a decade of heavy capital spending with uncertain payoffs.”

The Strategic Consequence

Who benefits from this mispricing? The early investors and the management teams that time their equity issuance at the peak of the hype cycle. NTT’s IPO allowed the Japanese government to sell shares at an inflated price. Today, companies like Arm, Palantir, and even data center REITs have used the AI frenzy to raise capital at valuations that assume perpetual hypergrowth. The insiders who sell at those prices lock in gains. The late buyers—pension funds, retail investors, and indexers—absorb the risk.

Who loses? Anyone who holds for the long term. The strategic consequence is a wealth transfer from patient capital to opportunistic sellers. The companies themselves are not necessarily doomed—they will survive, just like NTT survived. But the stock will deliver returns far below the market’s expectation. The lesson for institutional investors is painful: if the business model requires constant reinvestment of all free cash flow just to maintain the asset base, the equity is essentially a leveraged play on the cost of capital. When that cost rises, the stock craters.

What Most Commentary Gets Wrong

Most commentary on the safety of today’s tech stocks focuses on monopoly power, network effects, or the inevitability of AI adoption. These are true as descriptive statements but irrelevant for valuation. The NTT analogy shows that even a perfect monopoly can be a terrible investment if the underlying economics are capital-intensive and the growth story is overdone. The mainstream narrative says: “These companies are building the future, so they must be valuable.” The hard truth is that building the future is expensive, and the companies that build it often do not capture the value—their suppliers and customers do.

For example, Nvidia makes the GPUs that power AI. Its revenue surged, but its cost of sales also surged as it ramped up production. The chip industry is notoriously cyclical, and every boom leads to overcapacity. The same will happen in data centers: once the AI buildout matures, utilization rates will fall, and the price of compute will drop. The companies that own the pipes will be left with massive depreciation charges and no pricing power. The analogy is not to Microsoft’s software margins but to telecom carrier margins after the dot-com bust.

The Hard Business Lesson

There is a simple but brutal business lesson in the NTT story: do not confuse size with safety. NTT was the largest company in the world, and it was a disaster. Today’s blockbuster tech stocks are large, but large is not a moat. The real moat is a low need for reinvestment—software margins, subscription revenue, and customer lock-in. Infrastructure assets do not have that moat. They are commoditized by the very capital that builds them.

For investors, the practical takeaway is to examine free cash flow after all maintenance capex. If a company’s cash flow is positive only because it is growing and delaying replacement spending, that is a red flag. The safety investors assign to these stocks is an illusion created by a bull market narrative. The next time you hear that a tech IPO is a safe bet because it owns critical infrastructure, remember NTT. The telephone monopoly was as safe as they come, and it lost investors a fortune. The only safety in capital-intensive tech is to buy after the boom when the assets are cheap. But that is never where the story is.

The real truth is boring: blockbuster technology does not automatically make blockbuster stocks. The value flows to whoever controls the scarce resource. In today’s world, that scarce resource is not compute—it is attention, distribution, and data. The infrastructure providers are merely the vendors. Vendors do not get the richest multiples. The NTT lesson may be thirty-six years old, but it has never been more relevant.