Although Wall Street has spent the past three years arguing about whether AI capex is a bubble, one figure from Nvidia CEO Jensen Huang makes it all pretty clear.
A single 500-megawatt data center now requires 30,000 truckloads to build, and that figure does not include the separate power plant needed to feed it. Huang calls this “the single largest infrastructure buildout in human history.” But the more interesting question is what history actually says happens to the companies that supply buildouts at that scale.
The numbers behind the rhetoric are wild.
NVIDIA (NASDAQ:NVDA | NVDA Price Prediction) closed Q4 FY2026 with $68.13 billion in revenue, up 73.2% YoY, with Data Center revenue of $62.31 billion and networking revenue up 263% YoY.
Corning (NYSE:GLW), the picks-and-shovels fiber supplier inside those data centers, reported Optical Communications revenue of $1.846 billion, up 36% YoY. I have owned Nvidia for over 15 years now, and I have studied this kind of supplier setup long enough to recognize the shape of it.
What’s particularly notable is that Huang describes the moment as an opportunity to “revitalize American manufacturing for the first time in several generations.” That echoes the American railroad buildout of the late 1800s, when private capital funded a multi-decade physical project that reshaped which companies dominated the US economy for generations.
The Pattern History Keeps Repeating
Every great American buildout has followed the same five-step pattern.
Private capital floods a physical project at unprecedented scale. Supply-chain bottlenecks emerge first, so materials suppliers benefit early. Domestic manufacturing capacity expands to meet sustained demand. Some operators overbuild and fail when the demand ramp lags projections. Surviving infrastructure underpins the next economic cycle for decades.
Are you invested enough in this cycle?
The rails minted Carnegie’s steel empire. Electrification minted General Electric and Westinghouse. The interstate system minted the cement and equipment giants. In each case, the picks-and-shovels suppliers compounded for years before the headline operators sorted themselves out.
The Modern Suppliers
Corning is the closest modern analog to those original suppliers. It holds over 70% market share in specialized glass for AI data centers, its new facilities are creating 3,000 manufacturing jobs, and the company has now closed an up-to-$6 billion multiyear deal with Meta plus two more large long-term agreements with hyperscale customers of similar size and duration. CEO Wendell Weeks told analysts that “these agreements taken in total are driving so much growth, John, that you’re going to see expansion across all of our major optical operations.”
Huang has obvious skin in the game. He runs the company that sells the picks. The supply commitments back the story anyway: $95.2 billion in supply-related commitments, partnerships with OpenAI for at least 10 gigawatts of NVIDIA systems, CoreWeave for more than 5 gigawatts of AI factories by 2030, and South Korea for over 250,000 NVIDIA GPUs. Crucially, the buildout is funded by “private cash-flushed enterprises,” not federal debt, with construction unions “leading the charge” across the Rust Belt and South.
Nathaniel Whittemore of The AI Daily Brief frames the runway: enterprises are only “5 to 20%” fully onboarded into agentic systems today, and at “60 or 70%” the compute shortage becomes “monumental.” He calls the whole project “a sustained, likely decades-long project to get access to the compute that we are going to need for the next phase of the global economy.”
The Cautionary Mirror
The honest bear case is Corning’s own past. The late-1990s fiber-optic buildout was directionally correct. The internet did need that fiber. The timing was disastrous.
Many operators went bankrupt before demand caught up, and Corning itself peaked in 2000 before collapsing in the fiber glut. Today’s buildout is funded by cash-flush hyperscalers rather than debt-loaded telecoms, but the pattern of supply leading demand still applies. If enterprise AI adoption stalls below that 60 to 70% threshold, the buildout will outpace usage.
The valuation math already reflects some timing risk. Corning trades at roughly 59x forward earnings, a multiple that historically sat far lower for a specialty materials company. NVIDIA changes hands at $225.83, up roughly 74% over one year and about 1,490% over five. Polymarket assigns a 99% probability that Q1 FY2027 Data Center revenue clears $60 billion. The crowd is already paying for execution today.
Long term, Wall Street still heads higher in the decades to come, and the surviving suppliers of every great American buildout went on to dominate their industries for generations.
You might lean into the supplier thesis if you believe agentic AI adoption keeps climbing toward Whittemore’s 60 to 70% line.
Or if you think enterprise demand stalls, the 1990s fiber glut is the chapter to reread. Huang’s 30,000 truckloads tell us this buildout is real. And it is spectacular.
History tells us how it usually ends. Suppliers win early. Operators get sorted out late. The infrastructure quietly powers the next era.
Top Gaining Stocks
Top Losing Stocks