Downstream From the Data Center: An Industrial Boom—or Another Glut?

The AI data-center boom is driving investment in industrial equipment makers such as Caterpillar, Cummins, Eaton, Ford, Vertiv, Schneider Electric and Corning, with Caterpillar investing $725 million in large-engine production, Cummins projecting data-center sales to grow 80% to $9 billion by 2030, and Eaton's data-center sales rising from 14% of its business in 2023 to 21% in 2025. The risk is that manufacturers expand faster than AI demand materializes, potentially turning scarce infrastructure into commoditized capacity and squeezing margins, as seen in Corning's telecom fiber boom experience.

TL;DR — Key Takeaways - The AI data-center boom is creating demand far beyond construction, driving investment in generators, turbines, batteries, transformers, cooling equipment and optical fiber. - Companies including Caterpillar, Cummins, Eaton, Ford, Vertiv, Schneider Electric and Corning are expanding capacity to capitalize on AI infrastructure demand. - Corning’s experience during the telecom fiber boom shows that a technology can succeed spectacularly while suppliers still suffer from overcapacity, collapsing prices and delayed demand. - The biggest risk is that manufacturers expand faster than AI demand materializes, turning scarce infrastructure into commoditized capacity and squeezing margins. - The likely outcome is not simply boom or bust, but a sorting of suppliers based on diversification, financing, customer commitments and their ability to redirect production. When we talk about the economic benefits of AI data centers, we usually focus on two categories of jobs. There are the thousands of construction workers required to build these enormous facilities and the smaller number of permanent employees who operate them once they open. Those jobs matter. But they represent only the most visible part of a much larger industrial story. The AI data-center boom is creating demand far beyond the fence line. It is reaching into factories that manufacture generators, turbines, switchgear, batteries, cooling systems, transformers and optical fiber. Companies that were not ordinarily considered part of the technology industry are redirecting investment, expanding production and making acquisitions to feed an AI infrastructure supply chain that seems unable to get enough equipment. A recent Wall Street Journal article https://www.wsj.com/business/big-manufacturers-find-new-demand-in-equipping-ai-data-centers-14e869ee captures the scale of this pivot. Caterpillar, Cummins, Eaton and Ford are among the industrial companies racing to capture the money flowing into data-center construction. This is not speculative revenue projected to arrive sometime in the next decade. It is showing up in orders, backlogs, capital investments and factory expansions now. Caterpillar is investing $725 million https://investors.caterpillar.com/news/news-details/2025/Caterpillar-Invests-in-U-S--Manufacturing-and-Future-Workforce-Skills-Training/default.aspx to expand large-engine production as electricity generation becomes one of its most profitable businesses. Cummins expects its data-center sales to grow approximately 80% to $9 billion by 2030. Eaton’s data-center-related sales grew from 14% of its business in 2023 to 21% in 2025. Ford, facing excess electric-vehicle battery capacity, plans to invest $2 billion in a new energy-storage business, partly aimed at data centers. The pattern extends beyond those four companies. Vertiv is expanding production of AI-ready cooling equipment after reporting a 24% year-over-year increase in second-quarter sales https://investors.vertiv.com/news/news-details/2026/Vertiv-Reports-Strong-Second-Quarter-2026-with-Diluted-EPS-Growth-of-53-Adjusted-Diluted-EPS-Growth-of-60-Raises-Full-Year-2026-Guidance-Across-All-Key-Metrics/default.aspx . Schneider Electric says it has increased its data-center factory square footage by 270% in two years https://blog.se.com/datacenter/2026/07/29/fast-track-ai-growth-with-a-modular-data-center-approach/ . These are factories, industrial jobs, logistics networks and suppliers spread across communities that may be hundreds or thousands of miles from the facilities they will ultimately support. That is the broader economic footprint of the AI buildout. The data center may be where the capital is concentrated, but its effects are spreading through much of industrial America. The question is what happens after everyone expands. Selling Picks and Shovels The manufacturers feeding this boom will reasonably describe themselves as the picks-and-shovels providers of the AI gold rush. They do not have to predict whether OpenAI, Anthropic, Google, Meta or some company that does not yet exist ultimately discovers the richest vein. Every contender needs power, cooling, connectivity and electrical equipment. Selling picks and shovels appears safer than mining for gold. But that analogy provides less protection than it first appears. Selling picks and shovels is a wonderful business while picks and shovels remain scarce. When enough manufacturers expand production to serve the same gold rush, shovel production can outrun shovel demand. Buyers gain leverage, prices and margins contract, and factories built for the boom can become excess capacity. No company understands that risk better than Corning. Corning is one of the most compelling beneficiaries of the current AI infrastructure buildout. Meta has entered a multiyear agreement worth up to $6 billion https://investor.corning.com/news-and-events/news/news-details/2026/Corning-and-Meta-Announce-Multiyear-up-to-6-Billion-Agreement-to-Accelerate-US-Data-Center-Buildout/default.aspx for Corning to provide the optical fiber, cable and connectivity products needed for its U.S. data centers. The agreement includes a new cable-manufacturing facility in Hickory, North Carolina, along with expanded production across Corning’s other North Carolina operations. Corning also has a long-term partnership with NVIDIA https://investor.corning.com/news-and-events/news/news-details/2026/NVIDIA-and-Corning-Announce-Long-Term-Partnership-To-Strengthen-U-S--Manufacturing-for-AI-Infrastructure/default.aspx under which it plans to expand U.S. optical-connectivity manufacturing capacity tenfold and domestic fiber-production capacity by more than 50%. Thousands of GPUs inside an AI factory must constantly communicate with one another. That requires an extraordinary amount of high-performance optical connectivity. Corning is once again selling the glass that connects a technological revolution. It has done that before. When the Internet Left Corning Holding the Bag During the telecom boom of the late 1990s, investors believed that whoever controlled the physical pipes carrying internet traffic would own the future. Telecommunications companies raised hundreds of billions of dollars and raced to lay national and international fiber networks. Corning supplied the glass. The underlying prediction was correct. Internet traffic was going to explode. Bandwidth would become essential to practically every part of the economy. Corning’s stock climbed above $113 in 2000 as fiber represented approximately 40% of its revenue. Then the market discovered that telecommunications companies had installed far more capacity than customers could use at the time. The demand eventually came, but it did not come on the schedule required to support the debt, valuations and production capacity built around it. Fiber sitting unused in the ground became known as dark fiber. Carriers failed, capital spending collapsed and suppliers were left with factories designed for orders that were no longer arriving. Corning’s stock fell to approximately $1.10 by October 2002. Thousands of jobs disappeared, billions of dollars were written down, and the company mothballed its optical-fiber facility in Concord, North Carolina. According to Corning’s SEC filings https://www.sec.gov/Archives/edgar/data/24741/000119312510027165/d10k.htm , part of that plant did not reopen until 2007. The important point is not that the internet failed. It succeeded beyond practically every expectation. The cheap bandwidth created by the fiber glut became part of the foundation for cloud computing, streaming video, social media, mobile applications, software-as-a-service and the digital economy. The companies operating above the fiber layer built some of the most valuable enterprises in history. Corning helped make all of that possible, but it did not capture the greatest fortune created by the revolution it enabled. Being right about the technology was not the same as being right about the investment cycle. The Indispensability Trap Comes to the Factory This is an industrial expression of what I call the Indispensability Trap in my forthcoming book, The Indispensability Trap: How Becoming Essential Caps the Fortune—From the Railroads to AI, due in September. The trap is that foundational infrastructure can become essential, support enormous demand and enable an economic revolution while competition, abundance, regulation and commoditization cap the returns of its builders. The greatest value eventually migrates to the businesses that use the cheap, abundant infrastructure rather than the companies that financed and manufactured it. The fiber experience contains an important lesson for the current AI debate. Demand for bandwidth was growing rapidly in 2003, even as bandwidth prices fell roughly 90%. Demand growth and price compression were not opposites. In a commodity market, they were part of the same process. Falling prices encouraged more consumption, while the prospect of growing consumption attracted more capital and capacity. The AI infrastructure bulls may therefore be correct about aggregate demand. Enterprises, consumers and autonomous agents could consume vastly more intelligence than current forecasts contemplate. Every gigawatt being planned today may eventually find a workload. But “eventually” is not a sufficient business model for factories, employees or investors living quarter to quarter. The buildout can be rational in aggregate while ruining particular builders. Capital can concentrate in energy, foundries, data centers and industrial equipment while the durable margins migrate toward chips, networking, orchestration, proprietary data, applications, distribution and the companies that use inexpensive intelligence to reorganize actual work. Or, as I put it in the book: Capital in the ground floor, margin in the upper floors. The suppliers of AI infrastructure are not outside the wager. They are making a different wager. They are betting that aggregate demand will arrive quickly enough to absorb everything they are building before their once-scarce picks and shovels become commodities. Berkeley Lab’s latest data-center energy forecast https://eta.lbl.gov/publications/united-states-data-center-energy-2025 demonstrates the size of that uncertainty. Its central estimate has data centers consuming 11.8% of U.S. electricity by 2030, but its scenarios range from 9.5% to 15.3%. Both ends of that range represent enormous growth. They nevertheless imply materially different demand for generators, turbines, fiber, batteries, switchgear and cooling systems. Manufacturers must build factories and place equipment orders now without knowing where reality will land inside that range. They also cannot know how many announced data-center projects will obtain power, permits, financing and customers—or how much more efficient chips, cooling systems and AI models will become before those projects open. Corning’s current stock performance shows how sensitive investors already are to that uncertainty. When the Meta agreement was announced in January, Corning shares reached their highest intraday level since September 2000 https://www.wsj.com/livecoverage/stock-market-today-dow-sp-500-nasdaq-01-27-2026/card/corning-stock-jumps-on-meta-deal-for-data-center-cables-WNySfp7keYrFHztgIOLR . The historical symmetry was hard to miss. Then, in July, Corning shares fell more than 20% https://www.reuters.com/business/corning-falls-16-after-forecasting-weaker-than-expected-q3-sales-2026-07-28/ after the company issued a third-quarter sales forecast that came in only slightly below Wall Street expectations. The decline threatened to erase more than $23 billion in market value. The AI fiber boom had helped propel the stock upward, but the market demonstrated how little tolerance it had for even a modest suggestion that growth might arrive more slowly than expected. Not Every Supplier Faces the Same Risk None of this means another telecom-style collapse is inevitable. The fiber boom was especially destructive because many pure-play carriers financed construction with borrowed money that had to be repaid before customer demand arrived. When traffic failed to materialize on schedule, they had no profitable businesses to cushion the fall. Many of today’s industrial suppliers are better positioned. Caterpillar, Cummins and Eaton are diversified companies whose equipment can serve utilities, factories, hospitals, microgrids and other markets. Generators, power-management systems and batteries can be redeployed. Ford is using capacity originally built for electric vehicles rather than starting entirely from zero. Corning’s multiyear agreements with Meta and Nvidia provide more visibility than speculative orders from debt-fueled telecom startups did 25 years ago. That does not eliminate the risk. It distributes it unevenly. Diversified manufacturers may be able to redirect capacity if AI data-center demand slows. Companies making highly specialized cooling equipment, prefabricated data-center systems or components designed around a particular architecture may have fewer alternatives. Some suppliers could experience a sharp correction, others a long period of margin pressure, and the strongest manufacturers may absorb the slowdown by serving the broader electrification and grid-modernization markets. The likely outcome is not a simple choice between endless boom and spectacular bust. It is a sorting process based on financing, diversification, customer commitments and how easily production can be redirected. The industrial benefits occurring now are real. AI infrastructure is creating jobs, reviving factories and generating demand across a supply chain much greater than the data-center industry itself. Much of the equipment being produced will remain useful regardless of which AI companies ultimately win. But the history of Corning reminds us that essential infrastructure can be both economically transformative and financially unforgiving. Twenty-five years ago, Corning helped produce the abundance that made bandwidth cheap enough to build the modern internet. It was indispensable to the revolution, yet the greatest fortunes migrated to companies operating above its glass. Now Corning is expanding again to supply the AI economy. So are Caterpillar, Cummins, Eaton, Ford, Vertiv, Schneider Electric and dozens of companies throughout their supply chains. They are all selling picks and shovels into a gold rush that may be larger than anything we have seen before. They are also betting that this time demand will arrive before abundance does.