NextFin News - The companies that build the power, cooling, and electrical systems behind the artificial-intelligence data-center boom are doing something unusual: they are preparing for it to end. While Alphabet, Amazon, Meta Platforms, and Microsoft commit more than $650 billion in capital spending for 2026, the suppliers feeding the build-out — Siemens, Schneider Electric, Vertiv, Xylem — are quietly diversifying away from their most profitable customers, hedging against the day the AI capex cycle turns.
The Boom Is Real, but the Order Book Is the Warning Light
The scale of the spending is hard to overstate. The four largest hyperscalers are expected to spend more than $650 billion in 2026 expanding AI capacity, according to compiled market data. JPMorgan Global Research has raised its estimate for debt financing tied to the AI build-out to $4.1 trillion and expects hyperscaler capital expenditures to reach $650 billion this year before exceeding $1.1 trillion in 2027. U.S. hyperscalers spent an estimated $400 billion in 2025, with spending this year projected to approach $700 billion, per CoBank's Knowledge Exchange.
Yet the physical build is failing to keep pace with the announcements. Close to half of planned U.S. data center builds in 2026 are projected to be delayed or canceled, held back by shortages of transformers, switchgear, and batteries, and by grid interconnection queues that stretch for years. A Sightline Climate report found that power constraints and grid equipment shortages could delay 30% to 50% of data center projects this year. The trend is not new: of 110 projects planned to come online last year, 26% were delayed and 10% quietly revised their target operating dates. S&P Global projects data center power demand will more than double between 2026 and 2030, with almost 50 gigawatts of capacity expected on the grid by 2028 — a build rate that strains utilities already facing political opposition. Projects worth at least $156 billion were blocked or delayed in 2025 alone.
The suppliers sit at the narrowest point of this bottleneck, and they see the slowdown first. Their response — spreading revenue across industrial, utility, and non-AI customers rather than riding the hyperscaler wave — is the clearest signal yet that the trade is being managed as a cycle, not a permanent regime.
Why Suppliers Are the Canary in the AI Coal Mine
The transmission mechanism runs through the order book, not the earnings headline. A chip designer like Nvidia can report record revenue while the physical infrastructure chain is already rolling over, because chip orders are placed 12 to 18 months before a facility opens. Electrical equipment, cooling systems, and power distribution gear, by contrast, are ordered when construction actually starts. When interconnection queues stall or financing tightens, the first line items cut are the ones furthest from the shovel-ready site.
That is why supplier behavior matters more than hyperscaler guidance right now. Siemens said data-center demand pushed its revenue up more than a third in the quarter through December, with CEO Roland Busch stating that demand "has considerably exceeded our expectations." Schneider Electric narrowly beat first-quarter revenue expectations on the strength of the global AI data-center buildout, reinforcing its role as one of the most sought-after suppliers, and expanded its liquid-cooling portfolio through the acquisition of U.S. specialist Motivair. Vertiv, whose data-center exposure accounts for roughly 80% of sales, saw its shares climb 64% in 2026 alone.
"Despite the unprecedented amount of capital being deployed and the inherent risks associated with a high degree of market concentration, numerous signs suggest AI infrastructure spending has a long runway."
That was Jeff Johnston, lead digital infrastructure economist at CoBank, making the bull case. But strong current quarters are not the point. The point is what these companies are doing next. Each is using today's AI-driven margins to rebuild the parts of the business that do not depend on a handful of cloud giants. That is rational risk management — and it is also a confession that the customer base is dangerously concentrated. When your largest customers can collectively cut spending by hundreds of billions with a single board decision, diversification is not growth strategy; it is survival strategy.
Cyclical Wave or Structural Shift? The Verdict Is Both — and That Is the Problem
This is the judgment the market keeps dodging, and it determines everything. The AI data-center build-out contains two distinct forces that the market is pricing as one.
The cyclical leg is the hyperscaler capex surge itself. It meets every test for a mean-reverting cycle: it is front-loaded (capacity is being built ahead of demand, not behind it), it is concentrated (four buyers account for the bulk of spending), and it is already showing the classic symptoms of overbuild — delays, component shortages, and a widening gap between spending and revenue. AI-related services are expected to deliver only about $25 billion in revenue in 2025, roughly 10% of what hyperscalers are spending on infrastructure. History offers the comparison: the fiber-optic build-out of the late 1990s saw capital pouring into dark fiber that generated little near-term cash flow. Telecom carriers ordered equipment years in advance, and when revenue failed to materialize, capex collapsed by more than half between 2000 and 2002. The suppliers who survived were the ones with broad industrial customer bases — the ones who could sell the same switches and power gear to factories, hospitals, and utilities — not the pure plays most levered to the boom.
The structural leg is different and will outlast the cycle. Data centers are becoming a permanent, larger share of the electricity grid and industrial equipment demand. Even if hyperscaler spending normalizes, the baseline level of data-center power demand will remain far above the pre-AI era — S&P's forecast of demand more than doubling by 2030 does not require the boom to continue at its current pace. Electrification, grid modernization, and industrial automation are independent, durable demand streams that suppliers like Siemens, Schneider, and Eaton can lean on.
The market's error is blending the two. It has rewarded pure-play data-center suppliers as if the cyclical surge were permanent, while the suppliers themselves are pricing in the opposite. The divergence between stock-market enthusiasm and corporate hedging is the story.
The Second-Order Question Nobody Is Asking: Who Pays When the ROI Gap Widens?
The first-order consequence of a capex slowdown is obvious: supplier revenue falls. The second-order effect is more dangerous, and it is not priced in. If hyperscalers slow spending because AI revenue fails to materialize, the loss does not stop at the equipment vendors. It travels up the capital stack.
JPMorgan's estimate of $4.1 trillion in debt financing tied to the AI build-out means banks, private credit funds, and bondholders have significant exposure to projects whose economics depend on continuous hyperscaler demand. A delay is not merely a timing issue when debt service is due. The recent wave of circular financing — in which chipmakers invest in AI companies that then use the cash to buy the chipmakers' products — has amplified the leverage in the system. Nvidia's agreement to invest up to $100 billion in OpenAI, with the two companies signing a letter of intent to deploy at least 10 gigawatts of Nvidia systems, is the most visible example. Jensen Huang framed the partnership as the continuation of a decade-long collaboration:
"Nvidia and OpenAI have pushed each other for a decade, from the first DGX supercomputer to the breakthrough of ChatGPT."
But when financing and demand are linked within the same ecosystem, a slowdown in one corner reverberates through all of them. The $100 billion is not a grant; it becomes revenue for Nvidia only if OpenAI's data centers generate enough cash flow to keep buying chips. That is the circularity the market has celebrated as validation. It is also the feedback loop that can turn a demand slowdown into a credit event.
This is why investors have already begun rotating within the AI trade — trimming semiconductor exposure in favor of software companies and the cloud providers themselves. The rotation is visible in the market's own pricing: semiconductor and equipment names tied to facility construction have underperformed the cloud giants as capex concerns mounted, reflecting a bet that the hyperscalers — not their suppliers — will capture the AI revenue when it finally arrives.
The Counter-Thesis: This Time, the Spending Cannot Stop
The strongest argument against the bust case is not optimism — it is compulsion. Hyperscalers cannot slow spending without losing the AI war. If one player pauses while rivals keep building, it cedes model quality, cloud share, and enterprise contracts that are nearly impossible to win back. CoBank's Jeff Johnston put it directly: "numerous signs suggest AI infrastructure spending has a long runway." OpenAI reported revenue of about $2 billion per month earlier this year, up from an estimated $2 billion for all of 2023 — a growth curve that, if sustained, would justify continued infrastructure investment.
This argument is real, and it explains why the bust has not happened yet. But it confuses a reason to keep spending with a reason for suppliers to stay concentrated. Even if hyperscalers never stop building, the pace of build-out is what suppliers are hedging against — and the pace is already slowing as power constraints bite. The counter-thesis also rests on a fragile assumption: that revenue growth translates into profits large enough to service the capital stack. It does not, not yet. Spending $650 billion to generate $25 billion in AI services revenue is a 4% cash yield on invested capital before financing costs — a return profile that cannot survive a rise in borrowing costs or a pause in order growth.
The falsifying signal is specific: if hyperscaler capital expenditures grow less than 15% in 2027 — well below the roughly 70% pace the market has priced — while AI services revenue remains under 20% of infrastructure spending, the cyclical-overbuild thesis is confirmed and supplier multiples should compress. Conversely, if AI revenue catches up to spending within four quarters, the structural-shift case wins and today's supplier valuations look cheap.
What Comes Next: Beneficiaries, the Exposed, and the Watch List
In the short term, sentiment will swing on hyperscaler guidance and any sign of capex discipline. Stocks most levered to pure data-center demand — the cooling and power-equipment names that derive the majority of revenue from hyperscale AI builds — face the widest multiple compression if spending slows. The exposed also include memory and semiconductor-equipment makers whose orders track facility construction rather than end demand. Vertiv's 80% data-center concentration, which looked like a premium during the boom, becomes a liability the moment hyperscaler orders soften.
The medium-term beneficiaries are the diversified industrial-electrical conglomerates that used AI margins to rebuild non-data-center booklets. Siemens, Schneider Electric, and Eaton can fall back on grid modernization, factory electrification, and utility capex if hyperscaler orders soften. Utilities and independent power producers with secured interconnection queues hold a different kind of optionality: in a world where 30% to 50% of projects face delays, the ones with power already in hand become the bottleneck owners.
Long term, the structural demand for data-center power and cooling is intact regardless of the capex cycle. The winners will be the companies that treated AI as one large customer segment among several, not as the entire business model.
Watch three signals. First, hyperscaler capital expenditures versus the 15% growth threshold — that measures demand. Second, the share of data-center projects reaching commercial operation on schedule — that measures physical feasibility, and it is where the power and equipment shortages will show up first. Third, AI services revenue as a percentage of infrastructure spending — that measures whether the economics ever close. A fourth, quieter signal is supplier order-to-book ratios: when the canary stops singing, the order book tells you before the earnings call does.
The data-center boom will leave a larger grid and more computing capacity behind it — but the suppliers betting their valuations on the boom never slowing are making a cyclical wager and calling it structural. The order books already know the difference.
Explore more exclusive insights at nextfin.ai.
