NextFin News - Kevin Murphy, chief executive of Ferguson plc, delivered a pointed assessment of the U.S. artificial-intelligence buildout on Oct. 1, 2026: the boom has moved beyond chips and servers into the physical infrastructure that keeps data centers running, and the next phase of growth will be constrained less by demand than by the availability of electricity. Appearing on "The Close" with Romaine Bostick and Emily Graffeo as the final quarter of the year began, Murphy framed the open question for U.S. manufacturing: after emerging from a multiyear slump earlier in 2026 on several tailwinds, is the momentum still holding? For the largest plumbing and heating products distributor in North America, the answer is already visible in its order book.
Ferguson's own results show the AI buildout is now showing up in industrial distribution, and management has been willing to put capital behind that view. In the quarter ended June 30, 2026, the company reported net sales of $8.75 billion, up 4.6% year over year, with reported diluted earnings per share of $3.43, up 6.9%, and adjusted diluted EPS of $3.39, ahead of the $3.29 consensus. U.S. non-residential revenue grew 8%, HVAC revenue rose 11%, and industrial revenue jumped 18% — a compositional shift toward the large capital projects that data centers, chip fabs, and power-generation facilities represent. Management raised its full-year 2026 revenue outlook to mid-single-digit growth and lifted the low end of its adjusted operating margin guidance to a range of 9.5% to 9.8%.
The AI Buildout's Second Act Is Industrial, Not Digital
The first-order story of the AI boom is familiar: hyperscalers buy GPUs, build data halls, and race for model leadership. The second-order story — the one Ferguson's results are already reflecting — is that every rack of AI silicon requires a supporting industrial ecosystem that is far less glamorous and far more capital-intensive than the chips themselves.
Data centers need cooling equipment, piping, valves, pumps, HVAC systems, fire suppression products, and water infrastructure. These are not optional add-ons; they are the mechanical and plumbing systems without which a data center cannot operate. As AI workloads push power density per rack higher, the cooling challenge compounds: liquid cooling loops, chilled-water plants, and specialized flow-control equipment move from edge cases to standard specifications. A distributor with deep relationships in these categories, local branch coverage, and the ability to source and stage materials for large projects becomes a bottleneck-adjacent supplier — the kind of business that gets paid regardless of which model wins the intelligence race.
Ferguson's segment results map directly onto this thesis. Industrial revenue up 18% and non-residential up 8% in a quarter where overall sales grew 4.6% is not a broad-based cyclical recovery; it is a rotation toward the large capital projects that define the AI infrastructure layer. Chief Financial Officer Bill Brundage put it plainly on the company's second-quarter call, saying large capital projects "will build into the future and will be a tailwind over the next couple of years."
This is the mechanism behind Murphy's manufacturing comment. The tailwinds that pulled U.S. manufacturing out of its multiyear slump — infrastructure spending, onshoring of industrial capacity, and the data center buildout — are not evenly distributed. They concentrate in the industrial and non-residential end markets where Ferguson has been repositioning for years, pivoting away from the more cyclical residential repair-and-remodel exposure toward water infrastructure, HVAC, and large capital projects.
The company has put capital behind the view. In August 2026, Ferguson agreed to acquire FloWorks, an industrial distributor of technical valves and flow-control solutions, for an enterprise value of approximately $1.6 billion — roughly 10 times trailing adjusted EBITDA including expected synergies of about $45 million. FloWorks generated approximately $1 billion in revenue in 2025 and operates more than 60 locations across the U.S. and Canada, serving chemical, refining, power generation, semiconductor, pharmaceutical, and data center markets. The deal, expected to close in Ferguson's third quarter, lifts the company's total addressable market to approximately $400 billion from $340 billion. It is the eighth acquisition Ferguson has announced in 2026, a group representing roughly $1.4 billion in combined annualized revenue.
"FloWorks strengthens our leading position in high-growth industrial end markets, while adding meaningful capabilities and geographic coverage which we can leverage across our non-residential customer groups," said Kevin Murphy, chief executive of Ferguson. "Their expert teams, technical capabilities and strong OEM brands will further enhance our ability to provide essential water solutions for the specialized professional. We welcome their associates to Ferguson and look forward to our next chapter of growth together."
The stated rationale extends the logic: the acquisition strengthens Ferguson's position in the build-out of North American infrastructure around data centers, chip production, power generation, water, and also pharma and biotechnology.
The Binding Constraint Has Shifted From GPUs to Grid Power
Here is where Murphy's framing of the AI buildout gets consequential. The first bottleneck of the AI era was semiconductor supply — companies could not get enough GPUs. That constraint has loosened relative to demand. The binding constraint in 2026 is power.
Goldman Sachs Research projects U.S. data center power demand will climb from 31 gigawatts in 2025 to 41 GW in 2026 and 66 GW in 2027, with scheduled capacity additions reaching 13.6 GW in 2026 and 36.3 GW in 2027 — compared with realized additions of just 6.4 GW in 2024 and 8.5 GW in 2025. The share of U.S. data centers in total peak summer power demand is projected to rise from 4.1% in 2025 to 5.3% in 2026 and 8.5% in 2027. The International Energy Agency projects global electricity consumption by data centers will more than double by 2030, reaching approximately 945 terawatt-hours, with AI workloads the primary driver. McKinsey & Company forecasts U.S. data center load will rise from 25 GW in 2024 to more than 80 GW by 2030.
The grid is not keeping pace. Analysis of grid interconnection data has found waits in the largest U.S. data center markets — Northern Virginia, Phoenix, and Dallas — running four to seven years, and roughly half of planned U.S. AI data center builds in 2026 delayed or canceled because grid power is not available on schedule. Only about 3% of expected 2026 U.S. data center projects are designed for onsite power only; the remainder depend on utility interconnection, which means the utility queue, not the operator's balance sheet, sets the schedule. Goldman Sachs Research estimates approximately $720 billion in grid investment will be needed over the decade to accommodate the load growth.
This is a structural shift in the nature of the constraint, and it changes who benefits. When GPUs were scarce, value accrued to semiconductor designers and foundries. When grid capacity is scarce, value migrates toward the companies that can help customers secure, generate, and manage power: onsite generation, heavy electrical equipment, switchgear, transformers, cooling systems, and the flow-control and piping systems that connect them. Ferguson's FloWorks acquisition — with its exposure to power generation, semiconductor, and refining — is a direct read on this migration.
The capital intensity backs this up. Creditsights projects the top five hyperscalers will spend approximately $602 billion on capital expenditure in 2026, up 36% year over year, following roughly $443 billion in 2025 and $256 billion in 2024. Other industry estimates put the top five's 2026 data center infrastructure spend between $775 billion and $800 billion. Global data center capital expenditure is forecast to grow from approximately $434 billion in 2024 to more than $1 trillion by 2029, with some projections reaching a cumulative $1.7 trillion by 2030. This is the largest corporate capital expenditure cycle in recorded history, and a meaningful share of it is now flowing into the physical plant rather than the IT equipment.
Cyclical or Structural? Both — and They Must Be Separated
The critical judgment for investors is whether the AI buildout is a cyclical wave that will mean-revert or a structural regime shift that will not correct on its own. The answer is both, and conflating them produces the wrong conclusion.
The structural leg is real. The build-out of data center capacity to support AI workloads is a multi-year reallocation of corporate capital driven by a technological regime change, not a demand pulse. The evidence: hyperscaler capex is being directed at long-lived physical assets; the addressable market for industrial distributors serving these projects is being permanently re-rated — Ferguson's own acquisition math lifts its TAM by $60 billion; and the power constraint is a physical, not financial, bottleneck that cannot be resolved by a change in sentiment. A regime change in the underlying demand for compute, backed by permanent additions to grid and generation capacity, is structural by definition.
The cyclical leg is equally real and equally dangerous to ignore. Hyperscaler spending is front-loaded relative to revenue monetization, and any disappointment in AI-driven earnings would trigger a capex review. Inventory builds — Murphy has said inventory supporting large capital projects will continue to increase as Ferguson's backlog grows — create a double-edged dynamic: rising inventory amplifies growth on the way up and forces destocking on the way down. And the industrial distributors' exposure is leveraged to project timing: a data center delayed by a multi-year interconnection queue does not cancel its order, but it does defer revenue recognition and compress near-term growth rates.
So the correct read is a structural supercycle with a cyclical cadence. The direction is up for the physical infrastructure layer over a multi-year horizon; the path is lumpy, gated by power availability and hyperscaler capital discipline.
The Counter-Thesis, and What Would Prove It Right
The strongest case against this view is the bubble argument: hyperscalers are spending hundreds of billions of dollars on AI infrastructure ahead of any commensurate revenue, and when monetization disappoints, capital expenditure will be cut sharply. History offers a template — the fiber-optic buildout of the late 1990s left vast amounts of dark fiber and devastated the suppliers that had scaled up to serve it. If AI revenue fails to catch up to capex, the industrial distributors that positioned themselves as picks-and-shovels beneficiaries could face the same destocking cycle that hit telecom equipment vendors after 2001.
The argument has force. It is backed by the observable fact that hyperscaler capex growth of 36% to 73% year over year far exceeds the growth in cloud and AI revenue, and by the memory of previous infrastructure bubbles. It attacks the core thesis at its foundation: if the spending stops, the physical-infrastructure trade was never structural, only cyclical.
But the counter-thesis misses a key difference. The fiber bubble was financed by speculative carriers building capacity in search of demand that never arrived. Today's AI buildout is funded by the largest, most cash-generative companies in the world, spending on capacity that is already committed and largely leased or utilized by their own AI and cloud businesses before construction completes. The demand is pulled by existing workloads, not pushed by speculation. That does not make the trade risk-free — it makes the downside a growth deceleration, not a collapse into zero.
The falsifying signal is quantifiable: if the top hyperscalers' combined capital expenditure growth decelerates below roughly 20% year over year for two consecutive quarters, or if data center utilization and leasing rates show material deterioration while vacancy rises, the structural thesis for the physical-infrastructure layer is wrong and the cyclical destocking case takes over. Until that signal prints, the weight of evidence favors a multi-year buildout with power as the gating factor.
What Comes Next: Scenarios and Signals
The implications split cleanly by time horizon.
Short term (next one to two quarters): Sentiment and inventory drive the tape. Ferguson's guidance of mid-single-digit revenue growth and 9.5% to 9.8% adjusted operating margin for 2026 assumes second-half momentum holds. The risk is a pause in large-project bidding if power delays push construction schedules out; the opportunity is continued outperformance in industrial and HVAC as backlog converts to revenue. The company also expects to remain within its targeted net debt to adjusted EBITDA range of one to two times after the FloWorks closing, and has said it will resume buybacks when leverage moves back toward the lower end of that range.
Medium term (one to three years): Fundamentals dominate. The FloWorks integration, the conversion of a growing backlog, and the continued shift of hyperscaler spending into physical plant should support above-market growth for distributors positioned in data center-adjacent categories. Companies with scale, local execution, and M&A capability — Ferguson's stated playbook — are best placed to capture share as the market consolidates around capable suppliers.
Long term (three years and beyond): The structural question resolves. If the power grid expands — through transmission investment, onsite generation, or regulatory reform of interconnection queues — the buildout extends into a decade-long cycle. If power remains constrained, growth compresses into a shorter, more intense burst followed by a plateau. Either way, the companies that own the relationships and distribution networks in cooling, flow control, and power-adjacent industrial products will have been permanently re-rated relative to the pre-AI era.
Scenarios:
- Base case: Hyperscaler capex grows in the mid-20s to low-30s percent range through 2027; power constraints delay but do not cancel projects; industrial distributors grow mid-to-high single digits with margin expansion from mix shift toward large capital projects.
- Upside case: Interconnection reform or a surge in onsite generation unlocks the delayed pipeline; a wave of deferred projects hits the market simultaneously, producing a multi-year order surge for flow control, cooling, and power equipment suppliers.
- Downside case: AI monetization disappoints, hyperscaler capex growth falls below 20%, and the delayed-project pipeline shrinks rather than converts; distributors face destocking and multiple compression back toward pre-AI valuation levels.
What to watch: quarterly hyperscaler capex disclosures, grid interconnection queue data from independent trackers, Ferguson's large-capital-project commentary and backlog conversion, and any shift in inventory days that would signal a change from accumulation to destocking.
The bottom line: Murphy's message is that the AI buildout is no longer a technology story — it is an infrastructure story, and infrastructure is built with industrial products, funded by the world's largest balance sheets, and gated by a power grid that has not kept pace. The investors who treat it as a chip trade are looking at the last cycle; the ones who understand it as a physical-constraint trade are positioned for the next one.
更多独家洞察尽在 nextfin.ai.
