Beyond the GPU: AI's Real Bottleneck Is Power and Cooling Infrastructure

The dominant signal in this claim set is not a new NVIDIA-specific operating datapoint, but the accelerating construction of the infrastructure required to deploy AI compute at scale. Across power distribution, cooling, networking, optical connectivity, manufacturing, testing, and grid infrastructure, demand is translating into measurable orders, revenue growth, expanding backlogs, and new production capacity.

Consider the circuit. Vertiv provides power, UPS systems, thermal management, modular infrastructure, controls, and services for mission-critical environments ^2,5,12,17. Its portfolio is increasingly concentrated in liquid cooling, high-capacity power systems, and advanced thermal control—equipment becoming technically necessary as rack density rises ⁵.

For NVIDIA, this is ecosystem evidence rather than direct company-specific confirmation. The claims provide no direct NVIDIA revenue, order, margin, or product-update datapoint. They do, however, show that accelerated computing is creating requirements well beyond the GPU itself: more power delivery, greater thermal capacity, faster optical links, denser interconnects, and more integrated systems. The breadth of this demand supports the thesis that AI capital expenditure is producing a multi-layered infrastructure opportunity.

The evidence is recent, with most observations published between 28 July and 10 August 2026. Corroboration is strongest for Amphenol’s IT datacom growth ¹³, Schneider Electric’s organic performance ⁶⁴, Qnity’s interconnect growth ²⁵, Vishay’s point-of-sale momentum ³⁵, Vertiv’s guidance increase ^5,12, and Monolithic Power Systems’ data-center-related growth ¹⁶. The principal counterweight is valuation. Vertiv, Eaton, and Quanta have already experienced substantial multiple expansion, leaving AI-infrastructure beneficiaries exposed if growth, project timing, or hyperscaler spending disappoints ⁵.

Demand Is Broadening Across the Physical Data-Center Stack

The strongest recurring signal is that data-center demand is reaching the physical layer between electricity generation and usable computing capacity. Vertiv and Eaton are explicitly positioned in this layer ^11,61. Vertiv’s exposure includes UPS systems, power distribution, busway, thermal systems, rack systems, and integrated data-center infrastructure ^4,18.

Vertiv recorded 24% second-quarter growth ⁵, carried approximately $15 billion of backlog ⁵, and guided to approximately 35% organic growth in the third quarter ¹². Management described an accelerating commercial pipeline in both size and sales-cycle velocity across hyperscalers, colocation providers, NeoClouds, enterprises, products, and geographies ¹². That breadth matters: the demand is not confined to one customer class or deployment model.

Schneider Electric provides a broader corroborating read-through. Second-quarter organic revenue growth was 16.5% ⁶⁴, while first-half organic growth reached 14% ⁶⁴. Adjusted EBITA grew 22% organically ⁶⁴, and revenue reached a record level ⁶⁴. North America grew 23.1% organically, China and East Asia 19.7%, and South Asia & International 12.3% ^62,64. Data Center & Networks remained exceptionally strong, with strong double-digit sales growth ⁶⁴, while Products grew 13% organically, primarily through volume ⁶⁴.

The same pattern appears among component and connectivity suppliers:

• Amphenol’s IT datacom business achieved 63% organic growth, supported by 22% sequential growth and 94% year-over-year order growth ¹³. Orders increased 14% sequentially ¹³, and management expected another mid-teens sequential increase in the third quarter ¹³.
• Corning’s Optical Communications business generated $2.072 billion of second-quarter sales, up 32% ^8,51.
• Qnity’s Interconnect Solutions business grew 28% organically, compared with mid-to-high-single-digit growth for the broader printed-circuit-board market ²⁵.
• Flex reported growth of more than 70% in Power, 35% in CPI, 20% in ITS, and 12% in RMS ⁴.
• Celestica’s communications segment grew 62%, while its CCS segment grew 84% ⁹.
• Monolithic Power Systems’ Communications business grew approximately 80% year over year, while Enterprise Data grew 45% sequentially ¹⁶.

These observations identify the bottlenecks and investment opportunities surrounding accelerated computing: electrical distribution, optical links, interconnects, power conversion, thermal systems, and contract manufacturing. They also provide indirect support for continued demand for NVIDIA’s data-center platforms because suppliers across the same deployment chain are reporting unusually strong order and revenue growth.

Thermal Management and Power Density

Liquid cooling is the clearest technical theme. Vertiv is directly embedded in the data hall, where liquid cooling is becoming increasingly necessary ⁵. Its strategic exposure includes liquid cooling, high-capacity UPS equipment, and advanced thermal control ⁵. The company is vertically integrating and acquiring capabilities in server-side cooling, heat rejection, fluid management, and modular-frame construction ¹².

The acquisition of Strategic Thermal Labs strengthened cold-plate design, server-side liquid cooling, high-density thermal validation, and system-level thermal engineering ¹⁷. ThermoKey expanded facility-side capabilities ¹⁷. Together, these acquisitions support participation across nearly the full AI thermal architecture, from chip-level heat removal to facility heat rejection ¹⁷.

Vertiv is investing approximately $50 million to expand Ohio production of liquid cooling, chilled-water systems, and high-density thermal infrastructure ¹⁷. Its Tognana, Italy campus is intended to double regional chiller capacity by the end of 2026, and a new thermal-testing laboratory is expected to be completed in early 2027 ¹⁷. The Ironton, Ohio expansion is expected to be operational in the second quarter of 2027 ¹⁷. These are investments in structural increases in rack density, not merely in a temporary order surge.

The economics are changing with the density. Vertiv’s content opportunity was estimated at approximately $3.25 million–$3.75 million per megawatt, with management activity and product development continuing to corroborate that range ¹². Content per megawatt rises as data-center density and architectural complexity increase ¹², and larger integrated projects carry higher equipment content per megawatt ¹². Revenue per megawatt—and particularly gross profit per megawatt—may therefore be more informative than total data-center capacity alone ¹⁷. Each AI deployment is becoming more valuable not only because it contains more accelerators, but because each megawatt requires more power, cooling, networking, and systems content.

Other companies reinforce the power-and-cooling theme. Vertiv, Modine, nVent, Schneider Electric, and Flex have been identified as beneficiaries of liquid-cooling demand ⁴. Vertiv, Eaton, nVent, and Modine benefit from power-distribution and liquid-cooling expansion ¹³, while Vertiv, Eaton, Schneider Electric, ABB, and nVent have exposure to increased power and electrical-infrastructure demand ⁴. nVent reported 6% revenue growth ²⁶, and Power & Process revenue also grew 6% ⁴⁷.

Not every thermal-related business benefits equally. BorgWarner’s Turbos & Thermal segment declined 4.3% organically to $1.442 billion, with adjusted operating profit broadly flat at $225 million ³⁴. The relevant distinction is exposure to high-density data-center cooling, not thermal technology in general.

The transition toward higher-voltage architectures is a further long-term catalyst and an execution variable. Vertiv planned to release its 800 VDC portfolio in the second half of 2026 ¹⁷, alongside expected commercial offerings from Schneider Electric, Eaton, and Delta ⁷¹. Broader data-hall-level 800V deployment is targeted for 2028 ¹². Adoption will depend on facility readiness, customer standards, commissioning, and system-level compatibility. Is this truly negligible, or have we missed a coupling? In power systems, the answer is rarely found in the component alone.

From Components to Integrated Systems

Vertiv is moving from relatively discrete equipment orders toward larger integrated projects ¹². Its opportunity now extends from individual equipment to integrated powertrain and thermal-chain architectures, installation, commissioning, controls, software, and lifecycle services ¹². Vertical integration and technology acquisitions support this full-stack strategy ¹², while integrated power and thermal systems can create higher switching costs and deeper customer integration ^12,17.

The installed base also supports recurring service revenue through maintenance, spare parts, monitoring, upgrades, replacement components, and lifecycle services ¹⁷. This resembles the broader shift from selling individual accelerators toward complete AI computing platforms and ecosystems. As systems become more integrated, performance depends on the interaction of GPU, networking, power delivery, thermal management, software, and facility architecture.

Suppliers capable of coordinating these interfaces may capture more value and become harder to displace. Vertiv’s ability to secure integrated power, thermal, and service scope in Taiwan, Delta Electronics’ home market, illustrates how global full-stack vendors can capture the customer interface in strategically important local projects ¹². It also signals increasing competition between global full-stack vendors and local specialists such as Delta ¹².

The full-stack model is not without friction. Larger projects involve more complex revenue recognition and more demanding commissioning coordination ¹². Vertiv must coordinate multiple factories, external suppliers, customer milestones, and commissioning schedules ¹², creating quarter-to-quarter volatility ¹². Second-quarter revenue timing shifts were attributed primarily to multiphase project execution and temporary supply-chain dynamics, with affected revenue expected to be delivered in the second half of 2026 ¹². The relevant execution risks are supply-chain interdependencies, facility readiness, commissioning, and the timing of 800V adoption ¹². For NVIDIA, the parallel is direct: strong end demand does not prevent complex deployments from being delayed by power availability, networking integration, cooling readiness, or construction schedules.

Orders, Backlogs, and Capacity Expansion

Order evidence is unusually strong across several infrastructure-adjacent suppliers. Hammond Power Solutions reportedly saw data-center orders equal to 53% of its preceding-quarter backlog, while backlog increased 74% in a single quarter ⁶². Mayville Engineering booked a record $50 million of data-center awards in the first quarter, achieved 71% organic data-center revenue growth, and expects data-center revenue to exceed 20% of 2026 revenue, with a fourth-quarter exit rate approaching 25% ²⁶.

Viavi’s data-center revenue excluding Spirent more than doubled. Data-center revenue represented approximately 50% of NSE revenue, and the NSE segment grew 69.2% year over year ³⁹. Laboratory and production revenue grew at a double-digit sequential rate in absolute dollars, with production identified as the fastest-growing area ³⁹. Management expected robust data-center growth for several quarters, accelerating co-packaged-optics revenue in December and recognizing most incremental optical-circuit-switching revenue over the following several quarters ³⁹.

Additional semiconductor and equipment signals include Kulicke & Soffa’s approximately 28% sequential-growth guidance to $310 million for the June quarter and a further 5%–10% growth guide for September ⁴⁰. ASE communicated qualitatively strong July net-revenue growth and expected third-quarter ATM revenue growth of 11%–13% sequentially ^7,15. Vishay’s distributor point-of-sale rose 20.5% year over year, with more than half of its sequential revenue increase coming from Asia; industrial revenue grew 30.1%, backlog increased 18%, and average selling prices rose approximately 2% sequentially ³⁵. Its combined telecom, computing, and consumer category increased 28.4% year over year ³⁵. European seasonality affected third-quarter guidance, while distribution represented 58% of revenue, up from 55% in the prior quarter ³⁵.

These data support a positive ecosystem outlook, but current orders must not be confused with sustainable end demand. Vertiv’s approximately $15 billion backlog and strong pipeline provide visibility ⁵, while its growing installed base adds service stability ¹⁷. Yet backlog conversion depends on project timing, customer milestones, supply availability, and commissioning. The claims do not establish how much reported demand is incremental AI demand versus broader cloud, networking, enterprise, or replacement spending. That distinction is especially important for NVIDIA, whose growth expectations are more sensitive to hyperscaler and sovereign-AI investment than those of many component suppliers.

Financial Quality and the Need for Qualification

Vertiv raised full-year adjusted operating-margin guidance by 50 basis points ^5,12, citing operational execution, productivity, and favorable price-cost ¹². Favorable price-cost performance despite tariffs indicates pricing power in mission-critical electrical equipment ¹², although tariffs partially offset margin improvement ¹². Advance payments and milestone collections supported working-capital performance and free-cash-flow conversion ¹², shifting some cash-flow pressure onto customers ¹². That apparent cash-conversion strength should not be treated as entirely equivalent to recurring, earnings-backed cash generation ¹².

Schneider Electric’s first-half cash conversion was 66%, while adjusted net income rose 28.8% organically on a constant-currency basis ⁶⁴. Software & Services represented 18% of second-quarter revenue. Field Services grew 5% organically and represented 10% of revenue, Digital Services grew at a high-single-digit rate, and AVEVA delivered high-single-digit organic growth led principally by subscriptions ⁶⁴. Annualized recurring-revenue growth was supported by upselling existing customers and new-logo wins ⁶⁴. These recurring and service elements are strategically relevant to NVIDIA’s software and platform monetization ambitions, although the claims provide no direct measure of NVIDIA’s software mix.

The broader evidence also demonstrates why headline growth requires examination. Viatris’s Brand segment grew 6% reported and 4% operationally, while Greater China grew 21% reported and 16% operationally ⁴². TEVIMBRA grew 18%, trailing BRUKINSA’s 31% growth and growing more slowly than BeOne overall ³³. Teleflex reported 28.9% GAAP revenue growth, materially supported by Biotronik; its surgical segment grew 9.2% pro forma constant currency, vascular grew 8%, and interventional growth declined to negative 1% ⁴⁶. ICU Medical reported 6% consolidated organic growth, with Infusion Systems accelerating from 6% to 12%, but Vital Care declined 4% organically and 32% on a GAAP basis ⁵⁰.

The same framework applies to AI infrastructure. Corning, Amphenol, Viavi, Flex, Celestica, Vishay, and Vertiv are showing compelling growth, but acquisitions, sequential comparisons, distribution dynamics, project timing, and backlog conversion can inflate or obscure the underlying run rate. For NVIDIA, the stronger confirmation remains sustained end-customer demand for deployed compute—not merely supplier enthusiasm or component orders.

Implications for NVIDIA

The claims collectively identify a reinforcing investment loop. NVIDIA’s accelerated-computing platforms drive demand for higher-density servers and networking. Higher density raises power-delivery and thermal-management requirements. Those requirements increase the content opportunity for Vertiv, Schneider Electric, Eaton, nVent, Modine, Flex, Amphenol, Corning, Qnity, Viavi, and Vishay. Their orders, revenue growth, capacity investments, and product roadmaps then provide external evidence that AI deployment remains an active capital-spending theme.

The central implication is that AI infrastructure is becoming a systems market. Vertiv’s integrated powertrain and thermal-chain architectures ¹², participation from chip-adjacent cooling through facility heat rejection ¹⁷, and Schneider’s combination of products, systems, field services, software, and subscriptions ⁶⁴ illustrate the shift toward coordinated platforms. This favors NVIDIA’s broader strategy of combining GPUs with networking, systems, and software rather than relying on accelerator silicon alone. It also increases the value of ecosystem compatibility: the more tightly power, cooling, networking, and compute are engineered together, the greater the advantage for vendors able to influence the full architecture.

The medium-term demand outlook is favorable but conditional. Vertiv’s backlog, guidance increase, accelerating pipeline, and capacity expansion ^5,12,17 indicate that customer commitments extend beyond a single quarter. Amphenol’s 63% IT datacom growth and mid-teens sequential outlook ¹³, Corning’s 32% optical growth ^8,51, and Viavi’s more-than-doubling data-center revenue excluding Spirent ³⁹ are useful cross-checks because they occupy different positions in the connectivity chain. Schneider’s growth across North America, Asia, China, and international markets ⁶⁴ further reduces the risk that the theme is geographically narrow.

Valuation remains the principal discipline. Vertiv has undergone an extraordinary multiyear re-rating ¹⁷, traded above $270 in the cited period ¹⁷, and generated a 66.80% year-to-date holding return at 30 June 2026 ⁶⁹. Its valuation, alongside Eaton and Quanta, has expanded ⁵, and elevated valuations are identified as a principal concern ⁵. A sharp price move combined with an extended valuation can produce substantial downside even when long-term fundamentals remain intact ¹⁷. Similar multiple-compression risk applies if AI-capex growth disappoints ⁵. This warning is relevant to NVIDIA because its valuation also depends heavily on confidence in the duration and magnitude of AI spending.

Operational risks are transferable as well. A slowdown in hyperscaler data-center construction would eventually affect Vertiv ¹⁷, while weaker technology spending or overstated data-center demand could reduce the opportunity for power producers and infrastructure providers ⁶⁰. Vertiv must maintain technology leadership as liquid-cooling architectures evolve ¹⁷, and competition from Schneider, Eaton, Delta, and other well-resourced infrastructure companies could intensify ^12,17. Scarcity pricing may normalize ¹², tariffs can erode margins ¹², and cash conversion may benefit from customer advances that are not representative of steady-state economics ¹². These are not direct NVIDIA-specific negatives, but they define the transmission channels through which a slower or more disciplined AI build-out could eventually reach GPU demand.

Scope of the Evidence

A substantial portion of the claim set concerns companies outside NVIDIA’s topic. Examples include Viatris ⁴², Ziff Davis ⁵³, ICU Medical ⁵⁰, Vital Farms ⁴⁴, Vontier ⁴³, Vistra ^56,57,60, Siemens Energy ⁶⁷, Aptiv ²¹, Veolia ⁶⁵, Trulieve ⁵⁵, Murphy USA ³⁸, Green Thumb ^29,55, Flex ⁴, Cyient ⁴¹, Curtiss-Wright ⁴⁷, Redwire ^37,54, Teleflex ⁴⁶, and many others. These claims should not be interpreted as direct NVIDIA evidence, although selected industrial, power, semiconductor, and connectivity observations provide useful thematic read-through.

Other peripheral claims concern Fortive, Ventas, and Vertiv ³; FIS ²²; Axon ⁵⁸; Frontdoor ⁴⁸; Chemed and VITAS ¹⁰; Vingroup ⁷⁰; Urban Company ^27,32; VTEX ⁴⁹; Gartner ¹⁹; Hillman ⁵²; Blade ³⁶; Raymond ⁵⁹; Fuel Tech ³⁰; DuPont ²⁰; Ball ²³; Solid Power ²⁸; Enterprise Products Partners ¹⁴; Ingram Micro ⁷²; BorgWarner ³⁴; and Travere ⁴⁵.

The claims concerning V Technology ^1,68, Viavi ³⁹, Amphenol ¹³, Qnity ²⁵, ASE ^7,15, Kulicke & Soffa ⁴⁰, Vishay ³⁵, and Celestica ^9,31 are the most relevant non-Vertiv corroboration for the NVIDIA ecosystem thesis.

Finally, market and positioning observations on Vertiv—its 17.26% single-session decline during a technology-sector selloff ⁶, subsequent approximately 7% gain ^24,66, selection for a conviction board ⁶³, 4.32% portfolio weight ⁶⁹, and prior re-rating ¹⁷—illustrate the volatility and sentiment sensitivity of AI-infrastructure equities. They are useful for topic discovery, but they should not be mistaken for a trading signal for NVIDIA without direct NVDA price, earnings, valuation, or guidance evidence.

Practical Conclusions

• The claims provide strong, recently corroborated evidence of a broad AI data-center build-out spanning power, liquid cooling, optical connectivity, interconnects, testing, and manufacturing. This is a meaningful positive ecosystem read-through for NVIDIA ^5,8,13,51,64.
• Rising rack density is increasing infrastructure content per megawatt and shifting value capture toward integrated power, thermal, networking, controls, software, and lifecycle-service architectures ^12,17.
• Vertiv’s backlog, pipeline, guidance, and capacity investments indicate durable demand, but project timing, commissioning, supply chains, 800V adoption, competition, and normalization of scarcity pricing remain material risks ^5,12.
• The strategic signal for NVIDIA is favorable but indirect. The cluster confirms strong surrounding infrastructure demand, while valuation and hyperscaler-capex sensitivity argue for monitoring actual end-customer deployments rather than extrapolating supplier growth indefinitely ^5,17.