- AI infrastructure is in early innings; liquid cooling is growing at an annual rate of about 45%, with data centers growing around 1% CAGR.
- Liquid cooling adoption is just 5% of data centers today, projected to reach 23% by 2028.
- Envent Electric (NVT): ~40% market share in liquid cooling; contributes 23% of revenue; forward P/E ~31; YTD up about 40%; shares around $94.
- Vertiv Holdings (VRT): YTD up roughly 19% as hyperscalers reaffirm “all systems go” capex for data centers.
- Broadcom: up about 51% YTD and 40% in the last three months; the highest analyst target cited at $420.
- Arista Networks: up about 27% YTD and 57% in three months; forward P/E ~64x; consensus target $157.88 implies ~11% upside.
The AI Infrastructure Supercycle: Why This Buildout Is Different
The AI buildout is not a software upgrade—it’s an industrial-scale infrastructure project. Think highways, railroads, and telephone lines, then layer on the internet buildout of the 1990s. The new wave of AI requires power-dense facilities, purpose-built cooling, high-speed networking, and specialized silicon. Crucially, much of it must be built from scratch. That’s why the theme is measured in years and, potentially, decades. Projects can take two to three years just to get going, and the commitments from hyperscalers are already in place. This is a long-cycle story with multiple on-ramps for investors.
A core takeaway: investors are not just paying for today’s earnings; they are paying for tomorrow’s capacity. The market validates that view, with several leaders trading at premiums that reflect future demand.
Liquid Cooling: Small Base, Big Ramp
Cooling is emerging as one of the highest-growth niches in AI infrastructure. Two facts frame the opportunity. First, the liquid cooling market is expanding at about a 45% annual clip. Second, only about 5% of data centers use liquid cooling today, but analysts project that figure to rise to 23% by 2028. When compute density rises and uptime is non-negotiable, efficient thermal management becomes indispensable.
Envent Electric (NVT) is a prime beneficiary. It holds roughly 40% share in liquid cooling and derives about 23% of its revenue from the category. The company collaborates with major hyperscalers and is building solutions for next-gen platforms—positioning it as a critical partner at the front end of the spend. Yes, the stock screens as premium (forward P/E ~31) after a roughly 40% YTD climb, around $94. But if adoption curves play out, today’s price could look modest in hindsight.
Vertiv Holdings (VRT) is the better-known cooling peer. Shares are up about 19% YTD, with a strong last three months as hyperscalers like Microsoft and Meta reiterated “all systems go” capital expenditures in the latest earnings cycle. The takeaway: fears of capex pullbacks have receded, and suppliers tied to first-wave buildouts remain well bid.
Silicon at the Core: Nvidia and Broadcom
No AI stack conversation skips Nvidia. The company remains the category leader with what was framed as roughly 80% market share, continuously refreshing its platform and locking in demand from hyperscalers that want the latest chips. While no single company can sustain hypergrowth forever, the cadence of innovation suggests Nvidia will remain central to AI infrastructure for years, not quarters. Exposure can be direct or via broad tech ETFs; either way, many investors will want it in the mix.
Broadcom plays a pivotal, adjacent role. Its application-specific integrated circuits (ASICs), high-bandwidth memory controllers, and networking chips give hyperscalers alternate paths to performance while reducing dependency on a single vendor. Partnerships span Alphabet, Meta, and Amazon. The stock is up about 51% this year and roughly 40% over the last three months, hovering near its consensus target and 52-week high. Analysts have been lifting targets since early September, with the highest cited at $420. If the market gets a year-end tech rally, potentially aided by lower interest rates, Broadcom stands to benefit.
Racks, Servers, and the Network Fabric: Super Micro and Arista
If Nvidia and Broadcom are the heart and circulatory system, then Super Micro Computer (SMCI) is the skeletal frame and muscle that makes everything work. The company designs customizable servers and racks that integrate GPUs and advanced chips, making it a go-to for hyperscaler deployments. A financial reporting issue earlier this year dented sentiment, but the structural need for tailor-made systems ties Super Micro to the data center construction cycle. The stock is up about 48% in 2025, roughly 4% in the last three months, and has met resistance around $60 twice; an analyst from NEM sees potential to revisit that level, with some targets higher.
Arista Networks is the nervous system—wiring AI clusters together with high-performance networking. It’s the dominant name with partnerships at Microsoft and Meta and, critically, the kind of “locked-in” demand that comes with hyperscaler commitments. Shares are up about 27% YTD and more than 57% over the last three months. Valuation is rich (forward P/E ~64x), but the consensus target near $157.88 still implies roughly 11% upside, with some houses, like Goldman Sachs, higher. Investors wary of chasing a sharp move might scale in over time; the multi-year runway argues against over-timing.
What If You Don’t Want to Pick Winners? The ETF Route
For investors whose heads spin at the number of moving pieces, the Pacer Data and Digital Revolution ETF (TRFK) offers a diversified on-ramp. It’s up about 29% for the year and 18% in the last three months, holds names like AVgo, Nvidia, and Arista, and smooths single-stock volatility. It even pays a tiny dividend. In an ecosystem where a handful of suppliers can swing on headlines, basket exposure may be the right first step.
Conclusion: Implications and Near-Term Catalysts
The AI infrastructure cycle is a build-from-scratch story that likely extends through this decade and possibly into the late 2030s or 2040. Power availability, cooling, server design, networking, and specialized silicon are all mission-critical, and hyperscaler commitments today look “rock solid.” This should support multi-year backlogs across providers tied directly to data center construction.
Investment implications:
- Favor companies with entrenched hyperscaler relationships; those commitments underpin revenue visibility.
- Be willing to pay up selectively for category leaders; in early innings, premiums reflect future cash flows.
- Consider scaling into elevated names to manage entry risk; momentum and valuation can coexist during capacity buildouts.
- Use ETFs like TRFK for diversified exposure if single-stock volatility is outside your risk budget.
Policy implications:
- The discussion highlighted the scale of this buildout relative to historic infrastructure cycles. While specific policies were not disclosed, the theme underscores sustained capital formation needs in power and cooling to support 24/7 operations.
Near-term catalysts to watch:
- Hyperscaler earnings updates confirming capex trajectories (“all systems go” was the last signal).
- Adoption milestones in liquid cooling as penetration moves off the current ~5% base.
- Analyst target revisions and potential year-end tech rally, aided by lower interest rates.
- Project timeline disclosures (many facilities take 2–3 years to ramp), which can firm multi-year revenue outlooks.
The throughline is simple: the AI era demands physical infrastructure. From
chips to chillers, the winners will be those that deliver reliable, power-efficient capacity at scale—and can execute across multi-year procurement cycles without missing milestones. For investors, the strategy is straightforward: stay anchored to the buildout, prioritize mission-critical suppliers with hyperscaler validation, and let time in the trend compound the returns more than timing the trend.
In other words, the AI revolution is not just a model race—it’s an engineering race. Those who own the bottlenecks of power, cooling, interconnect, and silicon will likely own outsized share of the economics. Position accordingly, stay patient, and watch the capex signals.
Date: September 21, 2025
This article is for information only and is not investment advice.