Meanwhile, the biggest money in the AI boom is increasingly being made by companies most investors never think about.
Because servers need cooling. Power. Thousands of cables connecting them. Without that, even the most powerful chip in the world does nothing.
All investors have been chasing Nvidia stock in recent months, convinced they're holding the AI bull by the horns. Meanwhile, Vertiv (VRT) — the leader in liquid cooling and UPS systems for data centers — delivered investors a return of +270% over the past 12 months, clearly outpacing Nvidia (+95% in the same period).
In the same period, Nvidia grew +95%. Vertiv's revenues grew 30% year-over-year, and its order backlog exceeded $15 billion.
The thing is, this isn't an isolated case. And the answer to the question "who is really making money in the AI boom" turns out to be more complicated than it seems.
Investment history loves to rhyme. In California in 1849, the biggest winners weren't those who found gold — they were the sellers of shovels, jeans, and tents. Levi Strauss never dug; he sold clothes to miners and built an empire.
In the 1990s dot-com bubble, the greatest fortunes were made not by portal owners, but by Cisco — the company supplying network infrastructure: routers and cables. When the bubble burst in 2001, many internet darlings went bankrupt. Cisco survived and kept growing.
The energy demands of large AI models are staggering. Hyperscalers — Google, Microsoft, Amazon, and Meta — will spend hundreds of billions of dollars this year on new data centers. The problem is that chips alone aren't enough.
AI servers generate extreme amounts of heat. Advanced cooling keeps them running at full capacity, stable power allows 24/7 operation, and high-quality cables and connectors ensure lightning-fast communication between systems.
Nvidia sells the gold. Someone has to supply the tools.
Before we get to the shovels — one honest word about the gold. Nvidia in fiscal year 2026: revenue $215.9B (+65%), Data Center $193.7B, gross margin 75%, free cash flow $96.6B annually. Forward P/E ~24×. This matters — we'll come back to it at the end.
Vertiv designs and delivers complete UPS systems, liquid cooling, and thermal management for server rooms. AI servers generate extreme heat — traditional air cooling can't handle power densities of 100–200 kW per rack. Vertiv's liquid cooling is essential for Nvidia chips to run at full capacity. Hyperscalers treat Vertiv as a "must-have." Main competitors: Schneider Electric and Eaton — but neither matches Vertiv's backlog or Nvidia integration.
Bloom manufactures stationary fuel cells that function like a mini power plant installed directly at the data center. Power grids can't keep up with AI demand, with connection queues measured in gigawatts. Bloom provides clean, reliable 24/7 power without grid dependency. It is the sole power supplier for Oracle Project Jupiter (up to 2.8 GW in New Mexico). In Q1 2026, Non-GAAP EPS grew nearly 15× YoY — from $0.03 to $0.44. Competitors in fuel cells (Plug Power, FuelCell Energy) are financially weaker — Bloom holds a unique position in SOFC. Warning: Forward P/E ~140× assumes flawless execution for many years. Highest potential and highest risk of the five.
A company with 160 years of history designing and building large gas boilers, steam turbines, and complete power blocks installed directly at data centers. In March 2026, it signed a $2.4 billion contract with Base Electron for 1.2 gigawatts of capacity for AI Applied Digital campuses — four 300 MW boilers. Warning: A company with a history of losses, concentration on a single contract, and an ongoing class action lawsuit. Speculation, not investment.
nVent supplies transformers, switchgear, enclosures, and internal wiring — everything between the power source and the server. Provides safe and scalable power delivery inside data centers. Organic orders grew 40% in Q1 2026, mainly driven by large AI data center projects from hyperscalers and colocation firms. The only company among the five that regularly pays a quarterly dividend ($0.21/share in 2026). Competitors: Eaton and Hubbell — but nVent has a stronger position in data center enclosures. A rare combination of growth and stable income.
Amphenol produces connectors, fiber optics, copper and optical cables, and interconnects for GPU servers. Leader in high-speed interconnects for GPU clusters — ahead of TE Connectivity and Molex (Koch Industries). The physical data highway between thousands of GPUs in a cluster, without which no AI cluster operates. After acquiring CommScope CCS in January 2026, backlog exceeded $8.4 billion. Note: In Q1 2026, the company set aside $290M in tax reserves related to a tax audit in China — a one-time event, but worth monitoring.
| COMPANY | SEGMENT | REVENUE GROWTH Q1 2026 |
KEY METRIC |
|---|---|---|---|
 NVIDIA (NVDA) |
Chipy AI | +69% | Market Cap > $3T |
 VERTIV (VRT) |
Cooling / Power | +30% | Backlog ~$15B |
 BLOOM ENERGY (BE) |
Onsite power | +130% | Guidance +80% (2026) |
 nVENT ELECTRIC (NVT) |
Electrical Infrastructure | +53% | AI Orders +40% |
 BABCOCK & WILCOX (BW) |
Bloki energetyczne | n/d | Contract $2.4B |
 AMPHENOL (APH) |
Cables & Connectors | +58% | Backlog ~$8.4B |
The larger a company's market cap, the harder it is to grow at the same percentage rate. Smaller, specialized infrastructure companies have a clear leverage advantage in this cycle — and notably, they benefit from the entire class of AI models simultaneously, regardless of which ultimately wins the market race.
Each of these companies has real exposure to the AI boom. These aren't small, undiscovered companies — Amphenol and Vertiv are well-known on Wall Street. The point is that retail investors often overlook them, focusing exclusively on chips. But exposure isn't everything — the price you pay matters too. And that's where it gets interesting.
The article shows who grew. But there's one question that needs to be asked directly: after such gains, are infrastructure companies still attractive? Hard valuation data calls for caution.
| Company | Forward P/E | Trailing P/E | Ocena |
|---|---|---|---|
| 52–54× | ~83× | Very expensive | |
| 140–157× | ekstremalnie | Extremely expensive | |
| 40.36× | 39.37× | Drogo | |
| 30–31× | ~39–42× | Most reasonable | |
| 140.85× | 6.47× | Very risky | |
NvidiaNVDA · Forward P/E · GuruFocus · Yahoo · IR |
24.45× | 40.51× | Cheaper than "shovels" |
One of the more surprising conclusions of this analysis: Nvidia at forward P/E ~24× is today valued cheaper than most of the infrastructure companies that were supposed to "beat" it. Vertiv at 52×, Bloom Energy at 140× — this is no longer a story about shovels, it's a story about how the market priced in the future with a significant premium.
Amphenol stands out as the only company in this group with a relatively rational valuation. The others require continued, highly dynamic revenue growth to justify current prices.
| Company | Adj. Op. Margin | Gross Margin | Main Competitors |
|---|---|---|---|
NvidiaNVDA |
~67% | 75% | AMD, Intel, custom silicon |
AmphenolAPH |
27,3% | 36,7% | TE Connectivity, Molex |
Bloom EnergyBE |
17,3% | 31,5% | Plug Power, FuelCell Energy |
VertivVRT |
20,8% | ~35% | Schneider Electric, Eaton |
nVent ElectricNVT |
~20% | ~40% | Eaton, Hubbell, Vertiv |
Babcock & WilcoxBW |
~3,5% | ~25% | GE Vernova, Siemens Energy |
One important note: forward P/E is not a verdict. With revenue growth of 30–130% annually, high multiples can be justified — provided the growth rate holds. That "if" is the biggest risk.
1. Amphenol (APH) — best risk/reward. Lowest P/E, highest margins in the group, M&A discipline.
2. Vertiv (VRT) — purest AI exposure. Expensive, but backlog >$15B and rising margins justify the premium.
3. nVent Electric (NVT) — solid and dividend-paying. Less exciting, but safest after Amphenol.
4. Bloom Energy (BE) — high potential with successful scaling. Extreme 140× valuation requires flawless execution.
5. Babcock & Wilcox (BW) — speculation, not investment. One contract, lawsuit, history of losses. Only for those tolerating high volatility.
The picks and shovels strategy is not without pitfalls. Data center construction delays, rising energy costs, supply chain issues, or sudden changes in energy policy can quickly shift the trajectory. High valuations of some companies — especially after such dramatic gains — leave little margin of safety.
A separate risk comes from the hyperscalers themselves. Google, Microsoft, Amazon, and Meta are actively developing their own solutions — custom power, advanced cooling, proprietary interconnects. In the long term, they may partially bypass external infrastructure providers. This isn't a tomorrow scenario, but worth keeping in mind.
It's also worth noting that Babcock & Wilcox, despite its spectacular stock gain, is a company with high concentration risk. One large contract with Applied Digital makes up a significant portion of the backlog, and a class action lawsuit surrounds disclosures related to that deal. This is an example of how the biggest gains come with the biggest volatility and execution risk.
Levi Strauss didn't dig for gold — and got rich off those who did. A beautiful story. But there's one detail that gets overlooked: Levi sold pants at a reasonable price. Not at 140 times future earnings.
Picks & shovels remains a valid thesis — but in 2026, the market discovered it long ago and paid a hefty premium. Vertiv 52×, Bloom Energy 140× forward P/E. Meanwhile, Nvidia — the company investors "abandoned" in favor of shovels — trades at ~24×.
The AI boom is real. Infrastructure is essential. But the cheapest company in this set turned out to be the gold, not the shovels.
| Company | 10-K (FY2025) | Earnings Call Q1 2026 |
|---|---|---|
| Vertiv (VRT) | SEC Filings IR | Motley Fool · 22.04.2026 |
| Bloom Energy (BE) | SEC Filings IR | Motley Fool · 28.04.2026 |
| Amphenol (APH) | SEC Filings IR | Motley Fool · 29.04.2026 |
| nVent Electric (NVT) | SEC Filings IR | Motley Fool · 01.05.2026 |
| Babcock & Wilcox (BW) | SEC Filings IR | SEC EDGAR |
| Nvidia (NVDA) | 10-K FY2026 · SEC | investor.nvidia.com |