From the chip to the chiller
Behind every AI computation lies a chain of physical infrastructure, starting with a single watt of electricity. It begins at a processor, the component that attracts the most attention, but to reach that chip, it must pass through cooling systems, power distribution units, transformers, substations and, ultimately, a generating plant. Each layer presents a business opportunity. In 2026, however, the layers furthest from the chip are demonstrating some of the strongest economics.
The reason is scarcity. In parts of the US, new grid connection requests now face waiting times of up to seven years, while data centres are projected to almost double their share of national electricity consumption, from around 6% today to 11% by 2030. When a resource becomes both essential and scarce, pricing power accrues to whoever controls it; that is why the most sophisticated capital has migrated towards power and electrical equipment.
The proof is in the order book
Consider Vertiv, which makes the power and cooling hardware that keeps Graphics Processing Unit (GPU) racks alive. First-quarter revenue rose around 30% to US$2.65 billion, Americas organic sales jumped 44% and operating margin widened 430 basis points to 20.8%. Yet the figure that matters is forward-looking. Its book-to-bill ratio is near 2.9 times: nearly three dollars of new orders for every dollar shipped. Companies do not generate that kind of demand if a bubble is on the verge of bursting. The same trend is evident across the electrical equipment sector: Eaton’s Electrical Americas data centre orders reportedly climbed about 240% y-o-y, while GE Vernova’s electrification segment booked US$2.4 billion of data-centre equipment orders in a single quarter, exceeding its total for 2025.
Micron’s late-June results made the point in the bluntest terms yet. Revenue more than quadrupled year-on-year to a record US$41.5 billion, gross margin hit an extraordinary 84.9%, and data-centre revenue surpassed US$25 billion in the quarter, a US$100-billion annualised run rate. Management guided the current quarter to US$50 billion, far above the Street’s US$43 billion. Yet, the most important disclosure was not a number on the income statement. Micron has signed 16 five-year “Strategic Customer Agreements”. These are take-or-pay contracts that lock in roughly US$100 billion in minimum revenue and US$22 billion in upfront customer cash, covering a fifth of its DRAM and a third of its NAND volume through 2030. That is the difference between a backlog and a bubble: customers are now writing cheques years in advance to guarantee supply.
See also: Local firms powering Singapore’s strong green revenues
At the base of the AI infrastructure stack sits the most defensible asset of all: reliable electricity generation. By 2026, every major hyperscaler had signed at least one nuclear deal, totalling some 9.8 gigawatts (GW) of committed capacity. Nuclear generation cannot be permitted or built overnight, which is why these long-term contracts have become so valuable.
For investors closer to home, the implications are equally relevant. Singapore’s data centre moratorium in 2019, followed by its selective, sustainability-led reallocation of capacity, has made regional data centre capacity increasingly scarce and valuable. This has strengthened the investment case for operators with secured power access and land banks, benefiting digital infrastructure REITs as well as companies such as Keppel and ST Engineering through their infrastructure and grid-related businesses. Micron’s expansion also includes a new HBM packaging facility in Singapore.
The crack in the foundation
All this supplier prosperity ultimately rests on customers continuing to spend beyond their means. Free cash flow across the top five US capex spenders is turning negative for the first time in about 35 years, with capex intensity reaching 34% of revenue, more than double the late-1990s internet peak. Increasingly, the gap is plugged with debt: roughly US$108 billion was raised in 2025, against projections of some US$1.5 trillion in total.
See also: ‘Generational’ investment opportunity in green economy: LSEG
Two risks follow. The first, and arguably the most significant for investors, is overbuilding. If AI revenue disappoints, a debt-funded buildout reverses fast and the fragile neocloud layer, sub-investment-grade tenants renting GPUs, snaps first. The second risk is increasing competitive pressure, even among today’s apparent winners. Reports that Nvidia tapped Samsung and SK Hynix for slices of next-generation HBM4 are a reminder that today’s pricing power is not permanent, even in a near-duopoly. There is also a broader consequence that many investment narratives overlook: memory prices have more than doubled over the past year, creating inflationary pressure across products ranging from smartphones to automobiles.
Implications for Investors
The temptation is to dismiss the entire AI ecosystem as a bubble and walk away. The spending is real, contractually committed, and already under construction. This is not the speculative excess of 1999 and Micron’s long-term, non-cancellable customer agreements provide compelling evidence of that distinction.
At the same time, valuations leave little room for disappointment. Vertiv trades at around 53 times forward earnings, while Micron’s market capitalisation has surpassed US$1 trillion following a rally of nearly 700%. Those are demanding valuations for any cyclical business, regardless of how compelling the long-term structural story may be.
A disciplined investment approach, therefore, remains essential. Companies exposed to electricity generation, grid infrastructure, and power equipment may warrant accumulation during periods of market weakness, where contracted cash flows and strong competitive moats support long-term value. Meanwhile, higher-growth equipment manufacturers and memory leaders may be better accumulated during market pullbacks rather than chased at record highs.
The AI revolution may ultimately be measured in intelligence, but its costs are denominated in electricity. Today, the smartest investors are reading the meter.
Wong Ming Wei is a dealing manager at PhillipCapital
