Not just Nvidia: These power and cooling firms are riding the trillion-dollar data centre boom

HONG KONG: While Nvidia is synonymous with the AI boom, a lesser-known group of power and cooling equipment suppliers is cashing in on a global data centre construction spree as developers race to avoid infrastructure bottlenecks.

Energy-hungry data centres have triggered a surge in demand for ​equipment ranging from transformers to advanced cooling systems, creating winners across Asia's supply chain, though earlier stock-price gains have moderated.

McKinsey forecasts nearly $7 trillion in data-centre ​investment globally by 2030; Nvidia last week said it expects AI spending to remain robust for years.


But building data centres fast enough to meet demand is becoming more difficult. Hyperscalers often want facilities delivered within six months, but grid connection delays can stretch as long as 24 months in some emerging markets and more than eight years in major developed markets, according to consultancy Pivotale AI.

Also read | For a century, journalists have had voluntary ethics code. Editors are giving it AI-era reboot

"Outside the industry circle, people are talking about (graphics processing units), but within the circle, people most ‌certainly question you about the ⁠lead time ⁠for generators and transformers," said Wing Kin Cheung, the CEO of digital infrastructure service provider BodaData.

Transformers convert high-voltage electricity from grids into levels suitable for servers, cooling systems and power distribution units.

AI SCRUTINY DEEPENS

Leading transformer suppliers including South Korea's HD Hyundai Electric and China's Hainan ​Jinpan Smart Technology reported surging demand in the first half of 2026 tied to AI infrastructure projects, particularly in North America.

HD Hyundai Electric recently said demand in Europe was rising as U.S. hyperscalers expanded investments in markets ​like Finland, Germany and Britain, while Middle Eastern demand remained strong.

Its order backlog rose 23% to $8.5 billion at the end of June from six months earlier, and it said it expects data centre demand to remain robust.

"We currently have an order backlog covering more than three years, with a substantial portion of production capacity for major power equipment secured for the coming three years," the company told Reuters.

Also read | Fear the reckless humans, not the rogue AI

"We are also ​in order discussions with some key customers for volumes scheduled for delivery as far out as 2030."

For Jinpan, new data-centre orders in the ⁠first half ‌more than quadrupled from a year earlier, while its related backlog nearly tripled.

As AI chips consume more electricity, equipment makers are also betting on technologies aimed at improving ​efficiency and reducing environmental impacts, amid ​growing public scrutiny over data centres' consumption of water and electricity.

Bank of America estimates power consumption per AI rack could climb to more than 1.5 megawatts ⁠by the end of 2030, nearly 100 times that of a conventional rack, citing Nvidia's roadmap.

One technology attracting greater ​attention is the solid-state transformer (SST), a device that replaces bulky magnetic coils and copper windings with semiconductors to transform and route electricity.

UBS estimates ​SSTs will increase power efficiency by around 4% and reduce costs. While commercial adoption remains in its early stages, the bank expects their penetration to climb to 40% in 2030 and forecasts that Chinese companies will gain share thanks to technological expertise and cost advantages.

HD Hyundai Electric and Jinpan said they are deepening SST development, while Taiwan's Delta Electronics, a major supplier of power infrastructure, said a small data centre is using its SSTs.

Delta told Reuters that demand for AI power, cooling and data centre infrastructure solutions remains a growth engine and it is expanding its production footprint across Thailand, the U.S. and China to meet increased demand.

COOLING RACE

Cooling systems are emerging as another growth area as operators struggle to manage the heat generated by powerful AI chips.

"Power and cooling basically go hand in hand; so basically the more power you use, ‌the more cooling you need to use because you generate heat," said Matty Zhao, Bank of America's Asia-Pacific head of research for basic materials, oil and gas.

The bank forecasts liquid cooling will account for 70% of new AI data-centre installations versus air cooling by 2030, up from about 30% today. Liquid cooling can reduce energy consumption ​by over 27%, McKinsey says.

Developers ​are also exploring unconventional approaches, including floating facilities, underwater data ⁠centres and servers in caves or tunnels.

That is creating opportunities for a broader range of suppliers.

"With the expansion of data center self-generation and the floating data center market, opportunities are also opening up to enter new markets for marine medium-speed engines," HD Hyundai Electric said.

Strong demand for thermal-management products is lifting Delta and local peers Asia Vital Components and Auras Technology, as well as China's Shenzhen Envicool ​Technology. All are suppliers in Nvidia's ecosystem.

SUPPLY CHAIN CONSTRAINTS

Despite surging orders, the stock-price gains of suppliers have moderated as investors question elevated valuations amid intensifying competition.

Delta's shares are up more than 90% this year, while HD Hyundai Electric has stayed largely flat, cooling from gains of more than 100% last year. China's Jinpan and Envicool have fallen nearly 30% and 20%, respectively, after surging 118% and 244% in 2025.

"Even if revenue increases, I think gross margin will probably remain at roughly this level," Delta Chairman Ping Cheng said in July.

"There are many variables in the market, including new product platforms, deployment delays and component shortages. These issues may become somewhat more serious in the second half of this year."

Bank of America's Zhao said investors should be aware of potential risks.

"Not everyone can win," she said. "You have to be cherry-picked for the leaders who actually get the customers."