The Biggest Winner in the Liquid Cooling Boom: NVIDIA’s Most Mysterious Unlisted Supplier in Taiwan
Liang-rong Chen
Hello everyone,
Last week’s 2025 COMPUTEX in Taipei reminded me of a popular science book I loved as a student: The Beak of the Finch.
It tells the story of the famous Darwin’s finches on the Galápagos Islands, which inspired On the Origin of Species. In that harsh environment, different species evolved distinct beak shapes—some strong enough to crack seeds, others long and slender for catching insects, and some so sharp they could draw blood.
Then came a once-in-decades rainstorm. The barren islands suddenly turned lush, food was abundant, and for a few generations, the finches’ beaks began to converge again in shape—no longer shaped solely by scarcity.
To me, the bustling booths of Taiwanese companies at Nangang International Exhibition Center felt just like those finches singing after the storm—joyfully thriving in the rare, once-in-20-years AI boom.
NVIDIA’s earnings this morning only reinforced that feeling: its data center revenue surged 73% year-over-year in Q1, sending its stock up as much as 5% in after-hours trading.
This massive wave of demand for AI infrastructure has drawn in companies from all corners: Accton（智邦）, originally a network switch maker; Wiwynn（緯穎）, a server manufacturer; Delta Electronics（台達電）, a power supply specialist; and even Vertiv, a U.S. firm that used to build control systems for nuclear plants.
Under pressure from hungry customers, they’ve all launched data center solutions—some more integrated than others. The boundaries between business domains are beginning to blur—just like the finches in the rainy season.
Some previously obscure niche players, pushed forward by major clients, are now doubling their revenue year after year, quietly becoming the “hidden champions” of the AI era: vital, yet barely visible.
Read more in this issue.
On the penultimate day of COMPUTEX, I attended a small briefing at a company headquarters in Neihu, housed in a building shaped like a giant heatsink.
The company's R&D engineer was presenting results from their “two-phase immersion cooling” technology, which uses a specialized liquid that vaporizes at just over 60°C to absorb chip heat.
This method far outperforms current liquid cooling systems and is expected to be used in NVIDIA’s next-generation servers in the coming years. Jensen Huang（黃仁勳） has already hinted that thousands of GPUs will be packed into a single rack—inevitably generating astonishing amounts of heat.
Even though no photos or recordings were allowed at this confidential event, I was genuinely excited to be there. This was the most mysterious player in NVIDIA’s Taiwan supply chain. And for the first time, during Computex 2025, it had finally opened its doors for a rare behind-the-scenes visit.
The term “hidden champion” may be thrown around too often these days, but in the case of Cooler Master, it couldn’t be more accurate.
It doesn’t get more “hidden” than this. While Taiwan’s leading thermal players AVC（奇鋐） and Auras（雙鴻） have seen their stocks surge, Cooler Master has refused to go public in its 33-year history, and its founder Roger Lin（林仁政） has never once spoken to the media.
Yet it’s a true champion behind the scenes, serving as the lead supplier of liquid cooling systems for NVIDIA’s next-generation GB200 and GB300 servers.
From Hidden to Hero: Cooler Master Steps Into the Light
The biggest highlight of Jensen Huang’s COMPUTEX keynote this time was a one-minute documentary about Taiwan’s supply chain, starting from TSMC’s CoWoS process to Foxconn's cabinet assembly at its US factory, documenting the production process of each important stage of NVIDIA's proprietary “hundred-million-dollar cabinet” GB200 NVL72.
The company showcased in the film as the representative liquid cooling supplier, Cooler Master, is none other than 訊凱.
Jensen Huang’s familiar voice narrated: “Custom liquid-cooled brass modules from Cooler Master, AVC, Auras, and Delta keep the chips at optimal temperature.” The scene then cut to quick connectors and brass heat sinks being processed, scenes filmed at Cooler Master’s Guishan（龜山） factory.
A senior executive at a major cooling company confided that Jensen Huang keeps pushing the technological limits. The “hundred-million-dollar cabinet” packed with 72 GPUs has already overwhelmed Taiwan’s cooling system suppliers, and the next generation will pack over 500 GPUs. “We don’t even know where the technology is heading.”
It’s no wonder Wiwynn Chairwoman Emily Hong (洪麗寗) remarked in her COMPUTEX speech that cooling technology has become the single most critical factor in AI data center design. She shared a line graph indicating that the demand for advanced cooling (referring to liquid cooling) is currently doubling every two years, and after 2028, it will quadruple every two years.
Morgan Stanley estimates that the liquid cooling components for a ‘hundred-million-dollar cabinet’ are worth $88,000, accounting for 2.9% of total BOM cost, making it the highest component cost apart from GPU and CPU modules.
Cooler Master is the biggest winner of this liquid cooling boom.
Based on information from anonymous Cooler Master executives and industry peers, this Taiwan's most talked-about unlisted tech company has several key characteristics:
First, high vertical integration. From quick connectors to massive coolant distribution units (CDUs), and MGX 44RU Manifolds as thick as fire hoses, “we make everything ourselves,” a Cooler Master executive said.
Cooler Master even developed its own AI-powered inspection machines, currently displayed at its headquarters showroom. Equipped with cameras, it autonomously detects surface defects on brass heat spreaders designed for the GB200. This machine will be widely deployed on production lines in the future, gradually replacing the current manual quality control process.
Second, remarkable R&D investment. A Cooler Master executive revealed that the company employs 1,200 R&D personnel, with R&D spending accounting for 5% of revenue, already at the level of Taiwan’s top tech companies.
This is almost on par with cooling leader AVC, which had an R&D revenue ratio of 6% in 2024. According to the company's 2023 sustainability report, it has “over 1,000” R&D engineers. Cooler Master employs approximately 8,000 people—slightly less than AVC’s workforce of 11,000.
Third, rapid growth driven by “the customer.” From the moment I entered Cooler Master's headquarters building, I had a sense of incongruity.
The interior of the exquisitely designed building, with its cool, futuristic style based on black and purple tones, stood in stark contrast to the utilitarian vibe of typical thermal solution providers.
This stems from Cooler Master's unique gaming heritage. The group started with the “Cooler Master” gaming brand (known as 酷冷至尊 in mainland China). It entered the cooling product OEM business in 1997 and began contract manufacturing for Nvidia in 2000.
The company declined to disclose revenue figures. An executive only stated that the company's revenue has grown eightfold over the past decade, and with the explosion of AI, “recent years have seen exponential growth.” Therefore, contract manufacturing now accounts for 80-90% of its business.
“It's all driven by ‘the customer,’” the executive said.
The Dark Horse That Got Jensen Huang Cheering: Meet Vertiv
But Cooler Master wasn’t the only surprise debutant this year. Another lesser-known member of NVIDIA’s “elite guard” made its debut at this year’s COMPUTEX.
That is, American company Vertiv, whose huge booth on the first floor of Nangang International Exhibition Center Hall 1 was quite eye-catching.
The black-themed exhibition area displayed large transformers and water cooling devices similar to those from Delta and AVC, plus a massive data center liquid cooling system configuration model.
On the second day of COMPUTEX, Jensen Huang made a dramatic sweep through the exhibition hall. When he passed by Vertiv, he raised his fist and called out, “Vertiv! Vertiv! Vertiv!” leaving the company’s executives both surprised and thrilled.
Vertiv is the world’s largest provider of comprehensive data center solutions, with Microsoft and AWS as major clients. It doesn't belong to the traditional definition of NVIDIA’s supply chain.
But as Jensen Huang repeatedly emphasized that Nvidia sells not GPUs but AI infrastructure, Vertiv became an indispensable piece of the puzzle.
At GTC 2024, when analysts asked Jensen Huang how to tackle the soaring power and cooling demands of AI GPUs, he replied that he wasn't too worried because of their cooperation with Vertiv. It shot Vertiv, a relatively recent NYSE entrant from 2020,.into the spotlight of the tech industry.
Originally part of Emerson Electric, Vertiv was spun off and sold to a private equity firm. Following its rebranding and a series of strategic acquisitions, the company repositioned itself as a comprehensive data center solutions provider.
Emerson, Vertiv’s former parent company, mainly catered to nuclear plants, thermal power stations, and large-scale power grids. With that kind of heavyweight pedigree, Vertiv’s move into the data center market felt like bringing a bazooka to a knife fight.
But with AI data centers’ power consumption surging, several super-large clusters like Musk’s “Colossus” with hundreds of thousands of GPUs consume as much power as a nuclear reactor. Vertiv’s heavy electrical experience makes it a market favorite. From 2023 to early 2024, the company's stock price soared nearly 10 times, and its current market value is as high as $41.8 billion.
What brought this surging U.S. AI stock into the spotlight at COMPUTEX 2025?
Veteran analyst and CIO of J&J Investment（騰旭）, Jonah Cheng（程正樺）analyzed on his personal Facebook that COMPUTEX has become an AI data center procurement hub, with Microsoft and AWS buyers attending.
As one senior exec put it, “Vertiv showed up in Taiwan to throw down the gauntlet.”
Hung-chieh Hsu, project manager at Vertiv Taiwan, explained Vertiv's business model: “We don’t make servers,” he said. Vertiv handles other infrastructure. As long as customers specify power consumption requirements and water/air cooling combinations, Vertiv will complete everything from design to construction. Customers only need to install the servers.
Technically speaking, Vertiv and Taiwan’s server OEMs should have stayed in separate lanes.
But now, Wiwynn and Gigabyte are starting to help customers build entire data centers, including designing power, cooling, and server room structures, stepping into Vertiv’s home turf.
“They grew from small to big, while we’re going from big to small,” H.J. Hsu explained.
Vertiv Crashes the Party. Who Are Taiwan’s Real Rivals?
In his view, Taiwan’s clearest competitor is Delta Electronics.
Delta already has a “Grid to Chip” complete data center power supply solution and has recently entered the cooling system field. However, its own products are still limited, mainly relying on external suppliers to build complete solutions similar to Vertiv’s.
“We do everything ourselves,” he said, pointing to the thick server racks in Vertiv’s exhibition area, which came from the acquisition of the German giant Knürr AG several years ago.
In fact, not only Delta, Wiwynn, and Gigabyte, but many Taiwanese manufacturers are trying every means to get a share of the booming AI data center market. For instance, network switch manufacturer Accton displayed a data center rack at its booth, specifically targeting enterprise AI solutions, intending to test the waters in the U.S. market.
The aforementioned major company executive admitted frankly that no one has much of a choice anymore. A few years ago, the market was booming across various sectors, such as automotive, smartphones, tablets, and general-purpose servers.
“Everyone had their own niche. But now, all the attention is on AI,” he lamented. “There's no choice. This is the only place with growth.”
As the AI arms race continues to intensify, potential competitors are emerging from unexpected quarters. According to a senior industry executive, the biggest threat to Delta may not be infrastructure-heavyweights like Vertiv after all.
But the real threat is closer than anyone expected. A tectonic shift is brewing inside the AI server itself—in power management, Delta’s home turf—where an unexpected wave of competition is on the verge of erupting.
Two heavyweight semiconductor giants are preparing to enter the fray, and one of them is none other than TSMC.
To find out what’s next, stay tuned for the next issue.
Read more here:
NVIDIA's Jensen Huang Talks Advancements and Partnerships at COMPUTEX Taipei 2025
Nvidia Supplier Wiwynn Deepens AI Push with Malaysian Data Center Deal
GB300 Is Here—And So Are New Taiwanese Supplier
UBS Analysts Spotlight AI as Bright Spot Amid Tariff Uncertainty and Taiwan Market Risks