1. Asia Vital Components raises its liquid cooling outlook
Asia Vital Components (AVC), the Taiwanese thermal management specialist, told investors on November 10, 2026 that it has begun shipping double-sided cold plates for AI servers and expects every GPU- and ASIC-based AI server to use liquid cooling by the end of 2027. The forecast accompanied record third-quarter revenue of NT$59.09 billion, up 51.8% from a year earlier.
The outlook places AVC firmly on one side of a debate that has run through the data center industry for several years: how quickly operators will abandon air cooling for the processors that train and run artificial intelligence models. AVC's answer is that the transition will be complete for the most demanding hardware within roughly two years.
It is worth keeping in mind who is making the call. AVC sells liquid cooling hardware, and a rapid, universal shift to that technology would directly benefit its business. The company's latest financial results, however, show that demand for its products is already growing quickly, which gives the forecast more weight than a bare projection would carry.
2. What does Asia Vital Components do?
Asia Vital Components designs and manufactures thermal management products, the hardware that pulls heat away from electronic components so they can keep running at full speed without damage. Its products range from fans and heat sinks to the liquid cooling modules now in demand for AI servers, along with server chassis.
The company is part of Taiwan's dense electronics supply chain, which builds much of the world's server hardware. In the AI boom, thermal suppliers have moved from a supporting role to a central one, because the latest accelerator chips generate far more heat than earlier generations of server processors, and that heat has to be removed reliably for the systems to function.
AVC works with cloud service providers on both graphics processing units (GPUs) and the custom chips those providers design for themselves. It is also among the preferred cold plate suppliers for Nvidia's upcoming Vera Rubin platform, the successor to the company's current generation of AI systems.
3. How a cold plate works
A cold plate is a block of metal mounted directly on top of a hot chip. Coolant flows through narrow channels machined inside the block, absorbs the heat coming off the processor, and carries it away to be released elsewhere in the cooling loop. The arrangement works much like a car radiator, except that it sits right on the component that needs cooling.
Liquid is far better at moving heat than air. A stream of coolant can absorb and transport much more thermal energy than the same volume of air pushed by fans, which is why liquid systems become necessary as chip power climbs. Air cooling also requires large heat sinks and powerful fans, which take up space and consume electricity of their own.
For data center operators, the trade-off is complexity. Liquid cooling requires pumps, piping, coolant distribution units and careful maintenance to prevent leaks. Operators have historically accepted that complexity only when air could no longer keep up. AVC's forecast amounts to a claim that, for AI servers, that point has effectively been reached.
The economics of AI computing reinforce that conclusion. Accelerator chips are expensive, and they deliver their full value only when they can run at sustained high performance. A chip that overheats has to slow itself down to stay within safe limits, wasting part of the investment. Cooling that keeps processors at stable temperatures therefore protects the return on some of the most costly hardware in any data center.
4. Why cold plates are getting a second side
Traditional cold plates cool a chip from one direction, sitting on top of the processor package. AVC says customers are increasingly choosing double-sided designs, in which a single plate assembly has to draw heat away from two directions at once.
The shift reflects changes in how AI server boards are built. As chips draw more power, the components that deliver that power, and the heat they produce, become harder to manage with a cooling surface on only one side. A double-sided plate addresses heat from both faces of the assembly, which makes the engineering more demanding but allows denser, more powerful systems.
Designing such a plate is not a simple matter of adding a second surface. Coolant has to be routed evenly so that neither side runs hotter than the other, and the plate must fit within tight mechanical tolerances on a crowded server board. Suppliers able to manufacture these parts reliably at volume are in a strong position as server designs evolve.
AVC is also developing cold plates for the next generation of chips, which are expected to carry higher thermal design power (TDP) ratings. TDP is roughly the amount of heat a chip is expected to produce under heavy, sustained workloads, and it is the figure cooling engineers design around. Each step up in TDP makes the case for liquid cooling stronger.
5. Record third-quarter results
The financial results AVC reported alongside its forecast suggest the liquid cooling shift is already showing up in its numbers. Third-quarter revenue reached a record NT$59.09 billion, an increase of 51.8% compared with the same period in 2025.
Profitability improved even faster than sales. Gross margin rose to 33.86%, a gain of 7.74 percentage points from a year earlier. AVC attributed the improvement to a growing share of higher-value liquid cooling products and a more favorable mix of products overall.
The margin trend is significant because it indicates that AVC is not simply selling more units; it is selling more expensive, more complex products that earn a larger profit per sale. Liquid cooling modules carry more engineering content than fans and heat sinks, and the company's results show customers paying for that content.
Momentum carried into the fourth quarter. AVC reported record revenue for October 2026 of NT$22.1 billion, extending its run of monthly highs.
Monthly revenue figures are closely watched in Taiwan's technology sector because listed companies there publish them regularly, giving investors an early read on demand well before quarterly results arrive. A record October suggests that the orders behind AVC's third-quarter performance did not fade as the year moved toward its final months.
6. Capacity expansion focused on Vietnam
To meet the demand it expects, AVC plans to increase its liquid cooling production capacity by more than 50% in 2027. Most of the additional output will come from its facilities in Vietnam.
Vietnam has become an increasingly important manufacturing base for Taiwanese electronics suppliers seeking to diversify production and serve customers that want supply chains spread across more than one country. For AVC, the expansion there is the practical foundation of its forecast: a projection that all AI servers will be liquid-cooled matters only if suppliers can build enough cold plates to make it happen.
Expanding production of precision cooling parts also takes time. New lines need equipment, trained staff and qualification by customers, who typically test components extensively before approving them for use in their servers. Announcing the 2027 expansion now gives AVC lead time to bring that capacity online as demand develops.
A capacity increase of that size is also a signal of confidence in customer orders. Thermal suppliers typically expand in step with commitments from server makers and cloud providers, since building new production lines ahead of demand carries the risk of idle factories if orders slow.
7. Custom ASICs set to overtake GPUs
One of the more notable points in AVC's outlook concerns the type of chip its products will cool. The company expects revenue tied to application-specific integrated circuits (ASICs) to exceed revenue tied to GPUs in its server business in 2027.
ASICs are chips designed to perform one specific job, rather than the broad range of tasks a general-purpose processor or GPU can handle. Large cloud providers have invested heavily in designing their own ASICs to run AI workloads, aiming to improve efficiency and reduce their reliance on off-the-shelf accelerators.
For a thermal supplier, the shift matters because each custom chip program brings its own cooling requirements. AVC's expectation that ASIC revenue will overtake GPU revenue suggests that cloud providers' in-house chips are moving into large-scale deployment, and that those chips, like the GPUs before them, are generating enough heat to require liquid cooling.
The development does not mean demand from GPU platforms is fading. AVC continues to supply cold plates for Nvidia systems and is preparing for the Vera Rubin generation. Rather, it reflects a broadening of the market, with custom silicon growing alongside, and in AVC's case faster than, the GPU segment.
8. What AVC's forecast means for data centers
If AVC's projection holds, data centers built to host AI hardware will need to be designed around liquid cooling from the start. That affects everything from floor layouts and plumbing to power budgets, since cooling is one of the largest sources of energy use in a data center after the computing equipment itself.
Liquid cooling can improve efficiency by removing heat more directly, reducing the need for large volumes of chilled air. As AI workloads grow, that efficiency becomes increasingly important for operators managing electricity costs and the strain on local power grids.
The next markers for AVC will be its continued monthly revenue reports, progress on the Vietnam capacity build-out during 2027, and the ramp of next-generation platforms such as Nvidia's Vera Rubin. Together, those will show whether the company's prediction of a fully liquid-cooled AI server fleet by the end of 2027 proves accurate.