Auras Technology: A Thermal Vendor Climbing the AI Liquid-Cooling Stack

Where does that leave the stock today? Fundamentally, Auras sits in the sweet spot of the AI-infrastructure stack where a rising rack budget forces more cooling content and favors suppliers that can deliver an integrated kit rather than a single part. Competitively, it looks stronger than a plain module vendor but weaker than a fully diversified infrastructure company like Delta or Vertiv. In capital-market terms, the market is pricing a lot of the near-term upside already. That does not make the business bad. It means the investment debate has shifted from “is the ramp real?” to “how much of the next three years is already in the price?”

Company vertical history and financial review

Auras was founded in 1998 and, according to its own sustainability reporting, started as a notebook-radiator designer, manufacturer, and seller. That origin matters because the company came out of the Taiwanese electronics ecosystem at a time when thermal constraints were moving from desktop boxes into thinner, hotter portable devices. The first problem was simple but valuable: make compact heat removal work at scale inside contract-manufactured electronics. The early business model was therefore not software, not branded systems, and not end-market ownership. It was design, process control, and manufacturing discipline inside somebody else’s hardware cycle. Public company material does not give a full pre-founder career narrative for Chairman Lin Yu-shen, so it is safer to say the disclosed evidence points to an operator-led manufacturing company rather than a venture-backed science project.

The listing path is clearer than some secondary databases make it look. TPEx and Goodinfo show Auras was founded on 1998-08-24, first traded on the emerging market on 2004-05-17, and then formally listed on the TPEx board on 2005-05-13. That resolves the common date mismatch in market databases. The capital-markets story at listing was a thermal-module maker serving growing personal-computing hardware volumes, not an AI-infrastructure supplier. The company’s own history also shows a familiar Taiwan pattern: R&D and sales anchored in Taiwan, with manufacturing pushed into China as scale rose. Auras says it set up a Kunshan factory in 2003 and, over the following decade, expanded production lines in Guangzhou, Chongqing, and Hefei. That was a classic cost-and-scale decision for the era.

The first stage of Auras’s history runs from founding through the mid-2000s. The growth driver was customer demand for notebook thermal modules and adjacent PC cooling parts. The constraint was credibility: thermal suppliers can design clever prototypes, but OEM and ODM customers want yield, reliability, and fast customization. The choice to build a China factory early, while keeping Taiwan as R&D and sales, was the company’s answer to that constraint. The lasting impact of this stage is not the product mix itself. It is the operating habit of building around customer designs and manufacturing them at repeatable quality. That habit still shows up in liquid cooling, where much of the value lies in custom geometry, validation, and integration inside a larger system.

The second stage runs roughly from the mid-2000s through the mid-2010s. Here Auras broadened from notebooks into servers, graphics cards, motherboards, gaming systems, smartphones, and other information-electronics products. The sustainability report says heat pipes and vapor chambers were increasingly self-made to meet customers’ customized cooling-module needs. That sounds incremental, but it changed the economics. The company was no longer just assembling bought-in parts. It was pulling more of the thermal stack in-house. That improved design control and likely helped margins over time, though the public filings do not provide a neat bridge for each product family. In industry terms, this was Auras learning to survive across several consumer and enterprise hardware cycles rather than depending on a single device category.

The third stage was the long pre-monetization investment in liquid cooling. Auras’s English sustainability report says the company had advanced liquid cooling cold-plate technology alongside air cooling for many years, and specifically says IBM-related liquid-cooling technology transfer happened about a decade earlier. The report also says Auras offered rear-door heat exchangers as a way for customers to add water-cooling to cabinets without ripping up the entire data-center layout. That is the important turn in the story. The company did not wake up in 2024 and decide to “do AI.” It spent years learning cooling topologies that only became commercially urgent once rack densities jumped and PUE pressure tightened. In hindsight, this was the most underrated node in Auras’s history because the market had little reason to pay for it until AI made it mandatory.

The fourth stage began around the pandemic period and accelerated sharply after 2023. Auras says it built a Thailand factory in 2020 and started mass production there in the first half of 2021. In the 2025 annual report, Chairman Lin explicitly links the production-base strategy to U.S.-China trade frictions and customer supply-chain management needs. Thailand was therefore a resiliency move and a customer-service move as much as a cost move. In a world where hyperscalers, server OEMs, and platform owners want geographic redundancy, that matters. The May 2026 board-approved land-and-building purchase in Samut Prakan shows the shift is still underway. The long-term effect is higher capacity and lower geopolitical concentration. The short-term effect is more capital tied up in the ramp.

The latest stage is the current AI-driven rerating. The 2025 annual report is unusually blunt about what the company believes it has become. It says Auras has moved from an air-cooling module manufacturer into a liquid-cooling rack-system supplier, and it lists water blocks, cooling distribution devices, manifolds, pumps, and quick connectors as product categories designed to accelerate customer adoption of Auras’s liquid-cooling technology. The same report describes quick connectors as a strategic material because they improve reliability and supply-chain flexibility while also giving room to optimize cost structure. That language is revealing. Commodity component businesses do not usually talk like that. System businesses do.

Recent management and governance changes fit this transition story. The chair historically also served as general manager, but on 2024-08-09 the board appointed Chih-Wei Chen as new president and general manager, with the change framed as a corporate-governance step. Lin remained chairman and the largest disclosed individual holder at about 14.35% as of the 2025 annual report cut-off. That combination is usually a plus for a Taiwanese industrial company in transition: founder-level ownership remains meaningful, but day-to-day management is handed to an executive team as the business gets more complex. The board structure includes independent directors, and the annual report does not flag a recent auditor change or major accounting dispute.

The financial vertical is where the narrative becomes concrete. Goodinfo’s consolidated operating series shows revenue at about TWD 14.26 billion in 2021, TWD 13.86 billion in 2022, and TWD 15.78 billion in 2023. The 2025 annual report and Yahoo revenue series together make the rest of the sequence clearer: 2024 revenue was TWD 15.78 billion and 2025 revenue rose to TWD 23.28 billion, up 47.6%. Profit moved faster than revenue in the latest phase: net income was about TWD 1.89 billion in 2024, TWD 2.57 billion in 2025, and TWD 1.16 billion in 2026 Q1 alone. Gross margin improved from 27.4% in 2025 to 29.7% in 2026 Q1, and operating margin reached about 17.8% in that quarter. This is not what a pure scale business looks like. It is what a favorable mix shift looks like.

The numbers also show why not to call Auras a finished compounder yet. Cash conversion has turned messy during the buildout. Goodinfo’s consolidated cash-flow history shows operating cash flow of TWD 1.44 billion in 2021, TWD 1.24 billion in 2022, TWD 1.97 billion in 2023, TWD 1.64 billion in 2024, and negative TWD 571 million in 2025. Summed over 2021–2025, operating cash flow covered only about 71% of net income. Excluding the 2025 working-capital and ramp distortion, 2021–2024 cash conversion was actually stronger than earnings. That split matters. The long-run business has not shown chronic low-quality earnings. It has shown a recent year where growth inflation in receivables, inventory, and capex temporarily broke the conversion pattern. For a growth investor, that is tolerable. For a valuation investor, it is the main reason not to wave away the current multiple.

The balance sheet remains usable rather than stretched. Cnyes shows 2026 Q1 cash and equivalents of about TWD 5.42 billion, receivables of TWD 10.28 billion, and inventory of TWD 7.54 billion. That is a lot of working capital, but it is what you would expect when sales almost doubled year on year, export sales represented 75% of 2024 revenue, and the company was expanding in Thailand while serving AI-server ramps. The bigger issue is not solvency. It is whether working capital normalizes as the liquid-cooling business matures, or whether Auras remains trapped in a build-and-chase model that keeps reported earnings ahead of free cash flow.

The price history tracks those stages closely. The stock was a modestly valued industrial hardware name for years. Goodinfo’s long-run series shows year-end closes of TWD 214 in 2020, TWD 202 in 2021, TWD 148 in 2022, then TWD 352.5 in 2023, TWD 672 in 2024, and TWD 1,010 in 2025. The 2022 drawdown came as end-market conditions softened and the pre-AI narrative had not yet been monetized. The 2023–2025 rerating combined multiple expansion with rising EPS, which is usually the most powerful kind of up-leg and the least forgiving if expectations slip later. In earlier years, the market valued Auras more like a cyclical component name. Today it values it like a scarce AI-infrastructure enabler. The center of gravity has changed because the business mix has changed, not just because liquidity chased a theme.

Business model, industry, competition, and current fundamentals

Auras’s official segment disclosure is still frustratingly simple: in the 2025 annual report, 100% of revenue is booked under thermal modules. That means investors do not get an audited split among legacy air cooling, vapor chambers, cold plates, manifolds, quick disconnects, CDUs, or service tools. The absence of segmentation is one of the report’s real blind spots. Still, management commentary fills in enough to understand the business machine. The growth center is now servers, and within servers, liquid cooling. BigGo’s summary of the 2026 Q1 call said server revenue reached 76% of sales and liquid cooling passed 55% of sales. Public broker-sourced reports in March and June 2026 said the ramp was being driven by GB300 water blocks, manifolds, CDU, and ASIC-server projects, with liquid-cooling share potentially reaching 60%–70% by year-end. Those are not audited lines, but they are directionally consistent with both the annual report and the monthly revenue pace.

The real profit source appears to be the move into higher-value liquid components, not the residual air-cooling base. Management has explicitly said product-mix optimization should support gross margin at or above 30%, and public market coverage has said liquid-cooling gross margin runs above the corporate average. This fits the business logic. Air cooling in notebooks and commodity PC hardware is a scale-and-yield business with modest differentiation. Liquid cooling adds design complexity, validation burden, leak-risk management, fluid interfaces, firmware in the case of monitored CDU systems, and much higher customer pain if something fails in the field. Those attributes usually support better pricing than a plain heat sink.

The cost structure also fits a business with real operating leverage but non-trivial reinvestment needs. Variable costs remain high because Auras is still a manufacturing company. Copper, machining, brazing, tubing, pumps, connectors, and module assembly all scale with volume. But the recent margin expansion shows that once R&D, validation, and customer qualification are done, additional volume can carry higher incremental margin. The same Q1 2026 management summary that cited a 29.7% gross margin also described operating expenses falling below 12% of sales, which is exactly what you would expect when revenue outruns the fixed-cost base. The hard part in a downturn would be cutting factory and engineering expense quickly without impairing qualification schedules for next-generation platforms.

The moat is real, but narrower than the stock’s most enthusiastic holders sometimes imply. The first real moat is customer co-development around difficult thermals. Auras’s disclosures stress customization across cold plates, manifolds, pumps, quick connectors, and monitored rack systems. Thermal design is full of geometry, flow, pressure-drop, leakage, serviceability, and materials trade-offs; once a supplier is qualified inside a reference design, replacement is possible but not frictionless. The second moat is partial vertical integration. The company spent years bringing heat pipes and vapor chambers in-house, and in liquid cooling it presents a one-stop menu from water block to CDU. The third moat is manufacturing footprint and the willingness to spend ahead of demand, especially in Thailand. Those are real moats because they still matter when demand is strong and when customers want multi-country supply. What Auras does not have is a monopoly moat. It operates in a field where other capable suppliers can win programs if they offer better cost, reliability, or integration.

Management looks credible on execution, with two caveats. The positive side is obvious: public filings show meaningful founder ownership, years of pre-AI investment in liquid cooling, and a governance step in 2024 that separated the chairman and general-manager roles. The annual report also shows no notable recent board dissent on key matters and no recent auditor churn. The caveat is that management’s long-term commentary is predictably optimistic. Lin publicly said the AI opportunity looked good through 2028, and management lifted the 2026 growth target from around 50% to around 70%. That may prove right. It also means the next few quarters matter a great deal, because the stock’s valuation now leans on management being roughly correct.

The industry backdrop is powerful. Nvidia and its ecosystem have made liquid cooling a physical requirement for the hottest rack-scale AI systems. Nvidia’s official GB200 NVL72 and GB300 NVL72 pages both describe fully liquid-cooled rack-scale platforms. SemiAnalysis framed GB200 NVL72 as roughly a 120 kW rack, far above a general-purpose CPU rack and well beyond the normal range for older air-cooled AI installations. Nvidia’s own sustainability blog described GB200 NVL72 as balancing unprecedented compute density with energy and water efficiency, while Uptime Institute has written that AI is embracing liquid cooling even though enterprise IT is still slower to follow. This is an industry moving from niche to infrastructure, but not yet to full standardization. That matters because it creates both growth and room for supplier churn.

Public market sizing points in the same direction even if the exact numbers vary by research house. Dell’Oro said in 2024 that liquid-based data-center thermal management would account for more than a third of the market by 2028. MarketsandMarkets put the global data-center liquid-cooling market at about US$4.07 billion in 2026 and projected US$27.65 billion by 2033. Reuters reported in March 2026 that liquid-cooling systems for AI data centers were a fast-scaling market and named Foxconn and Auras among key Taiwan-related players in the supply web. I do not put too much weight on any single TAM number. The more important conclusion is that the profit pool is moving up from chip-level parts into integrated racks, liquid loops, CDUs, and facility interfaces. That favors companies that can sell more than just one piece.

The cycle attributes are mixed. Auras sits in a capex cycle, a technology-iteration cycle, and a semiconductor-adjacent platform cycle. It is not defensive. It is not a pure semiconductor cycle either, because thermal content depends on rack power, platform architecture, and deployment timing rather than on wafer starts alone. In an upcycle, the variables that help most are rack density, liquid-cooling penetration, and share of system-level content. In a downcycle, the fragile points are pricing, working capital, and next-generation design wins. That makes Auras a higher-quality cyclical than a generic component vendor, but still a cyclical.

Geopolitics matters mostly through footprint and customer supply chains, not direct sanctions. Auras’s own reports link Thailand expansion to U.S.-China trade and customer supply-chain demands. Export sales represented 75% of 2024 revenue, which means FX and logistics matter. A stronger TWD is a translation and margin headwind for a Taiwanese exporter unless hedging and pricing keep up. The company’s sustainability report says it manages foreign-currency asset and liability positions based on working-capital needs and major-currency trends, which is prudent, but no exporter with this footprint is immune to a swift Taiwan-dollar move.

On horizontal comparison, Auras is best understood against three listed benchmarks and several private or specialist references. Asia Vital Components is the closest listed thermal peer in Taiwan: more directly comparable in cold plates and server thermals, richer-valued than old-school thermal vendors, and deeply exposed to AI cooling. Delta is much broader: its value comes from power electronics, infrastructure, and data-center systems as much as from thermals, which is why its multiple reflects a very different quality mix. Vertiv is not a direct manufacturing peer in cold plates, but it is a useful benchmark for what rack and facility cooling specialists look like at system scale. Boyd, Cooler Master, and CoolIT sit in the competitive field but are not straightforward listed comparables for a TWD valuation table. Public supplier reporting also varies: AVC and Cooler Master show up repeatedly in GB300 cold-plate coverage; Auras is repeatedly cited as supplying GB300 water blocks and related liquid modules; Boyd is explicit about being on Nvidia’s recommended-vendor list for cold plates, manifolds, and CDUs on GB200. That mixture tells you the field is deep, and no single supplier owns it.

† Auras price/sales is approximate, using TWD 99.59 billion market cap divided by 2025 revenue of TWD 23.28 billion. Other valuation metrics are Reuters mid-price statistics.

The business reason behind these differences is simple. AVC is the closest public thermal peer, so the market values it like a scaled AI-cooling winner. Delta gets a premium not because it competes head-on with Auras on every part, but because it sells into a broader system stack with more durable diversification and infrastructure exposure. Auras sits between those poles. It is more specialized than Delta and more system-leaning than legacy heat-sink vendors. That is why it deserves a better multiple than an undifferentiated thermal assembler, but also why it should not automatically command the same valuation insulation as a diversified infrastructure leader.

The current fundamentals are strong enough to support the story, but not so clean that the market can stop asking questions. The latest quarter was plainly good: revenue nearly doubled year on year, margins improved, and EPS hit a record quarter. Some third-party summaries conflict on the exact year-on-year growth number for Q1 2026, but the official quarterly revenue series and Goodinfo’s consolidated operating record imply about 93.7% growth, which is the figure used in this report. The bigger point is that growth is no longer coming from a scattered basket of consumer devices. It is being led by servers and liquid cooling. Management then lifted the 2026 revenue-growth goal to around 70% and said gross margin should hold above 30% as mix improves. That is what the market is trading.

The current bull case rests on four pieces of evidence. First, Auras is moving up the thermal stack into manifolds, QDs, and CDU systems rather than staying a one-part supplier. Second, the server and liquid-cooling mix have already risen sharply in management commentary. Third, monthly sales show the Q2 2026 run rate started from a high base rather than immediately fading. Fourth, the company’s overseas sales mix and Thailand expansion suggest it is serving global AI programs, not just domestic Taiwan demand. The current bear case rests on four different facts. First, the company still reports one revenue segment, so investors depend on management commentary for mix. Second, 2025 cash conversion turned weak during the scale-up. Third, public reporting indicates Auras is not the core cooling supplier for Nvidia’s Vera Rubin platform. Fourth, thermal competition is crowded, especially in cold plates and data-center liquid loops. None of these bear points kills the story. Together, they explain why the stock is no longer obviously mispriced despite the growth.

Valuation analysis

Auras’s historical valuation has clearly moved to a higher band. Using year-end price and EPS from Goodinfo, the stock ended 2020 and 2021 at roughly 15–17 times EPS, 2022 around 10 times, 2023 around 25 times, 2024 around 32 times, and 2025 around 36 times. At the 2026-06-18 close, Goodinfo showed about 30.6 times trailing earnings. That places the current multiple below the most euphoric part of the recent rerating, but still far above the pre-AI valuation center. The market is not valuing Auras as a cyclical rebound anymore. It is valuing it as a structural AI-cooling winner.

Peer valuation says the same thing, with one important warning. Reuters showed about 39.4 times trailing earnings for AVC and about 79.5 times for Delta on mid-price statistics. Auras at roughly 30.6 times looks cheaper than both. That does not automatically make it cheap. AVC itself is trading on scarce AI-thermal positioning, while Delta is a broader, higher-quality infrastructure compounder with different optionality. The right conclusion is that Auras is priced as a quality AI-cooling name, but not at the most extreme sector multiple on the board. The valuation discount to AVC partly reflects smaller scale and narrower disclosure; the discount to Delta reflects narrower scope and less diversification.

The cash-flow passthrough test is where the picture gets harder. Consolidated operating cash flow from 2021 through 2025 totaled about TWD 5.72 billion against net income of about TWD 8.08 billion, for an operating-cash-flow-to-net-income ratio near 0.71. Excluding the 2025 ramp year, the ratio over 2021–2024 was above 1.0. That suggests Auras does not have a chronic earnings-quality problem, but it does have a recent scaling problem: working capital and capex ran ahead of reported profit. The company’s own and Goodinfo’s historical capex data also show a sharp jump in fixed-asset spending in 2024–2025 versus earlier years. I therefore treat 2025’s negative free cash flow as heavily distorted by growth investment, but I do not ignore it. For maintenance capex, the best public approximation is around depreciation, roughly TWD 0.6–0.8 billion a year, while 2025 total fixed-asset additions were much higher, implying that most of the recent spending was growth capex rather than bare maintenance.

That leads to a normalized owner-earnings approach rather than a raw DCF on 2025 free cash flow. The valuation below assumes that the growth build is front-loaded, that maintenance capex settles near TWD 0.75 billion, and that the relevant question is not “what was 2025 free cash flow?” but “what owner earnings can the business sustain once the liquid-cooling ramp stops eating so much working capital?” I do not think a single DCF is the right main lens here because too much depends on platform timing and margin durability. Forward owner earnings and scenario multiples are a better fit for a cyclical hardware business in the middle of a structural upshift.

This is valuation-scenario analysis inside a research framework, not investment advice. The business reason behind the table is that Auras no longer needs heroic multiple expansion to justify a respectable value. It needs the earnings ramp to keep showing up in cash, and it needs its system-content share to travel into the next platform wave. At today’s price, the market is already paying for a healthy part of the base case. It is not paying only for the conservative case.

The expectation gap is therefore narrow and operational. Investors do not need another proof that AI racks are hot. They need proof that Auras can keep taking more wallet share per rack. The metrics that matter most are liquid-cooling revenue mix, server revenue mix, gross margin around the 30% line, monthly sales conversion into quarterly profit, and cash conversion as Thailand capacity scales. At the next important update, I think the market will care more about liquid-cooling mix, CDU/manifold traction, and 12-month cash generation than it will about abstract AI demand. That is the natural progression once a theme stock becomes an earnings stock.

The margin-of-safety check is not friendly. Against the conservative fair-value band of roughly TWD 800–860, the current stock price trades at a premium, so there is no obvious margin of safety on conservative assumptions. The most fragile assumption in the base case is not revenue growth by itself. It is the belief that mix gains keep gross margin around 30% while working capital normalizes. If I haircut that assumption to 70% of the intended improvement, the base-case valuation falls into the high-TWD-800s to low-TWD-900s. If earnings were flat for the next three years and the multiple drifted back toward a mid-teens thermal-hardware level, annualized returns from the current price would be poor. This is a good company but no longer an easy price. Waiting for a better entry is rational unless one believes Auras will prove a much stronger next-generation share position than the public record currently supports.

Margin-of-safety sufficiency verdict: not obvious.

Risk analysis

The largest business risk is next-generation share leakage. Public reporting repeatedly ties Auras to GB300 cold plates and manifolds, and management has talked openly about demand from Nvidia, AMD, and ASIC servers. But public reporting also says Auras is not the main supplier for Nvidia’s Vera Rubin platform, while peers such as AVC and Cooler Master continue to show up in cold-plate discussions. The probability is medium, the impact is high, and the observable indicators are references to supplier lists, management commentary on program ramps, and whether liquid-cooling mix keeps rising through 2026. The transmission path is direct: a share loss in the highest-content platforms would hurt mix first, then gross margin, then the valuation multiple that the market is willing to pay for structural share gain.

The second business risk is that thermal competition becomes more systematized and squeezes pricing. Boyd is explicit about supplying Nvidia-recommended cold plates, manifolds, and CDUs for GB200. Eaton makes similar claims around GB200 ecosystem cooling. Reuters has also shown how fast the liquid-cooling field is consolidating, with Ecolab buying CoolIT and Schneider buying Motivair. That means Auras is not entering a sleepy niche. It is entering a market where industrial companies, data-center specialists, and Taiwan thermal vendors are all trying to move up the stack. The probability is medium, the impact is medium to high, and the key indicators are gross margin, new customer wins, and whether Auras’s content expands from cold plates into more rack plumbing fast enough to stay differentiated.

The biggest financial risk is cash absorption rather than leverage in the classic sense. The company’s own history shows it can convert earnings into cash in normal years, but 2025 proved that a rapid AI ramp can reverse that quickly. Receivables and inventory are already large in absolute terms, and fixed-asset additions have accelerated as Thailand expands. The probability is medium, the impact is high, and the indicators are 12-month operating cash flow, receivables as a share of sales, inventory days, and how much of capex remains expansionary into 2027. The transmission path is familiar: if cash lags earnings too long, the market stops paying a structural-growth multiple and starts paying a manufacturing multiple.

FX is a quieter but real risk. Auras sold 75% of 2024 revenue outside Taiwan, with Europe and the Americas accounting for a much larger share than a year earlier, while the company’s own sustainability report says it manages exposures in currencies such as the U.S. dollar and Chinese yuan. A stronger TWD reduces translated revenue and can squeeze margins if pricing or hedging lags. The probability is medium, the impact is medium, and the indicators are the USD/TWD rate, management commentary on hedging, and gross-margin behavior during quarters when the TWD strengthens abruptly. This is not the kind of risk that breaks a company. It is exactly the kind that interrupts a momentum narrative.

A governance and external risk worth tracking is concentration of market belief around a few public narratives. Auras still discloses one revenue segment, which leaves investors dependent on management summaries and third-party supply-chain reporting for the pieces that matter most. When disclosures are that aggregated, the share price becomes more sensitive to rumors, broker notes, and selective supply-chain checks. The probability is medium, the impact is medium, and the indicators are discrepancies among public summaries, clarifying exchange filings, and whether management starts giving more structured product-mix disclosure. This is not an accounting-red-flag claim. It is a disclosure-quality claim. The business is changing faster than the audited segment reporting.

Cross-synthesis summary

Looking vertically across the whole journey, the capability Auras has genuinely proven is not “inventing cooling.” It is translating thermal complexity into manufacturable products fast enough for demanding hardware customers. That skill began in notebooks, matured across graphics cards, motherboards, servers, and gaming rigs, and is now being re-monetized in liquid cooling. The crucial point is that liquid cooling did not appear out of thin air. Auras invested into it years before capital markets cared, built around cold plates and hybrid data-center cooling concepts, and only then found itself in a market where AI racks made those investments economically central. That is why the current story has more substance than a simple momentum stock.

Past success came from several things at once. There was era tailwind in the rise of portable computing, gaming hardware, and denser servers. There was management capability in moving production into China early and Thailand later. There was technical accumulation in heat pipes, vapor chambers, and later liquid systems. There was also luck in timing: AI infrastructure has turned thermal management into a choke point exactly when Auras had the product set to attack it. I do not think luck alone explains the outcome. But I also do not think the market should price today’s economics as though the next five years are already industrialized. Success factors are present; permanence is still being tested.

Horizontally, Auras’s real advantage over competitors is breadth within the thermal stack without being as sprawling as a full infrastructure conglomerate. Compared with a narrower legacy thermal supplier, Auras can now sell a richer bundle. Compared with Delta or Vertiv, it is still narrower and more exposed to the fortunes of a few high-density platforms. Its weakness is therefore partly structural and partly transitional. Structural, because it will probably never have Delta’s breadth or Vertiv’s full infrastructure moat. Transitional, because the market still does not know whether Auras’s current liquid-cooling share is a stable franchise or the high-water mark of a fast platform handoff.

The current valuation is rewarding future success at least as much as past success. On trailing earnings, the stock is rich by its own long history. On forward earnings, it is not absurd if 2026–2027 play out cleanly. That is the central tension. Investors are not massively overpaying for a fantasy, but they are paying enough that execution misses matter. The market is most likely misjudging two things. It may be underestimating how much richer Auras’s wallet share per rack can become if manifolds, quick-connects, monitoring, and CDU scale faster than cold plates alone. It may also be underestimating how much next-generation share uncertainty and working-capital drag can lower the value of that growth. Those are opposite errors, and the share price sits between them.

Over the next year, the critical variables are liquid-cooling mix, gross margin, and cash conversion. Over three years, the crucial question is whether Auras remains inside the winning part of the next platform map, not just GB300 but the programs that follow it in Nvidia, AMD, and ASIC ecosystems. Over five years, the deepest issue is whether Auras becomes a genuinely more durable systems supplier or remains a fast-moving thermal vendor that periodically re-rates with each platform cycle. The company becomes a much better investment if the stock price resets while liquid-cooling mix and cash conversion keep improving, or if management proves that next-generation share is broader than the market currently assumes. The original judgment should be revisited if gross margin slips below 27% for multiple quarters, if liquid-cooling mix stalls around current levels, if cash conversion remains poor after the Thailand build, or if public evidence accumulates that Auras is being pushed out of the highest-content platforms.

Bull reasons

• Auras has spent more than a decade building liquid-cooling capability and now sells a broader rack-level bill of materials than a plain cold-plate vendor.
• The latest quarter showed a strong mix shift, with management saying server revenue reached 76% and liquid cooling exceeded 55% of sales, supporting a path to higher margin.
• 2026 monthly sales after the strong Q1 did not collapse; January through May revenue was still up 80.5% year on year.
• Production diversification into Thailand responds directly to customer demand for non-China supply and should improve strategic relevance with hyperscale and server customers.
• Official Nvidia platform disclosures confirm that the hottest new rack-scale systems are fully liquid cooled, keeping the industry backdrop favorable for Auras’s product set.

Bear reasons

• Auras still reports one revenue segment, so investors do not get audited disclosure on how much profit comes from cold plates, manifolds, CDUs, or legacy air cooling.
• Consolidated cash conversion deteriorated sharply in 2025, when operating cash flow turned negative despite higher earnings.
• Public reporting indicates Auras is not the core cooling supplier for Nvidia’s Vera Rubin platform, which weakens the assumption that today’s share automatically persists into the next generation.
• Competition is not limited to Taiwan thermal peers; Boyd, Eaton, CoolIT, Motivair, Vertiv, AVC, and Cooler Master all crowd the liquid-cooling field.
• The stock trades far above its pre-AI valuation center, leaving less room for error if gross margin or liquid mix softens.

Pre-mortem

A plausible three-year failure script is that Auras rides GB300 volume through 2026, but 2027 platform transitions favor AVC, Cooler Master, and other qualified suppliers on the highest-value next-generation cold-plate and rack-plumbing positions. Liquid-cooling mix stalls around the mid-50s instead of moving above 60%–70%, gross margin falls back from around 30% toward 25%–26%, and the market stops valuing Auras as a structural AI winner. If the multiple compresses from around 19x forward owner earnings toward 12x-14x while earnings also flatten, the stock could lose half its value.

A second failure script is more subtle. Demand stays decent, but Thailand expansion, receivable growth, inventory build, and a stronger TWD keep cash generation weak. Earnings keep rising, but free cash flow stays poor and capex remains elevated. The market then decides Auras is a capital-hungry manufacturer with theme exposure, not a self-funding compounder. In that case the stock does not need an earnings collapse to fall hard; it only needs the narrative multiple to shrink.

The final judgment is this: Auras is a better business than its old thermal-supplier label suggests, but the stock already reflects much of that discovery. The company has done the hard part of getting inside the liquid-cooling profit pool. It has not yet done the harder part of proving that this new position will survive platform transitions, competition, and the growing pains of a global capacity ramp. I think the business is worth following closely and worth owning at the right price. I do not think the current price offers a wide enough cushion for new money.

What worries me most is not whether AI demand exists. It does. The worrying part is the combination of thin product-level disclosure, a recent cash-conversion dip, and visible uncertainty around next-generation share. What would change my mind? Either a cheaper entry, or clear evidence over the next several quarters that Auras can translate the liquid-cooling mix shift into stronger cash generation while also broadening beyond GB300-dependent optimism.

Company-profile scores

• Fundamental quality: high
• Growth: high
• Moat: medium
• Financial soundness: medium
• Management credibility: medium to high
• Valuation attractiveness: low
• Risk level: medium to high
• Suitable investor type: long-term growth, with cyclical tolerance

Investment rating

• Rating: Hold
• One-line thesis: Real AI-liquid-cooling share gains are visible, but the current price already discounts much of the 2026–2027 mix and margin improvement.
• Current-price classification: acceptable hold
• Whether to wait for a better price: yes; a materially better entry begins below TWD 690, ideally with evidence that liquid-cooling mix continues rising and gross margin stays near 30%; the opportunity cost of waiting is that Auras could still rerate if CDU, manifold, and ASIC programs surprise on the upside.
• Target holding horizon: 1–3 years
• Expected annualized return: conservative about -8% to -5%; base about 0% to +3%; optimistic about +6% to +10%
• Max-loss risk: roughly 50% in a harsh de-rating scenario, triggered by next-generation share loss plus a margin reset and weak cash conversion
• Reassessment-trigger signals: gross margin below 27% for two consecutive quarters; liquid-cooling mix fails to move beyond the mid-50s by late 2026; 12-month operating cash flow remains weak after Thailand expansion normalizes; management or supply-chain evidence points to share loss in next-generation Nvidia or ASIC programs; working-capital intensity keeps rising faster than revenue

【Ideal Buy Price】640–690 TWD Basis: at least a 20% margin of safety below the conservative fair-value range of roughly TWD 800–860 derived from normalized 2026 owner earnings and an 18x-19x multiple.

Acceptable hold price: 860–1,220 TWD. Clearly overvalued price: above 1,300 TWD.

Key data table

‡ 2024 gross and operating margins are available in third-party synthesis, but this report relies more heavily on 2025 and 2026 Q1 audited or management-reported figures where the current debate sits.

The business reading on these numbers is that Auras is scaling profit faster than revenue because mix is improving, but cash has not yet caught up. In other words, the strategic transition is visible in the P&L and still incomplete in the cash-flow statement.

Tracking dashboard

The reason to track these particular signals is that they bridge the gap between the narrative and the business. Revenue growth without liquid-mix gains is less valuable than it looks. Margin without cash conversion is less durable than it looks. Auras publishes monthly sales on its own website; market services track receivables and inventory ratios; management events and news flow are where next-platform evidence usually appears first.

Research uncertainties

• The company still reports one operating line, which limits audited product-level analysis.
• Public sources differ on some quarterly growth summaries; this report privileges the official quarterly-series arithmetic where conflicts appear.
• Exact customer concentration and exact program-level share on Nvidia, AMD, and ASIC platforms are not publicly disclosed in a way that allows a clean verified share model.
• The public record supports strong GB300 participation, but next-generation supply-chain positioning remains partly opaque and still changes quickly.
• Maintenance capex is estimated from depreciation and historical spending behavior, not disclosed directly by management as a separate figure.

Sources

The most important primary materials used here were Auras’s 2025 annual report, Auras’s English sustainability reports, the company’s monthly-sales page, the TPEx/basic company profile data, and company-related exchange announcements surfaced through public market services. For industry context and current platform requirements, the report relied mainly on Nvidia’s official GB200 and GB300 pages, Lenovo and Supermicro product documentation, Uptime Institute, Dell’Oro, and Reuters. For current public-market valuation and cross-checks, it used Goodinfo, Reuters market pages, Yahoo Finance Taiwan, and selected Taiwan financial press reports where management guidance or supply-chain commentary was necessary and clearly attributed.

【Valuation Range】

• current: 1,070 (close as of 2026-06-18)
• bear (conservative · ideal buy zone): [640, 690]
• base (fair · acceptable hold zone): [860, 1220]
• bull (optimistic · above the clearly-overvalued line): [1300, 1450]

Other tickers mentioned

• 3017.TW: Asia Vital Components, the closest listed Taiwan thermal peer for AI liquid cooling and cold-plate comparison.
• 2308.TW: Delta Electronics, a broader power-and-cooling benchmark that shows what system-scope diversification looks like.
• VRT.US: Vertiv, the relevant benchmark for CDU and data-center cooling infrastructure economics.
• NVDA.US: Nvidia, whose GB200 and GB300 platforms define the current liquid-cooling demand wave.
• AMD.US: management-cited AI-server demand vector outside Nvidia.
• AMZN.US: public reports cite Amazon ASIC server programs as a growth driver for Auras’s liquid-cooling business.
• GOOGL.US: public reports cite Google as a CDU-related customer and because hyperscale liquid-cooling demand matters for Auras.
• SMCI.US: public reports connect Auras to water-cold-plate supply into Supermicro-related programs.
• ECL.US: buying CoolIT, relevant because it shows strategic consolidation around liquid-cooling specialists.

This report is based on public information and does not constitute investment advice. Markets carry risk; invest with caution.