AI bubble stocks

Core demand for GPUs comes from hyperscale data centers, making them highly sensitive to CapEx budget rationalization.

NVIDIA Corporation stands as a prominent provider of advanced graphics, computational, and networking solutions, operating across the United States, Taiwan, China, and numerous international markets. Its Graphics division encompasses GeForce GPUs, central to PC gaming and personal computing experiences, along with the GeForce NOW cloud gaming service and its supporting infrastructure, as well as dedicated solutions for various gaming platforms. For professional visualization, it provides Quadro and NVIDIA RTX GPUs for enterprise workstations, further offering vGPU software designed for cloud-centric visual and virtual computing, automotive platforms for in-vehicle infotainment, and the Omniverse software suite, facilitating 3D design and virtual world creation. The Compute & Networking segment is a cornerstone for AI, high-performance computing (HPC), and accelerated data center platforms. It integrates Mellanox networking and interconnect solutions, delivers automotive AI Cockpit technologies, fosters autonomous driving development through strategic agreements, and offers comprehensive autonomous vehicle solutions. This segment also manufactures cryptocurrency mining processors, supplies Jetson platforms for robotics and other embedded applications, and offers enterprise AI software, including NVIDIA AI Enterprise. These diverse offerings find widespread application across the gaming, professional visualization, data center, and automotive sectors. NVIDIA distributes its portfolio through a broad ecosystem, engaging original equipment and device manufacturers, system integrators, add-in board makers, retail channels, software vendors, internet and cloud service providers, automotive companies (both manufacturers and tier-1 suppliers), mapping firms, nascent technology ventures, and other industry stakeholders. A notable strategic partnership exists with Kroger Co. Founded in 1993, NVIDIA Corporation maintains its corporate headquarters in Santa Clara, California.

SK Hynix is a concentrated HBM supplier and therefore has strong exposure to incremental AI-server demand and constrained-capacity pricing.

SK Hynix is a principal HBM and server-memory supplier, so incremental AI capex directly increases demand for its constrained products.

AI-server demand is increasing requirements for HBM and other advanced memory while supply remains constrained. SK hynix, a direct memory producer, should benefit through higher memory pricing, favorable mix, and stronger semiconductor revenue, subject to eventual capacity additions and cycle reversal.

The hyperscaler financing wave supports additional AI-server deployments and therefore demand for HBM. SK Hynix is a leading HBM supplier positioned to capture premium pricing and allocation strength while capacity remains constrained.

Subject to historical memory cyclicality and sentiment regarding AI capex sustainability.

Micron directly supplies HBM and server memory used in AI accelerators and servers, giving it strong volume and pricing leverage if compute investment continues.

The larger AI-server deployment increases demand for high-bandwidth memory used with accelerators. Micron is a direct HBM supplier with sold-out capacity, creating positive volume, mix, utilization, and pricing exposure while shortages may delay downstream server deployments.

NVIDIA accelerator-volume and outlook changes transmit directly to HBM3E demand. Micron is one of the concentrated HBM suppliers, so sustained AI-capex expectations support HBM pricing, mix, utilization, and memory margins, while allocation constraints may limit near-term system shipments.

Micron Technology, Inc. is a global leader specializing in the development, manufacture, and sale of advanced semiconductor memory and storage solutions. Its operations are structured across four primary business segments: Compute and Networking, Mobile, Storage, and Embedded. The company's product portfolio encompasses a range of memory and data storage technologies. These include high-speed, low-latency Dynamic Random Access Memory (DRAM) components for rapid data retrieval; non-volatile, re-programmable NAND flash storage devices; and fast-read, non-volatile, re-writable NOR memory chips. These innovative solutions are offered under its well-known Micron and Crucial brands, as well as through private label partnerships. Micron's extensive offerings cater to a diverse array of markets and applications. This includes critical infrastructure like cloud servers and enterprise data centers, personal computing (client and graphics), mobile devices such as smartphones, networking equipment, automotive systems, industrial applications, and various consumer electronics. The company utilizes a multi-faceted approach to market its products, employing a direct sales force, independent sales representatives, and a network of distributors and retailers. Additionally, it leverages a web-based direct sales channel and collaborates with various channel and distribution partners. Founded in 1978, Micron Technology, Inc. is headquartered in Boise, Idaho.

Eoptolink is a rapidly expanding 800G/1.6T supplier integrated into hyperscale data-center demand.

Eoptolink participates in constrained 800G and 1.6T optical-module supply supporting AI infrastructure. It benefits from continued bandwidth migration, but faces competitive and procurement pressure if Semtech’s dedicated SiGe allocation improves the availability and qualification position of Semtech-based modules.

Lumentum directly supplies optical transceivers and lasers used as AI-cluster bandwidth and port counts expand.

Lumentum supplies lasers, optical components and transceiver products for high-speed AI data-center interconnects. The move from 800G toward 1.6T and 3.2T increases content per cluster and can improve order visibility and pricing when InP and EML supply is tight.

Marvell’s custom ASIC and optical-DSP businesses are highly sensitive to hyperscaler AI-capex cycles.

Extremely high dependence as hyperscalers are the primary customers for custom ASICs and 1.6T optics.

The optical portion of the agreement increases demand for high-speed DSPs, SerDes, and connectivity silicon used in AI clusters. Marvell may face competition from Qualcomm in custom silicon, but the planned 1.6T networking transition is a positive second-order demand signal for its custom-connectivity and optical-DSP businesses.

A sustained NVIDIA and hyperscaler AI-buildout outlook supports Marvell's custom silicon, optical DSP, and high-speed interconnect demand. The exposure is meaningful but less direct than NVIDIA's because revenue conversion depends on customer-specific networking and accelerator programs.

Innolight can benefit from constrained qualified 800G and 1.6T capacity, though it remains dependent on upstream lasers and substrates.

The target requires networked accelerator clusters rather than isolated servers, increasing demand for 800G, 1.6T, and future co-packaged optical links. Zhongji Innolight is identified as a leading optical-transceiver supplier with 1.6T production and CPO capability, providing direct exposure to AI-networking demand.

Broadcom captures AI capex through custom silicon and networking components, while remaining sensitive to hyperscaler spending cycles.

Broadcom’s custom accelerators, networking silicon and optical products depend on large AI-cluster procurement, so delayed frontier or hyperscaler projects can defer related orders.

Broadcom’s custom silicon, networking, and optical demand depends heavily on large AI-cluster procurement.

Broadcom supplies custom AI accelerators, networking silicon and optical connectivity used in large clusters. A reassessment of AI infrastructure demand could defer custom-silicon and networking procurement, although diversified hyperscaler demand and the need for high-speed interconnect provide partial resilience.

AMD’s AI accelerators directly participate in the demand cycle, although its exposure is constrained by dependence on external foundry and packaging capacity.

AMD's AI chipsets rely on advanced packaging, but it doesn't sell packaging or tools. It is a consumer of capex.

AMD’s data-center accelerators participate in the same cloud-capex cycle as NVIDIA’s products. If weaker neocloud economics cause providers to defer deployments, accelerator orders and related advanced-packaging and HBM demand could be delayed.

AMD is exposed downstream because its AI accelerators require HBM and advanced packaging capacity. The new lower DISCO outlook creates a modest risk of delayed back-end capacity or equipment conversion, although sustained AI accelerator demand remains an offsetting positive.

Coherent is identified as an 800G and 1.6T optical-transceiver supplier required by AI clusters.

Directly impacted by hyperscale transceiver buy-cycles for AI clusters.

Production deployment of custom accelerator clusters increases demand for optical transceivers, lasers, and photonic components connecting racks and data-center fabrics. Coherent can benefit from higher 800G and 1.6T networking volumes, without a disclosed project-specific order.

Applied Optoelectronics is directly exposed to 800G and 1.6T transceiver demand, although it is downstream from the substrate layer.

The development strengthens the demand signal for the 800G and 1.6T optical transceivers that Applied Optoelectronics manufactures for hyperscale networks. The benefit is material but less certain than for larger component suppliers because customer qualification, manufacturing capacity, and upstream laser availability constrain conversion of demand into revenue.

Super Micro captures AI-server system demand but not the upstream memory-fab capital-equipment cycle.

More specialized AI-cloud capacity expands the market for pre-engineered liquid-cooled rack systems.

Super Micro integrates memory-intensive AI and general-purpose servers but has less procurement leverage than hyperscalers. NAND and memory scarcity can raise component costs, delay deliveries, and compress system margins before repricing reaches customers.

Super Micro integrates memory-intensive AI and general-purpose servers and has less procurement leverage than the largest hyperscalers. Scarcer DRAM and HBM could raise component costs, delay deliveries, and compress system margins before repricing reaches customers.

Alpha Compute Corp. (Nasdaq: ALP) is a high-performance GPU infrastructure and confidential-compute technology company built for the artificial intelligence economy. Structured as a holding company focused on sovereign AI compute, Alpha Compute owns the infrastructure that powers modern intelligence and enforces privacy directly at the hardware level. That foundation positions the company to build and acquire businesses that depend on confidential compute and artificial intelligence. Alpha Compute's mission is to support clients, subsidiaries, and partners across critical sectors such as finance, defense, intelligence, and media as they adapt to a rapidly changing AI landscape. The company provides the core framework any organization needs to run secure, confidential computing environments. Alpha Compute is domiciled in the British Virgin Islands, with offices in New York, Los Angeles, Miami, Amsterdam, and Toronto.

Cerebras Systems Inc. is a leading innovator in artificial intelligence infrastructure. The company develops and manufactures an advanced AI compute platform, integrating proprietary hardware systems and software. This platform is delivered in rack-mountable units, suitable for deployment in data centers, scaling all the way up to supercomputer-level capabilities. At its core is the groundbreaking Wafer-Scale Engine (WSE), a unique chip that encompasses an entire silicon wafer. This innovation is specifically engineered to deliver superior performance and speed compared to conventional GPUs, addressing the intensive computational demands of inference, Generative AI, and a broad spectrum of other AI applications. Cerebras serves a diverse clientele, including leading hyperscalers, advanced foundation model laboratories, AI-native and digital-first businesses, large enterprises, and key players in Sovereign AI initiatives. With operations spanning the United States, Europe, the Middle East, Africa, and other international markets, Cerebras maintains a significant global footprint. The company was established in 2015 and is headquartered in Sunnyvale, California.

Shanghai Iluvatar CoreX Semiconductor Co., Ltd. engages in the research, design, and deployment of advanced GPGPU products and AI computing solutions in China. It offers various GPGPU chips and accelerators, along with its proprietary and full-stack software development kits, including performance-optimized libraries, AI training frameworks and inference engines, enabling powerful AI applications across training and inference scenarios. It primarily develops and sell GPGPU accelerators, the accelerators integrate GPGPU chips with memory modules, PCB, and other critical components, forming complete computing accelerators that are ready for direct application in the end computing systems. TG Series, a training-focused product line; and ZK series, a specialized product line designed for cloud and edge inference applications. The AI Computing Solutions provides GPGPU Servers and GPGPU Clusters. The company sells its products through direct sales force and distributors. Shanghai Iluvatar CoreX Semiconductor Co., Ltd. was founded in 2015 and is based in Shanghai, China.

Cambricon Technologies Corporation Limited specializes in designing fundamental processor chips engineered for diverse intelligent systems, including cloud-based servers, edge computing devices, and smart terminals. Their product lineup features intelligent processor intellectual property (IP) for terminal applications, along with smart accelerator cards and AI acceleration cards specifically developed for edge environments. Complementing their hardware, they also provide Cambricon NeuWare, a comprehensive software development platform designed to support their full spectrum of intelligent processors across cloud, edge, and endpoint domains. The company, established in 2016, has its headquarters in Beijing, China.

Applied Digital’s expansion requires external capital and customer conversion before facilities generate cash, so it is highly exposed to a reduction in speculative AI infrastructure orders.

Crealights Technology Co. Ltd. is a provider of optoelectronic interconnection products, including optical transceivers, active optical cables, and active electrical cables. The company's products are used for high-speed data transmission in applications such as AI data centers, and it possesses capabilities from silicon photonics chip design to high-speed optical module manufacturing.

Credo Technology Group Holding Ltd (CRDO) specializes in delivering advanced high-speed connectivity solutions for both optical and electrical Ethernet applications. Its operational reach extends globally, encompassing the United States, Mexico, Mainland China, Hong Kong, and various other international regions. The company's product offerings include integrated circuits (ICs), active electrical cables (AECs), and SerDes chiplets, all developed utilizing its proprietary serializer/deserializer (SerDes) and digital signal processor (DSP) technologies. Beyond hardware, Credo also provides intellectual property (IP) solutions, particularly focusing on SerDes IP licensing. Established in 2008, the firm's corporate headquarters are situated in San Jose, California.

Arista is a major switch vendor directly driving and supplying AI datacenter networking demand that requires 800G and 1.6T optics.

Arista supplies Ethernet switching infrastructure that connects the 800G and 1.6T optical modules used in AI clusters. MaxLinear's high-volume hyperscale deployments support continued switch-fabric demand, while scarce optics can delay shipments and increase procurement costs.

Fabrinet can capture outsourced assembly and packaging volumes as optical vendors scale 800G and 1.6T programs.

Fabrinet is an outsourced manufacturing and packaging beneficiary. A stronger 800G and 1.6T production cycle can increase optical-transceiver, silicon-photonics, and CPO assembly volumes, although Fabrinet captures manufacturing throughput rather than the full component economics.

Fabrinet provides optical packaging and precision optical, electro-mechanical, and electronic manufacturing services in North America, the Asia-Pacific, and Europe. The company offers a range of advanced optical and electro-mechanical capabilities in the manufacturing process, including process design and engineering, supply chain management, manufacturing, printed circuit board assembly, packaging, integration, final assembly, and testing. Its products include switching products, including reconfigurable optical add-drop multiplexers, optical amplifiers, modulators, and other optical components and modules that enable network managers to route voice, video, and data communications traffic through fiber optic cables at various wavelengths, speeds, and over various distances. The company’s products also comprise transceivers, tunable lasers, and transponders; and active optical cables, which provide high-speed interconnect capabilities for data centers and computing clusters, as well as Infiniband, Ethernet, fiber channel, and optical backplane connectivity. In addition, it provides solid state, diode-pumped, gas, and fiber lasers used in semiconductor processing, biotechnology and medical device, metrology, and material processing industries; and differential pressure, micro-gyro, fuel, and other sensors used in automobiles, as well as non-contact temperature measurement sensors for the medical industry. Further, the company designs and fabricates application-specific crystals, lenses, prisms, mirrors, laser components, and substrates; and other custom and standard borosilicate, clear fused quartz, and synthetic fused silica glass products. It serves original equipment manufacturers of optical communication components, modules and sub-systems, industrial lasers, automotive components, medical devices, and sensors. The company was incorporated in 1999 and is based in George Town, the Cayman Islands.

A delayed frontier-AI roadmap can push out wafer and advanced-packaging orders for accelerators and custom silicon.

TSMC captures foundry and packaging revenue from the AI capex cycle while carrying less direct deployment and utilization risk than hyperscalers.

The contracted GPU expansion supports upstream demand for advanced-node wafers and advanced packaging used in AI accelerators. Nscale's execution problems can delay the procurement schedule, but the broader shortage and large contracted capacity requirement support continued foundry utilization if deployments proceed or shift to other operators.

Additional AI infrastructure spending increases demand for advanced-node wafers and CoWoS packaging used by GPU, accelerator, and custom-silicon suppliers. Packaging constraints may delay shipments but support utilization and pricing.

Tower manufactures silicon-photonics PICs used in 1.6T optical transceivers, giving it direct foundry exposure to AI optical demand.

Tower is a silicon-photonics foundry beneficiary. The MXL update strengthens the demand read-through for alternative 1.6T optical architectures and wafer capacity, although Tower’s exposure is indirect and dependent on customer qualification.

Vertiv directly supplies thermal-management and liquid-cooling systems used in high-density AI facilities, giving it clear but non-project-specific physical-capex exposure.

Power-constrained AI facilities still require high-density power distribution and thermal-management systems, including liquid cooling. Delays can defer near-term orders, but redesigns, retrofits, and cooling requirements can preserve demand as projects are reconfigured or relocated.

Vertiv Holdings Co, headquartered in Columbus, Ohio, is a global provider of essential digital infrastructure technologies and comprehensive lifecycle services. The company specializes in the design, manufacturing, and servicing of critical systems vital for data centers, communication networks, and diverse commercial and industrial applications. Its extensive product range encompasses AC and DC power management, thermal control solutions, integrated rack systems, modular designs, and sophisticated management platforms for monitoring and governing digital environments. These solutions are fundamental to the operation of numerous modern services, including e-commerce, online banking, file sharing, video on-demand, energy storage, wireless communications, the Internet of Things, and online gaming. In addition to its hardware offerings, Vertiv delivers a full spectrum of lifecycle management services. These include professional deployment, ongoing maintenance, and optimization, supported by predictive analytics. Specific service capabilities extend to preventative maintenance, acceptance testing, engineering and consulting, performance assessments, remote monitoring, training, spare parts, and specialized critical digital infrastructure software. Vertiv's products are marketed under prominent brands like Liebert, NetSure, Geist, E&I, Powerbar, and Avocent. The company serves a wide array of sectors such as social media, financial services, healthcare, transportation, retail, education, and government. Its global reach spans the Americas, Asia Pacific, and Europe, the Middle East, and Africa, facilitated by a robust network of direct sales professionals, independent representatives, channel partners, and original equipment manufacturers.

Amkor can receive overflow advanced-packaging and testing work from the AI-connectivity supply chain, but its exposure is indirect and qualification-dependent.

Continued AI investment can increase outsourced assembly, advanced packaging and testing volumes for accelerators and HBM. The downside is that an AI-capex slowdown would reduce or postpone outsourced backend demand.

ASE provides outsourced assembly, packaging, and testing capacity that can absorb AI-related overflow.

ASE is a positive second-order stakeholder because AI-related back-end demand and equipment investment can increase outsourced packaging and testing volumes. ASE has substantial packaging and testing capital expenditure, so it can capture constrained advanced-packaging demand, although the investment cycle also creates depreciation and execution risk.

Advantest has meaningful backend equipment exposure because AI chips and HBM packages require more testing and qualification.

More advanced AI chips and constrained packaging capacity increase backend test intensity and equipment requirements. Advantest, a major automated test-equipment supplier, benefits as OSATs and chipmakers expand testing capacity for complex AI and high-bandwidth-memory products.

Applied Materials’ semiconductor-equipment revenue follows customer wafer-fab and packaging capital expenditure with a lag, making it exposed to postponed AI-related capacity projects.

More AI infrastructure spending supports leading-edge logic, memory and packaging capex, which increases equipment volume over the semiconductor investment cycle.

Applied Materials supplies equipment used in leading-edge logic, memory, and advanced-packaging expansion. If a pause causes chipmakers to defer AI-related capacity projects, equipment orders would weaken with a lag, even though existing bottlenecks remain tight.

Sustained AI-accelerator and HBM demand can induce additional advanced-node, memory, and packaging capital expenditure. Applied Materials supplies deposition, etch, and materials-engineering equipment, but the revenue effect follows customer capex with a lag.

ASMPT is an advanced-packaging equipment supplier positioned to benefit as manufacturers add packaging capacity for AI accelerators.

ASMPT supplies advanced-packaging equipment used as AI accelerator and HBM capacity expands. Stronger package-board demand supports the broader packaging-capex cycle, but the read-through from Ishii Hyoki's PCB equipment backlog to ASMPT is indirect and dependent on customer tool qualification.

Besi is an advanced-packaging equipment supplier; its exposure is to bonding-tool capex rather than direct DRAM or HBM procurement.

BESI is an advanced-packaging equipment supplier. Stronger back-end semiconductor demand and AI-related capacity expansion can increase orders for packaging tools, although equipment revenue depends on customer qualification and facility ramp schedules.

Lam Research benefits from AI-driven memory and packaging expansion but faces deferred equipment orders when those projects are postponed.

Lam Research benefits when memory, wafer-fabrication and advanced-packaging suppliers expand capacity for AI accelerators and HBM. A demand reassessment can defer those equipment orders, creating a lagged negative effect even if existing semiconductor bottlenecks remain tight.

Additional HBM, NAND, and advanced semiconductor capacity requires etch and deposition equipment. Lam can benefit from upstream fab and packaging capex induced by the AI infrastructure cycle, although equipment orders occur with a lag.

ASML captures longer-cycle lithography demand from allied and sovereign semiconductor capacity expansion, but not the near-term AI deployment spend directly.

ASML benefits from eventual leading-edge lithography investment induced by persistent AI demand, but with a longer timing lag than packaging suppliers.

Lower AI-related logic and memory investment would delay fab expansions and therefore EUV and DUV lithography orders. ASML benefits several steps upstream from AI infrastructure and would face delayed equipment revenue rather than an immediate demand collapse.

A durable NVIDIA-led accelerator cycle can require additional advanced-node, HBM, and packaging capacity, eventually supporting lithography-equipment orders. ASML's exposure is indirect and lagged because customers must first convert demand and bottleneck economics into fab-capex commitments.

Modine can capture thermal-management spending as AI rack density increases and liquid cooling becomes necessary.

Higher AI rack densities exceed the practical limits of conventional air cooling and require direct-to-chip or related liquid systems. Modine can capture thermal-management equipment demand as planned campuses move from construction into fit-out.

Eaton captures AI infrastructure spending through power-management and liquid-cooling equipment, but cooling is not its sole business.

Eaton supplies power-management equipment and liquid-cooling systems through Boyd Thermal. The LU-VE contract indicates that hyperscaler build-outs are generating demand not only for heat exchangers but also for the electrical and thermal infrastructure required to energize and cool AI facilities, supporting Eaton’s backlog and data-center revenue. Copper costs are a partial offset.

Schneider participates across electrical distribution, power management, and liquid cooling for high-density AI facilities, creating substantial but indirect project exposure.

Schneider supplies electrical distribution, power-management, and liquid-cooling systems for high-density data centers. The financed transaction supports the broader AI infrastructure-capex cycle, while adding another scaled competitor in thermal systems.

Delta supplies sidecar and in-rack liquid-cooling systems, creating direct but narrower AI infrastructure-capex exposure.

High-density AI racks require coolant distribution and thermal-management hardware. Delta is identified as a leading supplier of sidecar coolant distribution units, creating potential demand exposure as MEGA IDC transitions toward AI-ready fit-out.

nVent has a narrower exposure to racks, manifolds, and liquid-cooling distribution infrastructure within AI facilities.

High-density AI racks require specialized thermal-management infrastructure. nVent is identified as supplying customized chillers and related cooling equipment, giving it indirect exposure to the planned facility's cooling procurement without evidence of a project award.

Carrier has meaningful data-center cooling exposure, but the AI-specific capex opportunity is moderated by the shift toward specialized liquid systems.

Carrier participates in the expanding data-center cooling market, but the new information favors AI campuses that increasingly use specialized liquid cooling rather than conventional room-level air-cooling equipment. This creates relative mix and competitive pressure on Carrier’s conventional cooling exposure even if total cooling demand grows.

Higher-density AI-server deployment requires liquid-cooling and thermal-management systems, creating potential second-order demand for Envicool.

Higher-density AI servers require advanced thermal management and liquid-cooling equipment. Envicool is a named Chinese liquid-cooling supplier serving domestic hyperscale operators, so a multi-year computing-capacity deployment provides a positive second-order demand pathway, without proving a direct order.

Founded in 2005 and headquartered in Shenzhen, China, Shenzhen Envicool Technology Co., Ltd. specializes in the domestic production and sale of advanced thermal management solutions. The company's diverse product portfolio addresses a wide range of cooling needs, including AC and DC-powered air conditioners, heat exchangers, and integrated units for outdoor cabinets. They also deliver specialized cooling systems for telecommunications infrastructure, such as free cooling and split AC/DC air conditioners. Envicool serves industrial sectors with robust air conditioners, fan and filter assemblies, air-to-water heat exchangers, precise thermostats/hygrostats, heaters, and various oil and water cooling apparatus. Their expertise extends to comprehensive cooling solutions for transportation and data centers. Additional offerings encompass door/wall and floor-standing units, along with indoor environmental control devices like ceiling and wall-mounted air units, elevator air sterilizers, and mobile disinfection purifiers. The company leverages both liquid and air cooling technologies throughout its extensive product line.

Tongfei supplies liquid-cooling components and systems for high-density compute, giving it moderate second-order exposure to AI data-center expansion.

Sanhe Tongfei Refrigeration Co., Ltd. research, develops, produce, and sells of industrial temperature control equipment in China. The company offers smart equipment manufacturing includes MCO/MCOI chiller, MCW/MCWI water chiller, MCS immersion chiller, MWA air/water heat exchanger, MEO oil/air heat exchanger, MCA temperature and humidity conditioner, MEA heat exchanger for electrical cabinet, HE/HS electrophoretic special heat exchanger; renewable energy storage such as LCI inverter liquid cooling unit, ECA wall mounted industrial air conditioner, ECA cabinet-in mounted industrial air conditioner, LCI PCS-specific cooling equipment, LCI high voltage cascaded/central cooling, ECH industrial dehumidifier; and power transmission and distribution electric, includes HWA air/water heat exchanger, and LWA/LWW pure water cooling unit. It also offers Intelligent Computing Center includes DWW cabinet CDU, DWW in rack CDU, DMF manifold, DTK TANK, DHWA dry cooler, DCW integrated chiller, DWA mini fan wall, DWA Chilled Water Rear Door Heat exchanger, DCA External Cabinet Backpack Aircon, DCA built-in integrated precision aircon, DCA DX rack aircon, DCA DX small precision aircon, DCA DX computer room, precision aircon, DCA DX in row precision aircon, DCA DX small precision aircon, DCA DX computer room precision aircon, DCA DX in row precision aircon, single rack/single-row modular data center, single rack/single-double-row modular date center, and DMDC container. In addition, the company offers semiconductor manufacturing; hydrogen energy; and lasers. Sanhe Tongfei Refrigeration Co., Ltd. was founded in 2001 and is based in Sanhe, China.