{"field":"ai-infra","url":"https://www.frontiertechintelligence.com/data/ai-infra","as_of":"2026-11-16","cite":[{"key":"index","text":"Momentum, the activity index of Frontier Tech Intelligence, stood at 103.2 for AI Infra in the week of 9 Nov 2026 (+3.1% over 4 weeks, with the week of 12 Oct 2026 set to 100). Source: Frontier Tech Intelligence, https://www.frontiertechintelligence.com/data/ai-infra, as of 16 Nov 2026."},{"key":"top100","text":"In the AI Infra 100 (November 2026 (updated 15 Nov) edition) published by Frontier Tech Intelligence, TSMC ranks first, Alphabet second and SK hynix third. Source: https://www.frontiertechintelligence.com/data/ai-infra, as of 16 Nov 2026."}],"activity_index":{"value":103.15,"week_start":"2026-11-09","change":{"weeks":4,"from_week":"2026-10-12","from_value":100,"points":3.15,"percent":3.15},"base_week":"2026-10-12","series":[{"week":"2026-09-21","value":109.41,"weekly":114.54},{"week":"2026-09-28","value":104.37,"weekly":104.84},{"week":"2026-10-05","value":102.09,"weekly":103.19},{"week":"2026-10-12","value":100,"weekly":100},{"week":"2026-10-19","value":101.72,"weekly":111.27},{"week":"2026-10-26","value":103.9,"weekly":115.74},{"week":"2026-11-02","value":102.37,"weekly":101.73},{"week":"2026-11-09","value":103.15,"weekly":111.02}]},"top100":{"name":"AI Infra 100","edition":"November 2026 (updated 15 Nov)","members":[{"rank":1,"name":"TSMC","slug":"tsmc","layer":"Fabrication, memory & packaging","segment":"Foundry","country":"Taiwan","score":100,"positioning":"Holds roughly 72% of the global contract chip-making market and makes nearly every AI accelerator of NVIDIA, AMD, Broadcom, Amazon and Microsoft. It controls more than 90% of CoWoS packaging, the tightest chokepoint in the AI chip chain.","display_name":"TSMC (incl. Global Unichip)","score_delta":0},{"rank":2,"name":"Alphabet","slug":"alphabet","layer":"Compute demand & operators","segment":"Agents & reasoning","country":"United States","score":98.5,"positioning":"Sits on both sides of AI computing, as the largest buyer for its own use after Amazon and Microsoft. Its TPU customers include Anthropic, and AI Overviews reach 2.5 billion monthly users.","display_name":"Alphabet (incl. Google DeepMind, Wiz)","score_delta":6},{"rank":3,"name":"SK hynix","slug":"sk-hynix","layer":"Fabrication, memory & packaging","segment":"HBM & DRAM","country":"South Korea","score":96,"positioning":"Holds about half of the high-bandwidth memory market and is the memory supplier most tied to NVIDIA, so its allocation decisions set how many accelerators can ship in a quarter.","display_name":"SK hynix (incl. Solidigm)","score_delta":6},{"rank":4,"name":"NVIDIA","slug":"nvidia","layer":"Chips & accelerators","segment":"Accelerator design","country":"United States","score":94,"positioning":"Default supplier of AI accelerators, networking and systems, with an installed base no rival approaches.","display_name":"NVIDIA (incl. Enfabrica)","score_delta":0},{"rank":5,"name":"ASML","slug":"asml","layer":"Semiconductor equipment, materials & critical inputs","segment":"Lithography","country":"Netherlands","score":93,"positioning":"The sole supplier of extreme ultraviolet lithography, a chokepoint on every factory making the newest logic or memory chips. Its output sets the ceiling on how fast TSMC, Samsung, SK hynix, Micron and Intel can add advanced capacity.","score_delta":0},{"rank":6,"name":"Amazon","slug":"amazon","layer":"Compute demand & operators","segment":"Hyperscalers","country":"United States","score":91,"positioning":"The largest cloud, and anchor buyer of NVIDIA chips and of its own Trainium and Graviton chips. Supplies compute to Anthropic and OpenAI.","display_name":"Amazon (incl. AWS)","score_delta":2},{"rank":7,"name":"Samsung Electronics","slug":"samsung-electronics","layer":"Fabrication, memory & packaging","segment":"HBM & DRAM","country":"South Korea","score":90.5,"positioning":"The only company that is both a top-three supplier of high-bandwidth memory and DRAM and an advanced contract chipmaker, with a 2nm process at about 80% yield. It is the main alternative to TSMC for customers who cannot get wafers.","display_name":"Samsung Electronics (Samsung Group, incl. Samsung SDS)","score_delta":6},{"rank":8,"name":"Applied Materials","slug":"applied-materials","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"United States","score":88.5,"positioning":"The largest chip equipment maker and the process leader in deposition, etch and materials engineering, supplying every advanced chip factory. It develops the Kinex bonding platform with BE Semiconductor Industries.","score_delta":0},{"rank":9,"name":"Broadcom","slug":"broadcom","layer":"Chips & accelerators","segment":"Custom silicon & IP","country":"United States","score":88,"positioning":"Builds Google's TPU chip and OpenAI's accelerator, and is the dominant supplier of Ethernet switching for AI clusters. The second pole of the accelerator market after NVIDIA.","display_name":"Broadcom (incl. VMware)","score_delta":0},{"rank":10,"name":"Microsoft","slug":"microsoft","layer":"Compute demand & operators","segment":"Hyperscalers","country":"United States","score":87.5,"positioning":"A large buyer of NVIDIA and AMD chips, with its own Maia chip. Second-largest cloud and main channel for OpenAI models, leasing capacity from CoreWeave, Nebius, Nscale, Lambda and IREN, with 120,000 Foundry customers.","display_name":"Microsoft (incl. GitHub)","score_delta":0},{"rank":11,"name":"Synopsys","slug":"synopsys","layer":"Semiconductor equipment, materials & critical inputs","segment":"EDA & design tools","country":"United States","score":87,"positioning":"One of the two chip design software companies that every AI accelerator and custom chip passes through before it reaches a chip factory. It is also the largest vendor of licensed design components and, with Ansys, of chip and system simulation.","score_delta":0},{"rank":12,"name":"Lam Research","slug":"lam-research","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"United States","score":86,"positioning":"The leader in etching and deposition for memory chips such as NAND flash, DRAM and high-bandwidth memory. NVIDIA is a partner on equipment simulation, and Nikon on panel-scale production.","score_delta":0},{"rank":13,"name":"Advantest","slug":"advantest","layer":"Semiconductor equipment, materials & critical inputs","segment":"Test, measurement & validation","country":"Japan","score":85.5,"positioning":"The dominant supplier of the V93000 range of chip test systems for AI accelerators, custom chips, CPUs and memory. Every advanced-packaged AI chip passes through its testers.","score_delta":0},{"rank":14,"name":"Cadence","slug":"cadence","layer":"Semiconductor equipment, materials & critical inputs","segment":"EDA & design tools","country":"United States","score":85.5,"positioning":"One of the two companies that dominate chip design software, strongest in analog and custom design and verification hardware. Its tools sit in front of every custom accelerator program at the largest cloud companies, and NVIDIA is a named customer.","score_delta":0},{"rank":15,"name":"KLA","slug":"kla","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"United States","score":85,"positioning":"Holds the monopoly in inspection and measurement for chipmaking, at roughly 7.5 times its nearest competitor. Soitec is a named customer.","score_delta":0},{"rank":16,"name":"Oracle","slug":"oracle","layer":"Compute demand & operators","segment":"Hyperscalers","country":"United States","score":84.5,"positioning":"The fourth cloud, now the builder of OpenAI's Stargate sites. It rents others' chips and contracts campuses from developers such as Crusoe.","score_delta":0},{"rank":17,"name":"Arista Networks","slug":"arista-networks","layer":"Networking & interconnect","segment":"Switching & NICs","country":"United States","score":84,"positioning":"The specialist in Ethernet switching for AI clusters, built on Broadcom chips, competing with NVIDIA's InfiniBand and NVLink link technologies. It collaborates with OpenAI on large-scale deployment.","score_delta":0},{"rank":18,"name":"Foxconn","slug":"foxconn","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"Taiwan","score":84,"positioning":"The largest user of robots, with automation at millions-of-units scale in its own plants. An anchor of NVIDIA's partner network, it plans to put NVIDIA-powered humanoids on its Houston AI-server production line.","score_delta":4},{"rank":19,"name":"Micron","slug":"micron","layer":"Fabrication, memory & packaging","segment":"HBM & DRAM","country":"United States","score":84,"positioning":"The only US-based maker of both DRAM and high-bandwidth memory. Third in share of high-bandwidth memory but first to volume shipments of HBM4, for Vera Rubin, with new chip factories in Idaho, New York and Taiwan.","score_delta":0},{"rank":20,"name":"Tokyo Electron","slug":"tokyo-electron","layer":"Semiconductor equipment, materials & critical inputs","segment":"Test, measurement & validation","country":"Japan","score":83.5,"positioning":"Japan's largest equipment maker and one of the four global leaders in its core process steps. A supplier to the largest chipmakers, and a partner of NVIDIA on AI-driven factory automation.","score_delta":0},{"rank":21,"name":"Innolight","slug":"innolight","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"China","score":83,"positioning":"The largest optical transceiver maker in the world, with about 3 million 1.6T units shipped and more than 50% global share, selling to NVIDIA, Google and Microsoft. Its own silicon photonics chips yield 95% and cut costs 30%.","score_delta":0},{"rank":22,"name":"AMD","slug":"amd","layer":"Chips & accelerators","segment":"Accelerator design","country":"United States","score":82.5,"positioning":"The only open-market GPU alternative at rack scale, with OpenAI, Meta, Microsoft, Anthropic, Oracle and Humain as customers. Its EPYC server processors dominate the processor side of AI clusters.","score_delta":0},{"rank":23,"name":"CoreWeave","slug":"coreweave","layer":"Compute demand & operators","segment":"Neoclouds & GPU clouds","country":"United States","score":82,"positioning":"The largest independent operator of NVIDIA computing capacity. Microsoft, OpenAI, Meta and Anthropic are named customers on contracts where they pay whether or not they use the capacity.","score_delta":0},{"rank":24,"name":"Equinix","slug":"equinix","layer":"Data center build & operate","segment":"Colocation & wholesale","country":"United States","score":82,"positioning":"The largest provider of rented data center space, with more than 270 sites and more than 500,000 interconnections. It hosts eight of the top ten AI model providers and builds 60 MW-class sites with partners, not gigawatt campuses.","score_delta":3},{"rank":25,"name":"OpenAI","slug":"openai","layer":"Compute demand & operators","segment":"Agents & reasoning","country":"United States","score":82,"positioning":"A buyer, not an owner, of computing capacity, and the largest contracted one in the world. It commits future spend to Oracle, Microsoft, Amazon, CoreWeave, NVIDIA, AMD and Broadcom, and ChatGPT has 1.1 billion weekly users.","display_name":"OpenAI (incl. Orchestra Bio)","score_delta":0},{"rank":26,"name":"Vertiv","slug":"vertiv","layer":"Power, cooling & thermal","segment":"Cooling & thermal","country":"United States","score":82,"positioning":"The specialist that sells only into data centers, from rack power and backup power to liquid cooling and prefabricated modules, with more than 4,000 field service engineers.","score_delta":3},{"rank":27,"name":"Meta","slug":"meta","layer":"Compute demand & operators","segment":"Hyperscalers","country":"United States","score":81.5,"positioning":"A buyer, not a seller, of AI computing. The only one at the scale of the largest cloud companies with no cloud business, with AMD and NVIDIA as suppliers.","display_name":"Meta (incl. Conductor)","score_delta":0},{"rank":28,"name":"Arm","slug":"arm","layer":"Chips & accelerators","segment":"Custom silicon & IP","country":"United Kingdom","score":81,"positioning":"Its designs sit under nearly every custom data center processor, including those of NVIDIA, Google, AWS and Microsoft, with 1 billion Neoverse cloud cores deployed. No other licensor holds that position.","display_name":"Arm (SoftBank Group, incl. Graphcore)","score_delta":0},{"rank":29,"name":"Shin-Etsu Chemical","slug":"shin-etsu-chemical","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"Japan","score":81,"positioning":"The world's largest supplier of wafers and of photoresist chemicals. It says requests for many items exceed its capacity.","score_delta":0},{"rank":30,"name":"Corning","slug":"corning","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":80,"positioning":"Has multiyear supply agreements with Amazon, Meta and NVIDIA, and uses long-term contracts to share the risk of capacity expansion. Microsoft is a partner on hollow-core fiber.","score_delta":0},{"rank":31,"name":"Dell Technologies","slug":"dell-technologies","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"United States","score":80,"positioning":"The largest branded server vendor, and a main channel through which enterprise and sovereign buyers get NVIDIA and AMD systems. 6,500 of its PowerEdge AI systems are deployed at organizations.","score_delta":0},{"rank":32,"name":"Schneider Electric","slug":"schneider-electric","layer":"Power, cooling & thermal","segment":"","country":"France","score":80,"positioning":"The largest supplier of power and cooling for data centers worldwide, and the main European supplier of their electrical equipment.","score_delta":2},{"rank":33,"name":"SpaceX","slug":"spacex","layer":"Compute demand & operators","segment":"Launch services","country":"United States","score":80,"positioning":"Falcon rockets have flown 700 launches, and its Starlink network has passed 11,000 satellites and 8 million customers. Its AI arm is the only lab to own gigawatt-class computing outright with on-site power, rented mostly to Anthropic and Google.","display_name":"SpaceX (incl. xAI)","score_delta":0},{"rank":34,"name":"ASE Technology Holding","slug":"ase-technology-holding","layer":"Fabrication, memory & packaging","segment":"Advanced packaging","country":"Taiwan","score":79,"positioning":"Among the largest independent chip assembly and test suppliers worldwide, and an alternative to in-house packaging at chip foundries. AMD has standardized on it for advanced assembly, and it works with NVIDIA through SPIL.","score_delta":0},{"rank":35,"name":"Huawei","slug":"huawei","layer":"Chips & accelerators","segment":"China accelerators","country":"China","score":79,"positioning":"China's standard for AI chips under sanctions, with its own memory and a CUDA alternative. More than 1,000 Ascend supernodes are deployed, ByteDance has ordered its chips, and it depends wholly on chipmaker SMIC's yields for volume.","display_name":"Huawei (incl. Huawei Cloud)","score_delta":0},{"rank":36,"name":"Ibiden","slug":"ibiden","layer":"Fabrication, memory & packaging","segment":"Advanced packaging","country":"Japan","score":79,"positioning":"Demand for its substrates outpaces supply, and its own constraint is ABF film from a single supplier.","score_delta":0},{"rank":37,"name":"Quanta Computer","slug":"quanta-computer","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"Taiwan","score":79,"positioning":"A server maker, not a robot builder, in robotics. Its partner network includes NVIDIA, AMD, Arm, Oracle and Techman Robot.","display_name":"Quanta Computer (incl. Techman)","score_delta":1},{"rank":38,"name":"Eoptolink","slug":"eoptolink","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"China","score":78.5,"positioning":"Among the first at large-scale production of 1.6T modules, from a Thailand plant built for North American demand. Amazon and Meta are named customers, and it is one of China's two big module makers.","score_delta":0},{"rank":39,"name":"Amphenol","slug":"amphenol","layer":"Networking & interconnect","segment":"Cables, connectors & PCB","country":"United States","score":78,"positioning":"The largest interconnect manufacturer, which acquired CommScope's Connectivity and Cable Solutions business.","score_delta":0},{"rank":40,"name":"Coherent","slug":"coherent","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":78,"positioning":"The largest supplier of optical transceivers and lasers to AI data centers, with NVIDIA as a committed buyer of its optics. It names indium phosphide capacity as its bottleneck.","score_delta":0},{"rank":41,"name":"Ajinomoto Fine-Techno","slug":"ajinomoto-fine-techno","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"Japan","score":77.5,"positioning":"The sole supplier, with about 95% share, of the insulating layer in every high-end GPU, CPU and custom chip package substrate. A specialty-chemicals unit of a food company.","score_delta":-3.5},{"rank":42,"name":"Kioxia","slug":"kioxia","layer":"Fabrication, memory & packaging","segment":"","country":"Japan","score":77.5,"positioning":"Works with NVIDIA on storage designed for AI, including the Context Memory Storage and Storage-Next initiatives, and is securing multi-year supply agreements.","score_delta":0},{"rank":43,"name":"Alibaba","slug":"alibaba","layer":"Compute demand & operators","segment":"Hyperscalers","country":"China","score":77,"positioning":"Runs China's largest cloud and is its largest domestic buyer of computing. The Qwen App has 300 million monthly users across platforms, and Qwen and Wan models have been downloaded more than 600 million times.","display_name":"Alibaba (incl. Alibaba Token Hub, T-Head)","score_delta":0},{"rank":44,"name":"Digital Realty","slug":"digital-realty","layer":"Data center build & operate","segment":"Colocation & wholesale","country":"United States","score":77,"positioning":"The second-largest data center real estate investment trust, and the largest developer for big cloud customers among listed operators. It finances gigawatt campuses through joint ventures with private capital.","score_delta":0},{"rank":45,"name":"DISCO","slug":"disco","layer":"Semiconductor equipment, materials & critical inputs","segment":"","country":"Japan","score":77,"positioning":"Holds 60 to 80% of the market, a near-monopoly, in wafer dicing and grinding equipment and blades. Stacked memory needs wafers thinned to 30 to 50 microns, and it has shipped more than 4,000 laser saws.","score_delta":-3.5},{"rank":46,"name":"Marvell","slug":"marvell","layer":"Chips & accelerators","segment":"Custom silicon & IP","country":"United States","score":77,"positioning":"The second custom-chip supplier to the largest cloud companies, far smaller than Broadcom. Google is a customer, and the relationship now extends beyond its TPU chips.","score_delta":0},{"rank":47,"name":"Unimicron","slug":"unimicron","layer":"Fabrication, memory & packaging","segment":"Advanced packaging","country":"Taiwan","score":77,"positioning":"The largest company of its kind in Taiwan, and the main substrate supplier for the custom AI chips of four US cloud companies, as well as for NVIDIA and Google. Apple is also a customer.","score_delta":0},{"rank":48,"name":"Ciena","slug":"ciena","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":76.5,"positioning":"A specialist, not a broad networking vendor, with 1,700 customers worldwide. It is moving inside the data center with copper links.","display_name":"Ciena (incl. Nubis)","score_delta":0},{"rank":49,"name":"LITEON Technology","slug":"liteon-technology","layer":"Power, cooling & thermal","segment":"Optical & photonic interconnect","country":"Taiwan","score":76,"positioning":"The first electronics company listed in Taiwan, repositioned around AI data center power and cooling. Its 800-volt DC power systems are, it says, co-developed with NVIDIA.","score_delta":0},{"rank":50,"name":"Trane Technologies","slug":"trane-technologies","layer":"Power, cooling & thermal","segment":"Cooling & thermal","country":"Ireland","score":76,"positioning":"The largest maker of commercial air-conditioning and heating equipment, with master purchase agreements and long-term capacity commitments from the largest cloud firms. It bought LiquidStack and Stellar Energy for liquid cooling and modular plants.","score_delta":0},{"rank":51,"name":"Delta Electronics","slug":"delta-electronics","layer":"Power, cooling & thermal","segment":"Rack & server power delivery","country":"Taiwan","score":75,"positioning":"Develops its power racks and a 2.4 MW coolant distribution unit together with NVIDIA, and offers a prefabricated AI data center that it says cuts construction time by 60%.","score_delta":0},{"rank":52,"name":"Lumentum","slug":"lumentum","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":75,"positioning":"Its lasers come from its own indium phosphide chip factories. Google is a customer for its optical circuit switches under a multiyear agreement, and NVIDIA a strategic partner.","score_delta":0},{"rank":53,"name":"GE Vernova","slug":"ge-vernova","layer":"Power, cooling & thermal","segment":"Generation for data centers","country":"United States","score":74.5,"positioning":"The largest gas-turbine maker and one of three that control about two-thirds of turbine capacity under construction. Its turbine slots are the binding constraint on on-site power for data centers.","score_delta":0},{"rank":54,"name":"Inspur Information","slug":"inspur-information","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"China","score":74.5,"positioning":"The largest AI server maker in China, the domestic counterpart to Foxconn and Dell in rack builds for the largest cloud companies. Increasingly builds on Huawei Ascend and other domestic accelerators.","score_delta":0},{"rank":55,"name":"Supermicro","slug":"supermicro","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"United States","score":74.5,"positioning":"Sells whole data center building blocks, not only servers, and added several hundred enterprise and other customers in a single fiscal year.","score_delta":0},{"rank":56,"name":"Infineon","slug":"infineon","layer":"Power, cooling & thermal","segment":"Power semiconductors (GaN/SiC)","country":"Germany","score":74,"positioning":"Covers power conversion from the grid to the chip. It works with NVIDIA on 800 V direct-current power for data centers, with Delta Electronics as development partner.","score_delta":0},{"rank":57,"name":"Sandisk","slug":"sandisk","layer":"Fabrication, memory & packaging","segment":"HBM & DRAM","country":"United States","score":73.5,"positioning":"Independent of Western Digital and makes its flash with Kioxia. Works with SK hynix on the High Bandwidth Flash specification.","score_delta":4.5},{"rank":58,"name":"Aligned Data Centers","slug":"aligned-data-centers","layer":"Data center build & operate","segment":"Hyperscale developers","country":"United States","score":73,"positioning":"Its 51 campuses hold 6.4 GW of operational and planned capacity. Its owning consortium has Microsoft, NVIDIA and xAI among its partners, and its sale was the largest data-center acquisition on record.","score_delta":0},{"rank":59,"name":"Caterpillar","slug":"caterpillar-inc","layer":"Power, cooling & thermal","segment":"Field & outdoor","country":"United States","score":73,"positioning":"One of the two main suppliers of autonomous mine haulage. The world's largest maker of construction and mining equipment, with more than 550 mining trucks running its autonomous system on three continents.","score_delta":0},{"rank":60,"name":"Credo","slug":"credo","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":73,"positioning":"The copper-side counterpart to the optical-module makers, for links inside the rack and over short distances. Now adds retimers, optical signal chips and silicon photonics through DustPhotonics.","score_delta":0},{"rank":61,"name":"Eaton","slug":"eaton","layer":"Power, cooling & thermal","segment":"Rack & server power delivery","country":"Ireland","score":73,"positioning":"The broadest electrical range in the data center hall, extended into liquid cooling with the purchase of Boyd Thermal. With Vertiv and Schneider, one of three names on nearly every very large build.","score_delta":0},{"rank":62,"name":"QTS","slug":"qts","layer":"Data center build & operate","segment":"Colocation & wholesale","country":"United States","score":73,"positioning":"The largest private data center platform, building the Pennsylvania digital-energy hub with on-site generation.","score_delta":1},{"rank":63,"name":"Amkor Technology","slug":"amkor-technology","layer":"Fabrication, memory & packaging","segment":"Advanced packaging","country":"United States","score":72.5,"positioning":"Second-largest of the outsourced chip packagers and the US-based option. It has a 10-year advanced packaging agreement with TSMC and is building the first US advanced-packaging facility, in Arizona near TSMC's chip factories.","score_delta":0},{"rank":64,"name":"Besi","slug":"besi","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"Netherlands","score":72.5,"positioning":"The most accurate maker of chip-to-wafer bonding tools, with 21 customers for them. They align to 200 nm and place up to 2,000 chips an hour, and Applied Materials is a named partner.","score_delta":0},{"rank":65,"name":"Wistron","slug":"wistron","layer":"Data center build & operate","segment":"","country":"Taiwan","score":72.5,"positioning":"Builds servers around NVIDIA's RTX PRO 6000 Blackwell graphics chips for enterprise customers, and uses NVIDIA's Omniverse technology in its own server production.","display_name":"Wistron (incl. Wiwynn)","score_delta":0},{"rank":66,"name":"Anthropic","slug":"anthropic","layer":"Compute demand & operators","segment":"Agents & reasoning","country":"United States","score":72,"positioning":"Buys computing capacity rather than running its own data center sites, with up to 5 GW of Amazon Trainium chip capacity and about 5 GW of Google TPU chip capacity agreed. More than 100,000 customers run Claude on Amazon Bedrock.","score_delta":0},{"rank":67,"name":"ASM International","slug":"asm-international","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"Netherlands","score":72,"positioning":"Strongest in the newest logic chips and in advanced memory, and its tools are in high-volume production of 2nm chips.","score_delta":0},{"rank":68,"name":"Cisco","slug":"cisco","layer":"Networking & interconnect","segment":"Switching & NICs","country":"United States","score":72,"positioning":"The established networking vendor, now the second supplier of network switching for AI to the largest cloud companies, after Arista.","score_delta":0},{"rank":69,"name":"Entegris","slug":"entegris","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"United States","score":72,"positioning":"The materials-purity layer under every chip factory expansion, covering the most advanced chips, high-bandwidth memory and advanced packaging, with specialty chemicals and wafer carriers alongside.","score_delta":0},{"rank":70,"name":"Nebius","slug":"nebius","layer":"Compute demand & operators","segment":"Neoclouds & GPU clouds","country":"Netherlands","score":72,"positioning":"A cloud provider, not a robot maker, in humanoids. Microsoft is its anchor customer, with Meta also under contract, and NVIDIA is a partner for its robot developer platform.","score_delta":0},{"rank":71,"name":"MediaTek","slug":"mediatek","layer":"Chips & accelerators","segment":"Accelerator design","country":"Taiwan","score":71.5,"positioning":"Taiwan's largest company of its kind, and Google's second source for TPU AI chips alongside Broadcom, with expanded TPU orders.","score_delta":0},{"rank":72,"name":"Astera Labs","slug":"astera-labs","layer":"Networking & interconnect","segment":"Optical & photonic interconnect","country":"United States","score":71,"positioning":"The specialist in connections inside the AI server and rack. Its Scorpio switches join accelerators to CPUs and memory, and its retimers carry PCIe 6 signals.","score_delta":0},{"rank":73,"name":"Cambricon","slug":"cambricon","layer":"Chips & accelerators","segment":"China accelerators","country":"China","score":71,"positioning":"China's main alternative to Huawei for the largest cloud companies and AI labs. Its chips were used to train Ant Group's Ling model, and DeepSeek models ran on them from release day.","score_delta":0},{"rank":74,"name":"Celestica","slug":"celestica","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"Canada","score":71,"positioning":"Pairs high-volume manufacturing with rack-scale systems integration. It builds Google TPU systems, works with AMD on the Helios rack-scale platform and does the system integration for OpenAI's Jalapeño accelerator.","score_delta":0},{"rank":75,"name":"HPE","slug":"hpe","layer":"Networking & interconnect","segment":"Servers, racks & system integration","country":"United States","score":71,"positioning":"Has 52,000 GreenLake customers. Vultr chose it with NVIDIA for large AI data center deployments, and Oracle plans Juniper networking across its AI data centers.","score_delta":0},{"rank":76,"name":"Johnson Controls","slug":"johnson-controls","layer":"Power, cooling & thermal","segment":"","country":"Ireland","score":71,"positioning":"A building-systems maker moving into data center cooling. Its coolant distribution units are NVIDIA-certified, and it says an absorption chiller design cuts cooling electricity use by 44 percent at a 1 GW AI data center.","score_delta":0},{"rank":77,"name":"Renesas","slug":"renesas","layer":"Power, cooling & thermal","segment":"Accelerator design","country":"Japan","score":71,"positioning":"Starts from microcontrollers and extends into AI at two ends. More than 20 million gallium nitride devices shipped, and its devices support NVIDIA's 800V direct-current power design for AI data centers.","score_delta":0},{"rank":78,"name":"Vantage Data Centers","slug":"vantage-data-centers","layer":"Data center build & operate","segment":"Colocation & wholesale","country":"United States","score":71,"positioning":"Builds gigawatt campuses, for OpenAI and Oracle in Wisconsin and for Nebius in the UK. Up to 2.5 GW across Project Frontier and Project Lighthouse.","score_delta":0},{"rank":79,"name":"Alchip Technologies","slug":"alchip-technologies","layer":"Chips & accelerators","segment":"Custom silicon & IP","country":"Taiwan","score":70.5,"positioning":"Has sent more designs to manufacture on the CoWoS advanced-packaging process than any other design house. Its largest customer is a North American cloud company, for which it is ramping a 3nm accelerator.","score_delta":0},{"rank":80,"name":"Flex","slug":"flex","layer":"Power, cooling & thermal","segment":"Rack & server power delivery","country":"Singapore","score":70.5,"positioning":"It says it is the only provider of both complete cloud IT integration and a full power and cooling range at scale. It works with Cerebras on US manufacturing of its AI supercomputers.","score_delta":0},{"rank":81,"name":"Hitachi Energy","slug":"hitachi-energy","layer":"Power, cooling & thermal","segment":"Grid & power equipment","country":"Switzerland","score":70.5,"positioning":"A chokepoint supplier of transformers, with its main transformer lines booked out 3 to 4 years. It is building new transformer plants in Virginia and Mississippi, and works with AWS and NVIDIA.","display_name":"Hitachi Energy (Hitachi)","score_delta":0},{"rank":82,"name":"Lasertec","slug":"lasertec","layer":"Semiconductor equipment, materials & critical inputs","segment":"Deposition, etch & metrology","country":"Japan","score":70.5,"positioning":"The sole supplier of mask blank and pattern inspection for extreme ultraviolet lithography, so every such mask passes its tools. The second one-of-one in lithography after ASML, with Intel and TSMC as customers of its ACTIS A200HiT series.","score_delta":0},{"rank":83,"name":"Siemens Energy","slug":"siemens-energy","layer":"Power, cooling & thermal","segment":"Grid & power equipment","country":"Germany","score":70.5,"positioning":"The second of the three large gas-turbine makers and, through Grid Technologies, one of the two largest suppliers of transformers and switchgear. The European counterpart to GE Vernova, partnered with Eaton on modular data centers.","score_delta":0},{"rank":84,"name":"Linde","slug":"linde","layer":"Semiconductor equipment, materials & critical inputs","segment":"Critical minerals & gases","country":"United Kingdom","score":70,"positioning":"A gas supplier to chip factories and rocket launches. The largest industrial-gas company, with electronics its fastest-growing market and a long-term agreement to supply Samsung in South Korea.","score_delta":0},{"rank":85,"name":"Tencent","slug":"tencent","layer":"Models & applications","segment":"Hyperscalers","country":"China","score":70,"positioning":"Second among Chinese cloud companies, with its own Hunyuan AI models as a captive customer for its compute. It buys domestic AI chips, among them Enflame's, and says it can rent surplus compute to third parties.","score_delta":0},{"rank":86,"name":"Bloom Energy","slug":"bloom-energy","layer":"Power, cooling & thermal","segment":"Generation for data centers","country":"United States","score":69.5,"positioning":"Aimed at data centers whose grid connections are years away. Up to 2.45 GW of its fuel cells are planned for Oracle's Project Jupiter in New Mexico, and American Electric Power and Brookfield are also partners.","score_delta":0},{"rank":87,"name":"Cerebras","slug":"cerebras","layer":"Chips & accelerators","segment":"Accelerator design","country":"United States","score":69.5,"positioning":"Sells fast capacity for running AI models, not chips alone, to OpenAI and through AWS. OpenAI has agreed to deploy 750 megawatts of Cerebras computing.","score_delta":0},{"rank":88,"name":"Nittobo","slug":"nittobo","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"Japan","score":69,"positioning":"The dominant maker of T-glass, whose shortage substrate maker Unimicron names as limiting supply of high-end chip substrates.","score_delta":0},{"rank":89,"name":"ABB","slug":"abb","layer":"Power, cooling & thermal","segment":"Industrial arms & cobots","country":"Switzerland","score":68.5,"positioning":"Covers the electrical side of AI data centers, not the computing side. It is developing power and control systems for floating data centers with Hanwha Ocean, and supplies power infrastructure to VoltaGrid.","score_delta":0},{"rank":90,"name":"Lenovo Group","slug":"lenovo-group","layer":"Data center build & operate","segment":"Servers, racks & system integration","country":"China","score":68.5,"positioning":"A supplier of AI servers and services, with 5,800 customer AI deployments. It builds hybrid AI platforms with NVIDIA and serves millions of customers in 180 markets.","score_delta":0},{"rank":91,"name":"SUMCO","slug":"sumco","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"Japan","score":68.5,"positioning":"Second-largest silicon-wafer maker, with demand for its 300mm wafers rising on AI.","score_delta":0},{"rank":92,"name":"WUS Printed Circuit Co.","slug":"wus-printed-circuit-co","layer":"Networking & interconnect","segment":"Cables, connectors & PCB","country":"China","score":68.5,"positioning":"","score_delta":0},{"rank":93,"name":"nVent Electric","slug":"nvent-electric","layer":"Power, cooling & thermal","segment":"Cooling & thermal","country":"United Kingdom","score":68,"positioning":"An electrical maker with a growing data center line. Works with NVIDIA on liquid cooling for GB200 NVL72 systems and also sells cable management and power distribution.","score_delta":0},{"rank":94,"name":"Resonac","slug":"resonac","layer":"Semiconductor equipment, materials & critical inputs","segment":"Wafers, chemicals & gases","country":"Japan","score":68,"positioning":"Has narrowed its focus to semiconductor materials by spinning off its petrochemicals business.","score_delta":0},{"rank":95,"name":"Teradyne","slug":"teradyne","layer":"Semiconductor equipment, materials & critical inputs","segment":"Industrial arms & cobots","country":"United States","score":68,"positioning":"Parent of two robot makers, Universal Robots and MiR. Universal Robots has sold 110,000 collaborative robots worldwide, and 43 MiR robots work at Stellantis' Caen plant.","score_delta":0},{"rank":96,"name":"China Telecom","slug":"china-telecom","layer":"Compute demand & operators","segment":"Neoclouds & GPU clouds","country":"China","score":63,"positioning":"An operator in AI platforms and satellite services, with 439 million mobile subscribers. Xiaomi and Samsung support its satellite link on the Xiaomi 15 Ultra and Galaxy S25 Ultra.","score_delta":0},{"rank":97,"name":"Fireworks AI","slug":"fireworks-ai","layer":"Compute software","segment":"Inference & serving clouds","country":"United States","score":55,"positioning":"Runs on rented graphics-chip capacity and owns no data center sites.","score_delta":-12},{"rank":98,"name":"Baseten","slug":"baseten","layer":"Compute software","segment":"Inference & serving clouds","country":"United States","score":49,"positioning":"Runs customers' models on rented graphics chips across several clouds. A software layer with no owned chips or sites.","score_delta":-14},{"rank":99,"name":"OpenRouter","slug":"openrouter","layer":"Compute software","segment":"Inference & serving clouds","country":"United States","score":49,"positioning":"Fronts most open and closed model APIs and publishes the token-share data the industry quotes. Processes more than 10 trillion tokens per day, with routing controls for privacy and European data processing."},{"rank":100,"name":"Tata Consultancy Services","slug":"tata-consultancy-services","layer":"Agents & tooling","segment":"","country":"India","score":43,"positioning":"A services provider rather than a chip maker. Its WisdomNext platform governs and orchestrates vendor, internal and open-source generative AI models for enterprises, with Air France-KLM and Best Buy as customers.","score_delta":0}]},"momentum":null,"csv":"https://www.frontiertechintelligence.com/data/ai-infra.csv","license":"Free to cite with attribution and link: \"Frontier Tech Intelligence, https://www.frontiertechintelligence.com/data/ai-infra\"."}