Power & grids
Operate power generation facilities such as nuclear power plants and sell electricity to enterprises and power grids.
Supply chain bottlenecks The output of active nuclear power depends on the refueling shutdown schedule and maintenance progress. Restarting the unit also requires regulatory approval, component replacement and grid connection testing.
Manufacturing of gas turbines and power transmission and distribution equipment.
Supply chain bottlenecks Gas turbine output is affected by the delivery schedule of high-temperature resistant alloy blades and large castings and forgings. Grid equipment also requires electrical steel for transformer cores.
Sells electricity through its Florida utility and develops wind, solar and storage projects.
Supply chain bottlenecks Wind and solar projects have to wait for the delivery of wind turbines, components and energy storage cells. Even if the equipment arrives on site, it still has to wait for grid connection approval and transmission capacity to be in place.
Operates power plants, including nuclear and natural gas facilities, and sells electricity to homes and businesses.
Supply chain bottlenecks Gas-fired power plants rely on pipeline gas and turbine spare parts. The stable output of nuclear power units also depends on nuclear fuel and maintenance arrangements.
Operates power plants, including the Susquehanna nuclear plant, and sells electricity.
Supply chain bottlenecks The Susquehanna Nuclear Power Plant is dependent on nuclear fuel supply and unit availability. New machine room loads must wait for the grid connection plan to be approved.
Sells solid oxide fuel cell systems for on-site power at data centers and other facilities.
Supply chain bottlenecks The production expansion of solid oxide fuel cells is restricted by ceramic cells, stacking yield and gas access conditions.
Makes silicon carbide materials and power devices for electric vehicles, industrial equipment and power conversion.
Supply chain bottlenecks Silicon carbide devices must first have low-defect crystals and qualified substrates. The yield of large-size wafers determines how many devices can be produced.
Agreement with Meta to advance nuclear power projects to serve data centers.
Supply chain bottlenecks The reactor must obtain a nuclear license and available fuel, and then implement the delivery of nuclear-grade equipment and on-site construction before it can start supplying power.
Data centers & cooling
Provide power, backup power and cooling equipment for the data center.
Supply chain bottlenecks Liquid cooling delivery relies on pumps, heat exchangers and coolant distribution units. The entire system must be installed and commissioned alongside the rack.
Manufacture of power distribution and power protection equipment used in data centers.
Supply chain bottlenecks The data center power distribution equipment must wait for circuit breakers, busbars, transformers and other components to be available. Complete sets of switchgear must be certified before they can be installed.
Leasing data center cabinets to enterprises and providing network interconnection services.
Supply chain bottlenecks New cabinets require utility power access and backup power supply capacity. High-density cabinets also need to be equipped with liquid cooling pipelines.
Operates data centers, rents out GPU computing capacity and mines Bitcoin.
Supply chain bottlenecks Converting a site into a GPU cloud requires simultaneous implementation of computer room cooling, power supply, and GPU delivery. Mine-level power capacity cannot be directly regarded as available AI cabinets.
Builds and operates high-density data centers and leases capacity for AI and high-performance computing.
Supply chain bottlenecks The launch of high-density computer rooms depends on the delivery of substations, backup power supplies and liquid cooling systems, followed by completion of power-on and load tests.
Sells electrical protection equipment, server racks and liquid cooling systems for data centers.
Supply chain bottlenecks The cabinet liquid cooling components must be equipped with manifolds, cold plates and quick connectors. Leakage testing and interface compatibility determine whether it can be installed into the complete machine.
Designs gallium nitride and silicon carbide power chips for power conversion.
Supply chain bottlenecks Gallium nitride and silicon carbide power chips rely on external wafer manufacturing, packaging and testing. High-voltage devices also need to pass reliability verification.
Helios Campus delivers AI data center capacity to CoreWeave.
Supply chain bottlenecks The expansion of Helios requires the implementation of long-delivery equipment such as transformers, and the adjustment of power supply and cooling facilities to accommodate high-density servers.
Rent of high performance computer room to Fluidstack and Core42.
Supply chain bottlenecks Turning the computer room into a high-performance computing facility requires transformation of power supply, distribution, cooling and network. At each stage, power supply and equipment integration testing must be completed before delivery.
Equipment & foundries
Manufactures wafers for chip design companies and provides advanced packaging.
Supply chain bottlenecks Advanced processes rely on EUV lithography machines, photoresists and high-purity materials. High-end AI chips still need advanced packaging production capacity and qualified substrates.
Makes photolithography machines used to write circuits on wafers.
Supply chain bottlenecks EUV lithography machines rely on Zeiss precision optics and high-power light sources. Assembly and calibration of key modules determine the pace of delivery.
Provide chip manufacturing equipment such as deposition and etching for wafer factories.
Supply chain bottlenecks Deposition and etching equipment requires high-purity process gas lines, vacuum chambers, and radio frequency components. Long lead times for modules will slow down the assembly of the entire machine.
Manufactures chips for designers, focusing on mature and specialty processes.
Supply chain bottlenecks The wafer output of mature processes depends on silicon wafers, electronic gases, photoresist and equipment maintenance. Process switching also requires re-verification of yield.
Designs and sells PC and server processors and operates a chip foundry business.
Supply chain bottlenecks The mass production of advanced processes relies on the introduction of high numerical aperture and other lithography equipment and the process yield increase. Multi-chip packaging also requires an available substrate.
Supplies etching, deposition and cleaning equipment to chipmakers.
Supply chain bottlenecks The reaction chambers, plasma-resistant parts and radio frequency power supplies of etching and deposition equipment must meet purity and precision requirements.
Sells ion implantation equipment used to change the electrical properties of wafer materials.
Supply chain bottlenecks The production capacity of the ion implanter depends on the supply of high-voltage power supply, ion source and vacuum system and the verification of the whole machine.
Supplies wafer and packaging inspection equipment that helps chipmakers find manufacturing defects.
Supply chain bottlenecks Detection and measurement equipment relies on precision optics, motion platforms and highly sensitive detectors. The accuracy calibration of the entire machine limits the quantity that can be delivered.
Manufactures chips for designers, focusing on specialty processes for radio frequency, industrial and automotive applications.
Supply chain bottlenecks RF and silicon photonics processes rely on specialized substrates and certified process equipment. Changing wafers or process materials requires verification of yield.
Manufactures analog, radio frequency, power management and sensor chips for other companies.
Supply chain bottlenecks Special substrates, epitaxial layers and process equipment used in analog chips and silicon photonics processes determine the output of qualified wafers.
Provide advanced packaging and testing of AI chips for customers such as NVIDIA.
Supply chain bottlenecks Advanced packaging relies on qualified substrates, bumps and bonding equipment. After the production expansion, multi-chip assembly, heat dissipation and testing yields must be improved.
Accelerators & servers
Design GPUs for training and running AI models, and provide network equipment and software.
Supply chain bottlenecks GPU delivery requires advanced process wafers, advanced packaging and qualified HBM. Any link that fails to keep up will slow down the delivery of the entire machine.
Read company research
Design server CPUs and AI accelerators.
Supply chain bottlenecks Instinct accelerator relies on foundry wafers, advanced packaging and HBM. Packages composed of multiple core dies also need to verify the overall package yield.
Read company research
Sells PCs, servers and enterprise storage, including AI systems equipped with GPUs.
Supply chain bottlenecks The AI server must be equipped with GPU, CPU, memory, network card and power supply. High-density racks also need to wait for the liquid cooling components to be installed before the complete machine is tested.
Sells servers, storage and networking equipment and supplies high-performance computing and AI systems.
Supply chain bottlenecks AI clusters are delivered with accelerators, high-speed interconnects, and direct liquid cooling components. Multiple cabinets also require full interconnection and thermal load verification.
Builds and sells servers and rack-scale systems, including liquid-cooled GPU racks.
Supply chain bottlenecks Modular servers require simultaneous arrival of GPU motherboards, power supply and distribution, and liquid cooling units. The rack must pass a full load thermal test after assembly.
Designs and manufactures electronics for other brands, including data center racks and power components.
Supply chain bottlenecks Electronics OEM requires complete sets of circuit boards, connectors, power components and assembly lines. Being out of stock on any given part can block final assembly.
Manufactures and assembles equipment for other companies, including data center servers and networking gear.
Supply chain bottlenecks The manufacturing of high-speed switching equipment and servers relies on key components such as switching chips, high-speed PCBs, and optical modules. The entire machine also requires compatibility testing.
Licenses processor architectures and designs to chipmakers and earns license fees and royalties.
Supply chain bottlenecks CPU IP must be compatible with foundry process libraries and EDA tools. The core design must be physically implemented and verified on the target process.
Sells chips for mobile devices, cars and connected products and licenses wireless technology patents.
Supply chain bottlenecks Edge AI chips rely on advanced processes and packaging. Mobile phones and PC platforms must also be equipped with compatible memory, power management chips and drivers.
Makes analog and embedded chips used in industrial systems, vehicles and electronics.
Supply chain bottlenecks Analog chip output relies on equipment installation and yield ramping at 300mm fabs. After the wafer is completed, packaging and testing capacity must be arranged.
Makes power management, power and sensor chips for automotive, industrial and power systems.
Supply chain bottlenecks Power devices require qualified silicon or silicon carbide substrates, epitaxial production capacity and power packaging. Thermal performance and reliability verification will affect the volume.
Makes analog and mixed-signal chips for converting, processing and measuring physical signals.
Supply chain bottlenecks Precision analog and mixed-signal devices rely on specialized wafer processes, precision packaging, and calibrated test equipment.
Designs high-speed interconnect chips and modules that link processors, memory and devices in AI servers.
Supply chain bottlenecks PCIe and CXL interconnect chips rely on external wafer fabrication. Compatibility with CPU, GPU and rack high-speed channels also needs to be verified before mass production.
Sells hybrid cloud, enterprise AI software and servers and provides technology consulting.
Supply chain bottlenecks Enterprise AI services require available computing power, underlying models, and controlled data environments. The self-developed system hardware also needs to be equipped with processors and storage.
Develops trapped-ion quantum computers and offers cloud access and quantum systems.
Supply chain bottlenecks Ion trap quantum systems rely on frequency-stabilized lasers, ultrahigh vacuum components, and precision control electronics. The stability of the entire machine is restricted by these components.
Develop wafer-level AI accelerator and provide inference computing power.
Supply chain bottlenecks Wafer-level processors require a stable supply of foundry wafers and high-density power supply and liquid cooling components before they can be put into operation.
Networking
Design customized acceleration chips for cloud vendors and supply network chips.
Supply chain bottlenecks Custom accelerators rely on advanced processes, packaging, and high-speed interface IP. Switch chips also need to verify the performance of high-speed SerDes on the target process.
Provide high-speed Ethernet switches for data centers.
Supply chain bottlenecks Delivery of high-speed switches relies on commercial switching ASICs, high-speed circuit boards, and compatible optical transceiver modules.
Designing chips for data center optical interconnects and custom computing.
Supply chain bottlenecks Optical interconnect DSPs and custom chips rely on external foundries, advanced packaging, and high-speed interface verification. High-end optical modules also need matching optical devices.
Sells switches, routers and security products for enterprise and data center networks.
Supply chain bottlenecks Data center networking equipment requires a complete set of switching chips, optical transceivers and power supplies, and must pass interoperability testing before being put into the computer room.
Designs high-speed connectivity chips and active cables for data centers.
Supply chain bottlenecks Active cables require high-speed signal processing chips, custom copper cables, and connectors. The entire link must meet high-speed transmission and power consumption specifications.
Makes optical fiber, fiber cables and specialty glass for communications, displays and other markets.
Supply chain bottlenecks Fiber optic delivery relies on high purity glass preform and wire drawing capacity. From bare fibers to high-density optical cables, adaptation and termination are required.
Provides contract assembly and testing for optical communications gear and other precision optics.
Supply chain bottlenecks Optical module foundry requires a complete set of lasers, optical devices and driver chips. Precision alignment and batch testing determine qualified shipments.
Makes lasers, optical materials and transceivers for data center networking.
Supply chain bottlenecks High-speed optical devices rely on compound semiconductor wafers and epitaxial materials. Laser chips also undergo coupling, packaging and reliability testing.
Supplies optical chips, lasers and high-speed modules to data centers and communications networks.
Supply chain bottlenecks The output of optical communication lasers depends on compound materials such as indium phosphide, epitaxial yield and optical chip packaging capabilities.
Makes lasers and optical transceivers for data centers and communications networks.
Supply chain bottlenecks High-speed optical modules need to be equipped with homemade or purchased lasers, high-speed DSP and drivers, and then pass high temperature and link performance tests.
Supplies wireless access, IP routing and optical networking equipment and related software.
Supply chain bottlenecks Optical transmission equipment relies on coherent optical DSP, optical components and circuit board supplies. Before the network goes online, line and equipment commissioning must be completed.
Sells mobile, broadband and networking services to households and businesses.
Supply chain bottlenecks Fiber optic and wireless network expansion requires fiber optic cables, radio frequency equipment, site power and construction permits. Insufficient backhaul links limit available capacity.
Cloud & models
Rent computing power and storage through AWS, and develop Trainium training chips.
Supply chain bottlenecks AWS’s Trainium relies on external wafers and packaging. After the accelerator arrives, it still has to wait for power supply, cooling and high-speed network connection.
Read company research
Cloud computing, model hosting, and AI development services via Azure.
Supply chain bottlenecks The Azure cluster needs to receive accelerator cards, servers and high-voltage power supply and distribution equipment simultaneously. The new computer room still has to wait for grid connection and cooling-system commissioning.
Read company research
Operate Google Cloud and develop TPU chips and Gemini models.
Supply chain bottlenecks Google's TPU relies on advanced manufacturing, packaging, and high-bandwidth memory. Cluster size is limited by available power and optical interconnect equipment.
Read company research
Rent GPU clusters to model developers and enterprises.
Supply chain bottlenecks In self-built GPU campuses, power access and data center commissioning must be completed before the arrival of the new generation accelerator, otherwise the computing power cannot be brought online on time.
Read company research
Operate a GPU cloud platform for AI training and inference.
Supply chain bottlenecks GPU clusters rely on accelerator card supply, high-density computer rooms and campus optical fibers. Available computing power can only be formed after each batch of cabinets is powered on and tested for interoperability on the network.
Runs e-commerce platforms and sells storage, cloud computing and Qwen model services through Alibaba Cloud.
Supply chain bottlenecks Alibaba Cloud and Tongyi models rely on available AI acceleration chips, data center power, and distributed training networks. Chip imports are also affected by export licensing.
Runs a Chinese search and advertising business and offers AI cloud, ERNIE models and autonomous driving services.
Supply chain bottlenecks Baidu’s Wenxin model training and inference rely on accelerators, servers and distributed computing platforms. Self-developed chips also need to solve manufacturing and software adaptation.
Software & applications
Sells software that integrates data and deploys AI to businesses and governments.
Supply chain bottlenecks AIP can only run after accessing controlled data sources and available models. Rights restrictions, data formats, and local deployment environments determine which workflows can be connected.
Read company research
Sell customer relationship management software and AI sales assistants.
Supply chain bottlenecks Agentforce relies on unified and permissioned access to business data, model inference services, and connectors. Incomplete data can limit automated execution.
Software that provides automated IT services and enterprise workflows.
Supply chain bottlenecks Workflow agents rely on task records, knowledge bases, and external system interfaces. Identity permissions and interface availability determine the scope of automation.
Sells design, imaging, and video production software and provides generative AI tools.
Supply chain bottlenecks Firefly relies on appropriately authorized training materials and inference computing power. Image and video generation also requires sufficient storage and computing capacity.
Sells network, cloud and security operations protection products to enterprises.
Supply chain bottlenecks Hardware firewalls rely on specialized chips and OEM assembly. Cloud-side security analysis relies on log access, data processing and threat intelligence supply.
Read company research
Sells databases and enterprise software and provides cloud computing and GPU capacity through OCI.
Supply chain bottlenecks OCI’s GPU cluster must have accelerators, high-speed networks and grid-connected capacity. The new campus also relies on mechanical and electrical equipment delivery and contractor construction.
Sells enterprise software for finance, procurement, supply chain and human resources.
Supply chain bottlenecks Joule needs access to ERP data, permissions and available underlying models. Data migration and interface adaptation of the old system determine the deployment speed.
Provides cloud data storage, analytics and sharing, along with tools for building AI applications.
Supply chain bottlenecks The data platform relies on third-party clouds for computing, object storage, and regional capacity. AI functions also require access to inference computing power and controlled data.
Prepares training data for businesses and provides data labeling and AI model evaluation services.
Supply chain bottlenecks Training data engineering relies on annotators and reviewers with corresponding domain knowledge, authorized data, and secure computing environments.
Provides endpoint, cloud and identity security software through its Falcon platform.
Supply chain bottlenecks The security model relies on endpoint telemetry, threat intelligence and cloud processing capabilities. Changes in operating system interfaces can affect the signals that can be collected.
Provides content delivery, network security and edge computing, including AI inference on its edge network.
Supply chain bottlenecks Providing model inference at the edge requires configuring accelerators for nodes, as well as computer room space, power supply, and network interconnection.
Sells devices such as the iPhone and Mac as well as digital services. It integrates Apple Intelligence into supported devices.
Supply chain bottlenecks End-side AI relies on A-series and M-series chips with advanced processes, and is also limited by device memory and cooling budget.
Read company research
Runs Facebook, Instagram and WhatsApp, earns most of its revenue from advertising and develops Llama models.
Supply chain bottlenecks Training Llama requires available accelerator, HBM, computer room power and cluster interconnection. Self-developed accelerators also rely on external foundry and packaging.
Sells measurement instruments, valves and automation controls for factories and process industries.
Supply chain bottlenecks Industrial control systems rely on industrial-grade semiconductors, measurement sensors, and field controllers. Replacement of components restricted by specifications such as temperature resistance and vibration resistance.
Sells industrial automation hardware and software for controlling production lines.
Supply chain bottlenecks Automated production lines rely on programmable controllers, servo drives and machine vision hardware, and require integration testing with field equipment and control software.
Sells automated semiconductor test equipment and operates an industrial robotics business.
Supply chain bottlenecks Chip test equipment relies on high-precision test boards, probe interfaces and high-speed instrument modules. Different chip packages require special test procedures.
Sells electric vehicles and energy storage systems and offers FSD driver-assistance software.
Supply chain bottlenecks FSD training relies on the image data returned by the fleet and the computing power of the Cortex cluster. On-board deployment also requires mass-produced on-board inference chips and cameras.
Read company research
Develops, manufactures and sells vehicles and equips them with driver-assistance systems.
Supply chain bottlenecks Mass production of driver assistance systems relies on automotive-grade processors, sensors and software integration. After parts are replaced, they must be re-verified according to the model.
Develops driver-assistance and autonomous driving systems, including EyeQ chips and software.
Supply chain bottlenecks EyeQ autonomous driving chips rely on external foundry and packaging, and system mass production also requires automotive-grade cameras, calibration and vehicle integration verification.
Designs and installs warehouse systems that use robots to move, sort and stack goods.
Supply chain bottlenecks Automated warehousing deployment relies on motors, encoders and vision sensors for robots. After the mechanical and electrical installation, the robot and the shelf system must be jointly adjusted.
Streams films, series and live content. Subscriptions are its main source of revenue, alongside advertising.
Supply chain bottlenecks The recommendation model relies on legally available viewing data and cloud computing power. Video delivery additionally relies on Open Connect nodes, caching hardware, and network interconnects.
Read company research
Streams music, podcasts and audiobooks and earns revenue from subscriptions and advertising.
Supply chain bottlenecks Personalized recommendations rely on listening data and cloud computing. The availability of songs, podcasts, and audiobooks is also subject to content licensing.
Operates film, television, streaming services such as Disney+ and theme parks.
Supply chain bottlenecks Content cataloging and recommendation of streaming media rely on content metadata, cloud processing and encoding capabilities. The film libraries available in different regions are subject to copyright licensing restrictions.
Provides software and payment services for merchants to build stores, take payments and manage orders.
Supply chain bottlenecks AI product and operational assistants rely on structured catalogs, transaction data permissions and model services. Connecting payments and fulfillment also requires stable third-party interfaces.
Runs an advertising platform that helps app developers acquire users and sell ad inventory.
Supply chain bottlenecks Advertising models rely on in-app event callbacks and compliant attribution data available. Mobile operating system tracking permissions change available signals.
Operates ride-hailing, delivery and freight platforms and earns revenue from transactions on them.
Supply chain bottlenecks Platform scheduling relies on real-time positioning, map data and cloud computing. Self-driving transportation capacity also relies on partners to deliver suitable vehicles and systems.
The telemedicine platform uses Labs AI and MedMatch to assist in health information interpretation and personalized medical services.
Supply chain bottlenecks Medical AI relies on authorized health records, laboratory data, and clinical review. Data availability and privacy requirements limit the scope of deployment.
Use AI to analyze experimental data, screen targets, and design drug candidates.
Supply chain bottlenecks AI drug screening relies on high-quality experimental data, automated experimental equipment and training computing power. Candidates screened by the model must also be synthesized and verified experimentally.
Integrate generative AI assistants into financial and tax software such as QuickBooks and TurboTax.
Supply chain bottlenecks Accounting and tax preparation assistants rely on bank transaction data interfaces, up-to-date tax rules and available models. Missing accounting data limits automated processing.
Figma Make generates interactive prototypes using natural language.
Supply chain bottlenecks Figma Make relies on available model inference services and design components with read permissions. Generating prototypes also needs to be compatible with existing design systems.
Operate autonomous trucks with autonomous driving systems and provide related capabilities to freight customers.
Supply chain bottlenecks Unmanned trucks need to be equipped with lidar, cameras and on-board computing platforms, and then complete integration verification with mass-produced chassis, redundant braking and steering systems.
See & Spray uses machine vision to distinguish crops and weeds and spray according to target.
Supply chain bottlenecks Precise spraying relies on field cameras, edge computing modules and independently controllable nozzles. Illumination, occlusion and crop sample coverage can affect recognition.
Provide restaurant customers with voice AI that can answer calls and process orders.
Supply chain bottlenecks Restaurant voice AI relies on available voice reasoning computing power, menu data, and cashier interfaces. Stable audio input is also required in noisy environments.
Sales enterprise AI development platform and industry applications.
Supply chain bottlenecks C3 AI deployment relies on accessible enterprise databases, industrial sensor data, and authorized models. Connectors and data quality limit the applications that can be brought online.