H

Huawei

Hardware

Chinese telecom and devices conglomerate building Ascend AI accelerators and the CANN heterogeneous computing stack.

Total funding—

Valuation—

Employees213,000

Founded1987

HeadquartersShenzhen, China

StatusPrivate

Products

Selected offerings — the AI products recorded here, not the vendor's full catalogue.

• Atlas Hardware

  Huawei's line of AI training and inference processors (NPUs) and the systems built on them -- the platform brand for the silicon itself (still called Ascend in some regional markets and in the underlying chip generation names), sold standalone and in Atlas-branded servers and SuperPoD clusters, now recorded in their own separate hardware and software-stack products.

• Atlas 650E AI Server Hardware

  14U AI server powered by eight Huawei 950DT NPUs, rated at up to 12.4 PFLOPS at mxFP4, for on-premises AI training and inference in finance, government and healthcare deployments.

• Atlas 950 SuperPoD Hardware

  Rack-scale AI supercomputing cabinet built from 64 Huawei 950DT NPUs per cabinet and scalable to 1,024 NPUs over a UB Link fabric for trillion-parameter model training and inference.

• CANN Developer tool Sold on its own

  Huawei's heterogeneous compute architecture for its Ascend/Atlas NPUs, supplying the operator libraries, compiler and programming interfaces that bridge AI frameworks to the hardware.

• MindIE Developer tool Sold on its own

  Inference engine and serving framework for Atlas/Ascend hardware that deploys LLM and diffusion models behind unified APIs compatible with vLLM, OpenAI and Triton interfaces.

• MindSpore Developer tool Sold on its own

  Open-source AI framework originated by Huawei for building, training and deploying models with native distributed training, best optimised for Huawei's Ascend/Atlas processors.

• MindStudio Developer tool Sold on its own

  End-to-end development toolchain for Atlas/Ascend AI applications, covering custom operator development, model conversion and compression, accuracy debugging and performance profiling via MindStudio Insight.

Closest alternatives 1

NNVIDIAPublic · $5.1T9 of 9 capabilitiesShared product typeNamed in filingBuy data-centre AI computeCompare

Similar products 90

How this list is ranked ↗

AAMDPublic · $785.1BAtlas and AMD Instinct do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).8 of 9 capabilitiesShared product typeCompareDTDell TechnologiesPublic · $360.7BAtlas 650E AI Server and Dell PowerEdge XE9685L do comparable work on AI compute hardware, model training and server systems; both also serve buyers who need to buy data-centre AI compute; larger scale (public).5 of 9 capabilitiesShared product typeCompareIIntelPublic · $463.7BAtlas and Intel Gaudi AI Accelerator do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).4 of 9 capabilitiesShared product typeCompareQQualcommPublic · $174.9BAtlas and Qualcomm Cloud AI 100 Ultra do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; Huawei's scale not recorded.5 of 9 capabilitiesShared product typeCompareAAAxelera AIPrivate · $450M raisedAtlas and Axelera Metis AIPU do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; Huawei's scale not recorded.3 of 9 capabilitiesShared product typeCompareUUnslothPrivateMindSpore and Unsloth do comparable work on model training; both also serve buyers who need to run an open-weight model on your own infrastructure, serve a model in production and train or fine-tune a model; scale not recorded for either.2 of 9 capabilitiesShared product typeCompareCCiscoPublic · $433.3BAtlas 950 SuperPoD and Cisco Nexus Hyperfabric AI do comparable work on AI compute hardware, interconnect, model training and 1 more; both also serve buyers who need to buy data-centre AI compute and connect AI clusters; larger scale (public).7 of 9 capabilitiesShared product typeCompareGGraphcorePrivateAtlas and IPU (Intelligence Processing Unit) do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; scale not recorded for either.4 of 9 capabilitiesShared product typeCompareANArista NetworksPublic · $246.4BAtlas 950 SuperPoD and 7060X5 Series do comparable work on AI compute hardware and interconnect; both also serve buyers who need to buy data-centre AI compute and connect AI clusters; Huawei's scale not recorded.2 of 9 capabilitiesShared product typeCompareCSCerebras SystemsPublic · $44.4BAtlas and Cerebras WSE-3 Turbo (WSE-3T) do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; smaller scale (public).3 of 9 capabilitiesShared product typeCompareGGroqPrivate · $3.5BAtlas and Groq LPU (GroqChip) do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; smaller scale (private).2 of 9 capabilitiesShared product typeCompareHTHailo TechnologiesAcquired ($340M deal)MindIE and Hailo AI Software Suite both serve buyers who need to run an open-weight model on your own infrastructure and serve a model in production; Huawei's scale not recorded; Hailo Technologies is acquired, with no independent scale.3 of 9 capabilitiesShared product typeCompareTTenstorrentPrivate · $2.6BAtlas and Blackhole do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; smaller scale (private).4 of 9 capabilitiesShared product typeCompareSSambaNovaPrivate · $11BAtlas and SambaNova RDU do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; smaller scale (private).4 of 9 capabilitiesShared product typeCompareBBroadcomPublic · $1.7TAtlas and Broadcom custom silicon do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; larger scale (public).2 of 9 capabilitiesShared product typeCompareMTMarvell TechnologyPublic · $214.9BAtlas 950 SuperPoD and Marvell OCTEON DPU do comparable work on interconnect; both also serve buyers who need to buy data-centre AI compute and connect AI clusters; Huawei's scale not recorded.2 of 9 capabilitiesShared product typeCompareCCrusoePrivate · $10BAtlas 650E AI Server and Crusoe Spark do comparable work on AI compute hardware and server systems; both also serve buyers who need to buy data-centre AI compute; smaller scale (private).6 of 9 capabilitiesShared product typeCompareSHSK HynixPublic · $874.5BAtlas 650E AI Server and HBM3E do comparable work on AI compute hardware and model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).3 of 9 capabilitiesShared product typeCompareKKioxiaPublic · $177.9BAtlas 650E AI Server and KIOXIA CM9-V Series do comparable work on AI compute hardware; both also serve buyers who need to buy data-centre AI compute; Huawei's scale not recorded.1 of 9 capabilitiesShared product typeCompareDDecartPrivate · $4BCANN and Decart Optimization Stack do comparable work on GPU programming and model training; both also serve buyers who need to program GPU accelerators, run an open-weight model on your own infrastructure, serve a model in production and 1 more; smaller scale (private); ships API service, application and 2 more rather than the same layer.4 of 9 capabilitiesDifferent layerCompareMModularAcquired ($3.1B deal)MindStudio and Mojo do comparable work on GPU programming; both also serve buyers who need to program GPU accelerators; Huawei's scale not recorded; Modular is acquired, with no independent scale.3 of 9 capabilitiesShared product typeCompareSESamsung ElectronicsPublic · $1.2TAtlas 650E AI Server and Samsung Datacenter SSD do comparable work on AI compute hardware; both also serve buyers who need to buy data-centre AI compute; larger scale (public).2 of 9 capabilitiesShared product typeCompareOOllamaPrivate · $88M raisedMindIE and Ollama do comparable work on model hosting; both also serve buyers who need to run an open-weight model on your own infrastructure and serve a model in production; Huawei's scale not recorded; ships infrastructure service and model API rather than the same layer.2 of 9 capabilitiesDifferent layerCompareBBaiduPublic · $31.6BMindSpore and PaddlePaddle do comparable work on model training; both also serve buyers who need to run an open-weight model on your own infrastructure, serve a model in production and train or fine-tune a model; smaller scale (public).2 of 9 capabilitiesShared product typeCompareHRHorizon RoboticsPublic · $8.8BAtlas and Journey 6 Series do comparable work on accelerator silicon; both also serve buyers who need to buy data-centre AI compute; smaller scale (public).3 of 9 capabilitiesShared product typeCompareCTCredo Technology GroupPublicAtlas 950 SuperPoD and ZeroFlap / OmniConnect do comparable work on interconnect; both also serve buyers who need to connect AI clusters; scale not recorded for either.1 of 9 capabilitiesShared product typeCompareALAstera LabsPublic · $50.4BAtlas 950 SuperPoD and Aries Smart Cable Modules do comparable work on interconnect; both also serve buyers who need to connect AI clusters; smaller scale (public).1 of 9 capabilitiesShared product typeCompareALAyar LabsPrivate · $3.8BAtlas 950 SuperPoD and TeraPHY do comparable work on interconnect; both also serve buyers who need to connect AI clusters; smaller scale (private).1 of 9 capabilitiesShared product typeCompareLLightmatterPrivate · $4.4BAtlas 950 SuperPoD and Passage do comparable work on interconnect; both also serve buyers who need to connect AI clusters; smaller scale (private).1 of 9 capabilitiesShared product typeCompareRRanovusPrivateAtlas 950 SuperPoD and Odin do comparable work on interconnect; both also serve buyers who need to connect AI clusters; scale not recorded for either.1 of 9 capabilitiesShared product typeCompareRRunPodPrivateMindIE and Runpod Hub do comparable work on model hosting; both also serve buyers who need to serve a model in production; scale not recorded for either.4 of 9 capabilitiesShared product typeCompare44ParadigmPublic · $2BAtlas 650E AI Server and 4Paradigm SageOne do comparable work on server systems; both also serve buyers who need to buy data-centre AI compute; smaller scale (public).3 of 9 capabilitiesShared product typeCompareTMThinking Machines LabPrivate · $12BMindSpore and Tinker do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; smaller scale (private); ships API service rather than the same layer.2 of 9 capabilitiesDifferent layerCompareAMApplied MaterialsPublic · $380.7BAtlas and SEMVision H20 do comparable work on model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).2 of 9 capabilitiesShared product typeCompareLLambdaPrivate · $1.5B raisedMindIE and Lambda Inference do comparable work on model hosting; both also serve buyers who need to serve a model in production; Huawei's scale not recorded; ships infrastructure service and model API rather than the same layer.4 of 9 capabilitiesDifferent layerCompareNGNebius GroupPublic · $61.6BMindIE and Nebius Token Factory do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (public).4 of 9 capabilitiesShared product typeCompareAAlibabaPublic · $273.4BMindSpore and ModelScope do comparable work on model training; both also serve buyers who need to run an open-weight model on your own infrastructure, serve a model in production and train or fine-tune a model; larger scale (public); ships AI agent, API service and 4 more rather than the same layer.4 of 9 capabilitiesDifferent layerCompareBBasetenPrivate · $13BMindSpore and Baseten do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; smaller scale (private); ships API service, infrastructure service and 2 more rather than the same layer.3 of 9 capabilitiesDifferent layerCompareAAlphabet (Google)Public · $4.2TAtlas and Google Cloud TPU do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).5 of 9 capabilitiesShared product typeCompareFAFireworks AIPrivate · $17.5BMindIE and Fireworks Inference do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (private); ships API service and model API rather than the same layer.3 of 9 capabilitiesDifferent layerCompareAAmazonPublic · $2.8TAtlas and AWS Trainium do comparable work on accelerator silicon and model training; both also serve buyers who need to buy data-centre AI compute; larger scale (public).6 of 9 capabilitiesShared product typeCompareMModalPrivate · $466M raisedMindIE and Modal Inference do comparable work on model hosting; both also serve buyers who need to serve a model in production; Huawei's scale not recorded.3 of 9 capabilitiesShared product typeCompareOOraclePublic · $454.4BMindSpore and OCI Data Science do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; larger scale (public).3 of 9 capabilitiesShared product typeCompareTATogether AIPrivate · $8.3BMindIE and Together Dedicated Container Inference do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (private).3 of 9 capabilitiesShared product typeCompareLALiquid AIPrivateMindSpore and LEAP do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; scale not recorded for either; ships application and platform rather than the same layer.2 of 9 capabilitiesDifferent layerCompareAHArm HoldingsPublic · $282.8BAtlas and Arm CSS for Mobile 2 do comparable work on accelerator silicon; larger scale (public).3 of 9 capabilitiesShared product typeCompareKEKakao EnterprisePrivateMindIE and Kubeflow do comparable work on model hosting; both also serve buyers who need to serve a model in production; scale not recorded for either.4 of 9 capabilitiesShared product typeCompareCMComet MLPrivate · $70M raisedMindSpore and Comet Artifacts do comparable work on model training; both also serve buyers who need to train or fine-tune a model; Huawei's scale not recorded.2 of 9 capabilitiesShared product typeCompareHCHuawei CloudPrivateMindSpore and Huawei HiLens do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; scale not recorded for either.4 of 9 capabilitiesShared product typeCompareHFHugging FacePrivate · $4.5BMindIE and Spaces do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (private); ships API service, application and 2 more rather than the same layer.4 of 9 capabilitiesDifferent layerCompareCCCactus ComputePrivateMindSpore and Cactus Platform do comparable work on model training; both also serve buyers who need to train or fine-tune a model; scale not recorded for either; ships infrastructure service and platform rather than the same layer.2 of 9 capabilitiesDifferent layerCompareRReplicatePrivateMindIE and Replicate do comparable work on model hosting; both also serve buyers who need to serve a model in production; scale not recorded for either; ships infrastructure service rather than the same layer.2 of 9 capabilitiesDifferent layerCompareEEleutherAINonprofitMindSpore and GPT-NeoX do comparable work on model training; both also serve buyers who need to train or fine-tune a model; Huawei's scale not recorded; EleutherAI is a nonprofit, with no independent scale.2 of 9 capabilitiesShared product typeCompareGGitLabPublic · $8.3BMindIE and GitLab Duo Agent Platform Self-Hosted do comparable work on model hosting; both also serve buyers who need to run an open-weight model on your own infrastructure and serve a model in production; smaller scale (public).2 of 9 capabilitiesShared product typeCompareCCoreWeavePublic · $48.5BMindSpore and CoreWeave AI Cloud do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; smaller scale (public); ships AI agent, infrastructure service and 1 more rather than the same layer.3 of 9 capabilitiesDifferent layerCompareNCNAVER CloudPrivateMindSpore and AiTEMS do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; scale not recorded for either; ships API service, application and 1 more rather than the same layer.3 of 9 capabilitiesDifferent layerCompareTTencentPublic · $510.1BMindSpore and Tencent Cloud TI Platform do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; larger scale (public).3 of 9 capabilitiesShared product typeCompareDDatabricksPrivate · $190BMindSpore and Databricks Model Training do comparable work on model training; both also serve buyers who need to train or fine-tune a model; Huawei's scale not recorded.4 of 9 capabilitiesShared product typeCompareW&Weights & BiasesAcquiredMindIE and W&B Serverless Inference do comparable work on model hosting; both also serve buyers who need to serve a model in production; Huawei's scale not recorded; Weights & Biases is acquired, with no independent scale.4 of 9 capabilitiesShared product typeCompareKKuaishouPublic · $18.6BMindSpore and Vanchin do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; smaller scale (public).3 of 9 capabilitiesShared product typeCompareSSenseTimePublic · $7.7BMindSpore and SenseCore do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; smaller scale (public).4 of 9 capabilitiesShared product typeCompareLSLM StudioPrivateMindIE and LM Studio do comparable work on model hosting; both also serve buyers who need to serve a model in production; scale not recorded for either; ships AI agent and application rather than the same layer.2 of 9 capabilitiesDifferent layerCompareCCore42PrivateAtlas 650E AI Server and Core42 Sovereign AI Cloud do comparable work on AI compute hardware; both also serve buyers who need to buy data-centre AI compute; scale not recorded for either; ships API service and infrastructure service rather than the same layer.2 of 9 capabilitiesDifferent layerCompareVVercelPrivate · $9.3BMindIE and Vercel AI Gateway do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (private).3 of 9 capabilitiesShared product typeCompareCCoherePrivate · $7BMindIE and Model Vault do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (private); ships API service, application and 2 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareAASMLPublic · $670.1BAtlas and ASML DUV lithography systems both serve buyers who need to buy data-centre AI compute; larger scale (public).Shared product typeCompareTTSMCPublic · $2TAtlas and TSMC 3DFabric both serve buyers who need to buy data-centre AI compute; larger scale (public).Shared product typeCompareYYandexPublicMindSpore and Yandex DataSphere do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; scale not recorded for either.3 of 9 capabilitiesShared product typeCompareMMicrosoftPublic · $3.7TMindSpore and Azure Machine Learning do comparable work on model training; both also serve buyers who need to train or fine-tune a model; larger scale (public).4 of 9 capabilitiesShared product typeCompareLLabelboxPrivate · $189M raisedMindIE and Labelbox Foundry both serve buyers who need to serve a model in production; Huawei's scale not recorded.3 of 9 capabilitiesShared product typeCompareVVMwareAcquiredMindIE and VMware Cloud Foundation Private AI Services do comparable work on model hosting; both also serve buyers who need to serve a model in production; Huawei's scale not recorded; VMware is acquired, with no independent scale; ships infrastructure service rather than the same layer.3 of 9 capabilitiesDifferent layerCompareHHumainPrivateMindSpore and HUMAIN COMPUTE do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; scale not recorded for either; ships API service, agent platform and 1 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareMPMeta PlatformsPublic · $1.5TMindIE and Meta Model API do comparable work on model hosting; both also serve buyers who need to serve a model in production; larger scale (public); ships AI agent, agent platform and 2 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareCCivitaiPrivate · $5.1M raisedMindIE and Civitai API do comparable work on model hosting; both also serve buyers who need to serve a model in production; Huawei's scale not recorded; ships API service and platform rather than the same layer.2 of 9 capabilitiesDifferent layerCompareTTuringPrivate · $2.2BMindSpore and RL environments do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships agent platform and data service rather than the same layer.2 of 9 capabilitiesDifferent layerCompareIIBMPublic · $216.6BMindSpore and IBM watsonx.ai do comparable work on model training; both also serve buyers who need to serve a model in production and train or fine-tune a model; Huawei's scale not recorded.3 of 9 capabilitiesShared product typeCompareALAI21 LabsPrivate · $1.4BMindSpore and AI21 Post-Training do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships API service, agent platform and 2 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareUUiPathPublic · $7.2BMindSpore and UiPath AI Center do comparable work on model training; smaller scale (public).2 of 9 capabilitiesShared product typeCompareFFalPrivateMindIE and fal Serverless do comparable work on model hosting; both also serve buyers who need to serve a model in production; scale not recorded for either; ships AI agent, infrastructure service and 1 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareMAMistral AIPrivate · $13.8BMindSpore and Mistral Forge do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships AI agent, API service and 3 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareEElasticPublic · $10.2BMindIE and Elastic Inference Service do comparable work on model hosting; both also serve buyers who need to serve a model in production; smaller scale (public); ships agent platform, application and 2 more rather than the same layer.3 of 9 capabilitiesDifferent layerCompareSSiemensPublic · $244.7BMindSpore and Rapidminer AI Studio do comparable work on model training; both also serve buyers who need to train or fine-tune a model; Huawei's scale not recorded; ships AI agent, application and 1 more rather than the same layer.3 of 9 capabilitiesDifferent layerCompareTTolokaPrivateMindSpore and Toloka Train do comparable work on model training; both also serve buyers who need to train or fine-tune a model; scale not recorded for either; ships AI agent, data service and 1 more rather than the same layer.2 of 9 capabilitiesDifferent layerCompareHHeliconeAcquiredMindIE and Helicone AI Gateway do comparable work on model hosting; Huawei's scale not recorded; Helicone is acquired, with no independent scale; ships API service and platform rather than the same layer.2 of 9 capabilitiesDifferent layerCompareIMInsilico MedicinePublic · $4.3BMindSpore and Chemistry42 do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (public).2 of 9 capabilitiesShared product typeCompareSESchneider ElectricPublic · $200.4BAtlas 650E AI Server and NetShelter Power Rack 800VDC do comparable work on AI compute hardware; Huawei's scale not recorded.2 of 9 capabilitiesShared product typeCompareNNeo4jPrivate · $2BMindSpore and Neo4j Graph Data Science do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships agent platform, application and 1 more rather than the same layer.1 of 9 capabilitiesDifferent layerCompareSASnorkel AIPrivate · $1.3BMindSpore and Snorkel Flow do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships agent platform, data service and 1 more rather than the same layer.1 of 9 capabilitiesDifferent layerCompareRRunwayPrivate · $5.3BCANN and Runway Video Model Licensing do comparable work on model training; smaller scale (private).1 of 9 capabilitiesShared product typeCompareBFBlack Forest LabsPrivate · $3.3BMindSpore and Open Weights Licensing do comparable work on model training; both also serve buyers who need to train or fine-tune a model; smaller scale (private); ships application, infrastructure service and 1 more rather than the same layer.1 of 9 capabilitiesDifferent layerCompare

Closest alternatives are pairs a vendor page itself compares. Similar products are matched from what each product does; nothing in that group is asserted as a rivalry.

Sources

Employees

Company disclosure · 23 Sep 2026

Huawei's corporate-information page: "We have approximately 213,000 employees and we operate in more than 170 countries and regions". Stated directly, replacing a figure derived from the R&D share. Huawei is employee-held and carries no market valuation.

Founded

Wikipedia · 29 Aug 2026

Founded September 1987 in Shenzhen.

Revenue (TTM)

Regulatory filing · 31 Dec 2025

Huawei 2025 Annual Report, five-year financial highlights: 2025 revenue USD 126,018 million (CNY 880,941 million), "converted into United States dollars ("USD") using the closing rate at the end of 2025 of USD1.00 = CNY6.9906". The stored USD 125.8B did not follow from that rate.

Description

Official documentation · 6 Sep 2026

"The largest AI-silicon programme outside the US" carries no metric and no date, and Huawei's own Ascend site makes no size or ranking claim at all — it names CANN as the "heterogeneous computing architecture", which is what the sentence now says.

Open Huawei in the directory

Something wrong here? Send a correction — quote this company id: huawei.