{"id":"5c5c1d6d-90e3-4bc7-8eb3-03fac08da003","name":"Huawei Cloud","slug":"huawei-cloud","description":"Huawei Cloud is the public-cloud arm of Huawei Technologies, spun up as a dedicated business unit in 2017 and now operating 34 regions and 103 availability zones serving customers in more than 170 countries. Unlike the pure-play neocloud pack, it is a hyperscaler-scale operation embedded inside a $118 billion telecoms-equipment parent, and its differentiator is silicon: because Huawei has been on the US Entity List since 2019 and cut off from Nvidia's top-end accelerators, its AI cloud runs almost entirely on the in-house Ascend line (910B and, since 2025, 910C) rather than H100/H200/B200.\n\nThe flagship AI product is the CloudMatrix 384 rack system, which lashes 384 Ascend 910C chips together across 16 racks to deliver roughly 300 PFLOPs of dense BF16 compute, nearly double an Nvidia GB200 NVL72 at the rack level. Per SemiAnalysis it draws about 4.1x the power of a GB200 NVL72 with materially worse power-per-FLOP, a trade Huawei can make because Chinese customers have abundant grid capacity. TSMC is reported to have delivered around 1.05 million Ascend 910C dies across 2024 and 2025 before US pressure tightened, giving Huawei Cloud the largest domestic-Chinese pool of leading-edge AI silicon.\n\nThe customer base skews Chinese state-owned enterprises, banks, telcos, and government, plus internal use by Huawei's own PanGu foundation-model program; overseas expansion has concentrated on the Middle East, Southeast Asia, Latin America, and Africa where US export controls do not block sales. Positioning is both training and inference at hyperscale, and Huawei Cloud is consistently ranked the #2 IaaS provider in China behind Alibaba Cloud.","status":"active","founded":2017,"hq":"Shenzhen, China","hqLat":null,"hqLng":null,"hqGeocodedAt":null,"hqAddress":null,"hqPrecision":null,"website":"https://www.huaweicloud.com","fundingTotal":null,"type":null,"roles":[],"reviewStatus":"reviewed","lifecycleState":"active","supersededByCompanyId":null,"sources":[{"url":"https://www.huaweicloud.com/intl/en-us/about/","title":"About Huawei Cloud (official)","publisher":"Huawei Cloud"},{"url":"https://newsletter.semianalysis.com/p/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72","date":"2025-04-16","title":"Huawei AI CloudMatrix 384: China's Answer to Nvidia GB200 NVL72","publisher":"SemiAnalysis"},{"url":"https://en.wikipedia.org/wiki/Huawei","title":"Huawei","publisher":"Wikipedia"},{"url":"https://en.wikipedia.org/wiki/Huawei_PanGu","title":"Huawei PanGu","publisher":"Wikipedia"}],"keyFacts":[{"label":"Parent company","value":"Huawei Technologies (private, employee-owned; $118.2B group revenue in 2024)","sourceUrl":"https://en.wikipedia.org/wiki/Huawei"},{"label":"Global footprint","value":"34 regions, 103 availability zones, service in 170+ countries; 3 liquid-cooled AI computing centers disclosed","sourceUrl":"https://www.huaweicloud.com/intl/en-us/about/"},{"label":"AI silicon","value":"Huawei Ascend 910B and 910C (in-house); no Nvidia H100/H200/B200 due to US export controls","sourceUrl":"https://newsletter.semianalysis.com/p/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72"},{"label":"Flagship AI rack","value":"CloudMatrix 384: 384 Ascend 910C chips across 16 racks, ~300 PFLOPs dense BF16, 3.6x GB200 NVL72 aggregate memory","sourceUrl":"https://newsletter.semianalysis.com/p/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72"},{"label":"Ascend die supply","value":"~1.05 million Ascend 910C dies delivered by TSMC across 2024-2025 per SemiAnalysis; Huawei has not publicly disclosed its own inventory","sourceUrl":"https://newsletter.semianalysis.com/p/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72"},{"label":"Power profile","value":"CloudMatrix 384 draws ~4.1x the power of an Nvidia GB200 NVL72 per SemiAnalysis; total operational MW not disclosed by Huawei","sourceUrl":"https://newsletter.semianalysis.com/p/huawei-ai-cloudmatrix-384-chinas-answer-to-nvidia-gb200-nvl72"},{"label":"Enterprise customers","value":"2,600+ enterprises on the platform per Huawei Cloud's own disclosure","sourceUrl":"https://www.huaweicloud.com/intl/en-us/about/"},{"label":"Anchor tenant / internal use","value":"Huawei's PanGu foundation-model program trained on Ascend 910 clusters","sourceUrl":"https://en.wikipedia.org/wiki/Huawei_PanGu"},{"label":"US sanctions status","value":"Parent Huawei on US Entity List since 2019; blocks access to Nvidia top-end GPUs and leading-edge foundry capacity","sourceUrl":"https://en.wikipedia.org/wiki/Huawei"},{"label":"Positioning","value":"Hyperscaler both training and inference; #2 IaaS in China behind Alibaba Cloud; overseas push in Middle East, SE Asia, LatAm, Africa","sourceUrl":"https://www.huaweicloud.com/intl/en-us/about/"},{"label":"Public status","value":"Private; wholly owned by Huawei Technologies, itself employee-owned and unlisted"},{"label":"Hourly rental pricing","value":"Not publicly disclosed in Western coverage; Ascend instance pricing is quoted per project in-region rather than as a standard on-demand US-dollar rate"}],"aliases":[],"collisionRisk":"low","reviewNote":null,"atsProvider":null,"atsSlug":null,"factoryLocations":null,"manufacturingCapacity":null,"orgType":null,"burnSignal":null,"litigationEvents":null,"insurancePostureDisclosed":null,"unitEconomicsDisclosed":null,"tickerSymbol":null,"stockExchange":null,"secCik":null,"revenueUsd":null,"revenueBasis":null,"revenueAsOf":null,"employeeCount":null,"employeeCountBasis":"Huawei Cloud does not publish a standalone headcount; parent Huawei Technologies reported ~207,000 employees in 2024.","employeeCountAsOf":null,"marketCapUsd":null,"marketCapAsOf":null,"marketCapSource":null,"createdAt":"2026-08-28T01:48:31.674Z","updatedAt":"2026-08-28T03:45:17.421Z","jsonLd":{"@context":"https://schema.org","@type":"Organization","@id":"https://registry.deploy.report/companies/huawei-cloud","url":"https://registry.deploy.report/companies/huawei-cloud","name":"Huawei Cloud","description":"Huawei Cloud is the public-cloud arm of Huawei Technologies, spun up as a dedicated business unit in 2017 and now operating 34 regions and 103 availability zones serving customers in more than 170 countries. Unlike the pure-play neocloud pack, it is a hyperscaler-scale operation embedded inside a $118 billion telecoms-equipment parent, and its differentiator is silicon: because Huawei has been on the US Entity List since 2019 and cut off from Nvidia's top-end accelerators, its AI cloud runs almost entirely on the in-house Ascend line (910B and, since 2025, 910C) rather than H100/H200/B200.\n\nThe flagship AI product is the CloudMatrix 384 rack system, which lashes 384 Ascend 910C chips together across 16 racks to deliver roughly 300 PFLOPs of dense BF16 compute, nearly double an Nvidia GB200 NVL72 at the rack level. Per SemiAnalysis it draws about 4.1x the power of a GB200 NVL72 with materially worse power-per-FLOP, a trade Huawei can make because Chinese customers have abundant grid capacity. TSMC is reported to have delivered around 1.05 million Ascend 910C dies across 2024 and 2025 before US pressure tightened, giving Huawei Cloud the largest domestic-Chinese pool of leading-edge AI silicon.\n\nThe customer base skews Chinese state-owned enterprises, banks, telcos, and government, plus internal use by Huawei's own PanGu foundation-model program; overseas expansion has concentrated on the Middle East, Southeast Asia, Latin America, and Africa where US export controls do not block sales. Positioning is both training and inference at hyperscale, and Huawei Cloud is consistently ranked the #2 IaaS provider in China behind Alibaba Cloud.","identifier":"5c5c1d6d-90e3-4bc7-8eb3-03fac08da003","foundingDate":"2017","address":"Shenzhen, China","publisher":{"@id":"https://deploy.report/#organization"}},"framework_metadata":{"framework_schema_version":"0.1.0","verification_status":"verified","maturity_stage":null,"lifecycle_state":null,"architectural_position":{"cohort":null,"sub_cohorts":[]},"within_cohort_verified_vs_claimed_pair":null,"cap_flags":[],"verification_depth":{"sources_count":4,"primary_source_types":["knowledge-base"]}}}