Huawei Noah's Ark Lab

Huawei Noah's Ark Lab is the artificial intelligence research division of Huawei Technologies Co., Ltd., the Chinese telecommunications and consumer-electronics company headquartered in Shenzhen, Guangdong, China. The lab was established in 2012 with its initial Hong Kong base and has since expanded to additional research labs in Beijing, Shenzhen, Shanghai, London, Paris, and Edmonton (Canada). It develops the Pangu foundation model series (covering language, vision, multimodal, and scientific computing variants), the AI capability underlying Huawei's Ascend AI compute platform and CANN (Compute Architecture for Neural Networks) software stack, and published research output across machine learning, natural language processing, computer vision, decision-making AI, and other areas. As of April 2026, Noah's Ark Lab is one of the principal Chinese industrial AI research divisions, with integration into Huawei's broader product line spanning HarmonyOS, the Mate and Pura smartphone series, the cloud business, and the Ascend AI compute platform.

At a glance

• Founded: 2012 in Hong Kong; subsequently expanded to additional research labs in Beijing, Shenzhen, Shanghai, London, Paris, and Edmonton.
• Status: AI research division of Huawei Technologies Co., Ltd., a privately held Chinese company.
• Funding: Operates within Huawei's overall research-and-development budget. Huawei reported approximately CNY 180 billion (approximately $25 billion) in R&D spending in fiscal 2024, with Noah's Ark Lab representing a portion of that allocation on AI research.
• Senior leadership: Hua Wu, Chief Scientist of Huawei Noah's Ark Lab. Huawei senior leader. Yiqun Liu, Director of Noah's Ark Lab.
• Other notable leadership: Tong Xiao (Director, Edmonton lab), Zheng Zhang (Director, Hong Kong lab through 2020). Xian-Sheng Hua, former senior researcher at Noah's Ark Lab who later moved to Alibaba DAMO. Jian Li, Senior Researcher.
• Open weights: Yes, partial. Selected Pangu variants and other research code released through GitHub and Huawei's open-source MindSpore framework. The principal Pangu foundation model weights are predominantly released for Huawei Cloud commercial customers rather than as fully open weights.
• Flagship outputs: Pangu foundation model series (Pangu-Σ, Pangu-α, Pangu-Coder, Pangu-Drug, Pangu Tianshu, Pangu-CV, Pangu-Speech, Pangu-Weather), the MindSpore open-source deep-learning framework (originally developed at Noah's Ark Lab), published research output at NeurIPS, ICML, CVPR, ACL, AAAI, and other venues, and integration into Huawei's broader product line.

Origins

Huawei Noah's Ark Lab was established in 2012 in Hong Kong as Huawei's first dedicated AI research division, with founding leadership from senior Huawei research executives and recruitment from Hong Kong universities and other research institutions. The lab's name (referring to Noah's Ark) reflected the founding framing of preserving and advancing AI research capability within Huawei's broader research portfolio.

The 2012 to 2018 founding period built out Noah's Ark Lab's research capacity across the principal research areas: machine learning, natural language processing, computer vision, decision-making AI, and other areas. The lab opened additional research locations in Beijing, Shenzhen, Shanghai, and London through 2014 to 2017, with the Edmonton (Canada) lab subsequently established in 2018 in partnership with the University of Alberta.

The 2019 to 2021 period saw Noah's Ark Lab's transition under the US Department of Commerce restrictions on Huawei (the Entity List addition in May 2019, with subsequent commercial restrictions on US technology access). The lab's research-collaboration relationships with US universities and research institutions were curtailed; Noah's Ark Lab continued research output through Chinese, European, and Canadian research relationships. The August 2020 transition saw Huawei's HiSilicon-affiliated Ascend AI compute platform expand to address the supply-chain restrictions on advanced semiconductor access.

The 2021 to 2024 period saw Noah's Ark Lab's research output on the Pangu foundation model series. Pangu-α (April 2021) was a 200-billion-parameter Chinese-language autoregressive language model trained on Ascend compute, briefly the largest published Chinese-language foundation model. Subsequent Pangu variants extended the family across natural language processing, computer vision, multimodal AI, scientific computing (Pangu-Weather for weather forecasting, with published results at the European Centre for Medium-Range Weather Forecasts benchmarks), drug discovery (Pangu-Drug), and other application domains.

The 2024 to 2026 period has seen continued Pangu series iteration with Pangu-Σ (a trillion-parameter mixture-of-experts foundation model) and other variants. The Huawei Cloud commercial integration through ModelArts (the cloud AI platform) has anchored Noah's Ark Lab's commercial-product positioning. The continued cadence of Mate and Pura smartphone series releases with AI capability has continued the consumer-product integration.

Mission and strategy

Huawei Noah's Ark Lab's stated mission is to advance fundamental AI research and to integrate AI capability into Huawei's broader product line, with emphasis on foundation models, AI compute infrastructure, and industrial-research publication output. The strategic premise reflects Huawei's vertical integration across consumer electronics, telecommunications infrastructure, cloud computing, AI compute (Ascend), and the deep-learning framework (MindSpore).

The strategy combines four threads. First, the Pangu foundation model series with integration into Huawei Cloud's ModelArts platform and the broader Huawei product line. Second, the AI compute infrastructure through the Ascend AI accelerator chips and the CANN software stack, providing Huawei with vertical-integration alternatives to NVIDIA-based AI compute. Third, the MindSpore open-source deep-learning framework providing the principal alternative to PyTorch and TensorFlow within the Huawei AI ecosystem. Fourth, the published research output across major academic venues continuing the Noah's Ark Lab credibility internationally.

The competitive premise is that Huawei's vertical integration across consumer electronics, telecommunications, cloud computing, AI compute (Ascend), and the deep-learning framework (MindSpore), combined with the China-domestic supply-chain resilience under the US Department of Commerce restrictions, provide Huawei with a durable structural advantage as one of the principal Chinese industrial AI research and product organizations.

Distribution channels include direct integration into Huawei's product line (HarmonyOS, Mate and Pura smartphone series, the Ascend AI compute platform, the Huawei Cloud commercial cloud business with ModelArts), published research output through major academic venues, and open-source contributions through MindSpore and selected GitHub repositories.

Models and products

• Pangu foundation model series. Pangu-α (April 2021, 200-billion-parameter Chinese-language autoregressive language model), Pangu-Σ (trillion-parameter mixture-of-experts model), Pangu-Coder (code generation), Pangu-Drug (drug discovery), Pangu-CV (computer vision), Pangu-Speech (speech), Pangu-Weather (weather forecasting), Pangu Tianshu (multimodal scientific computing). Predominantly available through Huawei Cloud's ModelArts platform.
• MindSpore. Open-source deep-learning framework originally developed at Noah's Ark Lab; the principal alternative to PyTorch and TensorFlow within the Huawei AI ecosystem. Open-source through GitHub.
• Ascend AI compute platform. Huawei's AI accelerator chip line including Ascend 910 (training) and Ascend 310 (inference), with the CANN software stack as the principal compute platform for Pangu training.
• Huawei Cloud ModelArts. Cloud AI platform providing Pangu commercial access for enterprise customers.
• Multi-discipline research output. Published research output at NeurIPS, ICML, CVPR, ACL, AAAI, and other venues across machine learning, natural language processing, computer vision, decision-making AI, and other areas.

Distribution channels include direct integration into Huawei's product line, Huawei Cloud commercial distribution through ModelArts, the open-source MindSpore distribution, and published research output through major academic venues.

Benchmarks and standing

Huawei Noah's Ark Lab's evaluation framework combines foundation-model benchmarks (with Chinese-language and multilingual benchmark performance for the Pangu series), published research output, and commercial-deployment metrics through the Huawei Cloud ModelArts platform. The Pangu series has been characterized in Chinese AI industry coverage as one of the principal Chinese-language foundation model families, alongside Alibaba Qwen, DeepSeek, Z.AI, and other commercial Chinese foundation model families.

The Pangu-Weather scientific computing variant has been benchmarked against the European Centre for Medium-Range Weather Forecasts (ECMWF) numerical weather prediction systems, with published results in Nature. The scientific-AI positioning has continued to anchor Noah's Ark Lab's industrial-research credibility across application domains.

The Ascend AI compute platform has continued market-share growth in China through 2024 to 2026 under the supply-chain restrictions on NVIDIA H100 and H200 access. Chinese cloud and enterprise customers have transitioned compute infrastructure to Ascend; standardized comparison benchmarks against NVIDIA-based AI compute remain limited.

The published research output across major academic venues has continued through 2024 to 2026 alongside Chinese, European, and Canadian research-collaboration relationships.

Leadership

As of April 2026, Huawei Noah's Ark Lab's senior leadership includes:

• Hua Wu, Chief Scientist of Huawei Noah's Ark Lab. Huawei senior leader.
• Yiqun Liu, Director of Noah's Ark Lab.
• Tong Xiao, Director of the Edmonton lab.
• Senior research leadership across the principal research areas (machine learning, natural language processing, computer vision, decision-making AI) and across the lab locations (Hong Kong, Beijing, Shenzhen, Shanghai, London, Paris, Edmonton).

Departures and arrivals are continuous. Xian-Sheng Hua moved from Noah's Ark Lab to Alibaba DAMO; Zheng Zhang departed from the Hong Kong lab through 2020. The 2024 to 2026 period has continued senior research-talent recruitment and senior leadership transitions across the lab locations.

Funding and backers

Huawei Noah's Ark Lab operates within Huawei Technologies Co., Ltd.'s overall research-and-development budget. Huawei reported approximately CNY 180 billion (approximately $25 billion) in R&D spending in fiscal 2024, with Noah's Ark Lab representing a portion of that allocation on AI research. Huawei is privately held and does not publicly disclose Noah's Ark Lab's specific budget allocation.

The Huawei revenue base from telecommunications, consumer electronics, and cloud computing supports Noah's Ark Lab's research-funding commitment. Open questions on near-term resourcing are limited compared to standalone AI labs, given Huawei's revenue base and the Chinese government support for Huawei's continued operations under the US Department of Commerce restrictions.

Industry position

Huawei Noah's Ark Lab occupies a structurally distinctive position as one of the principal Chinese industrial AI research divisions, with integration into Huawei's broader product line spanning HarmonyOS, the Mate and Pura smartphone series, the Huawei Cloud commercial cloud business with ModelArts, the Ascend AI compute platform, and the MindSpore open-source deep-learning framework. The 2012 to 2026 research output and the Pangu foundation model series have continued to position Noah's Ark Lab as one of the principal Chinese industrial AI research divisions internationally.

Industry coverage has consistently characterized Noah's Ark Lab as one of the principal Chinese industrial AI research divisions, alongside Alibaba DAMO, Baidu, Tencent Hunyuan, and ByteDance Seed. The 2024 to 2026 period has continued the lab's research output and the vertical-integrated product positioning across the Huawei AI ecosystem.

Competitive landscape

• Alibaba Qwen / DAMO, Baidu, Tencent Hunyuan, ByteDance Seed. Chinese industrial AI research peers with overlap on Chinese-language foundation model research and commercial-cloud distribution.
• BAAI, Shanghai AI Laboratory. Chinese government-backed AI research peers.
• DeepSeek, Z.AI, Moonshot AI, Stepfun, MiniMax. Commercial Chinese AI startups with Chinese-language foundation model competition.
• NVIDIA Research. Industrial-research peer with overlap on AI accelerator silicon (NVIDIA GPUs vs. Huawei Ascend) and the underlying compute platform; commercial competition under the US Department of Commerce restrictions on NVIDIA access in China.
• Cerebras, Groq, Mediatek. AI accelerator silicon peers.
• Microsoft AI, Google DeepMind, Meta AI / FAIR, Apple Intelligence, Amazon AGI. Non-Chinese industrial AI research peers with overlap on multi-discipline industrial-research positioning.

Outlook

• The continued cadence of Pangu foundation model series releases through 2026 and 2027.
• The continued Ascend AI compute platform market-share growth in China under the supply-chain restrictions.
• The continued MindSpore open-source ecosystem expansion and the cross-institution research-cooperation.
• The competitive dynamic with the principal Chinese commercial AI startups (DeepSeek, Z.AI, Moonshot, Stepfun, MiniMax) on Chinese-language foundation model capability.
• The continued Huawei Cloud commercial expansion through ModelArts and the Pangu commercial-product integration.
• Continued senior research-talent recruitment and senior leadership stability through the 2026 to 2027 Chinese AI commercial expansion.

Sources

• Huawei Noah's Ark Lab official site. Lab reference.
• Huawei Cloud Pangu. Pangu foundation model commercial reference.
• MindSpore. Open-source deep-learning framework.
• Pangu-Weather Nature paper. Pangu-Weather scientific publication.
• Huawei Ascend AI compute platform. Ascend AI compute reference.