Baidu and Huawei Joining Forces in AI

This year at its annual Artificial Intelligence (AI) developer conference, Baidu Create 2019, Baidu announced a partnership with Huawei. Baidu’s open-source deep learning framework, PaddlePaddle, will support Huawei’s smartphone focused AI chipset, the Kirin series. This is perhaps the first time in the industry when two Chinese cloud AI giants made public announcements regarding chipset-related collaboration.

Targeting high- and mid-tier smartphones, the Kirin series System-on-Chip (SoC), namely Kirin 970 and 980, features its proprietary Neural Processing Unit (NPU). Found on Huawei’s premium and mid-tier phones, such as P30 Pro, Mate 20, and Honor 20, the NPU has built-in support for TensorFlow Lite and Caffe2 models. Now, it will also support PaddlePaddle.

On the other hand, Baidu has been investing in AI for quite some time. At Baidu, PaddlePaddle has been deployed in existing products and services, such as ad Click-Through Rate (CTR) prediction, large-scale image classification, Optical Character Recognition (OCR), search ranking, computer virus detection, and search recommendation. However, PaddlePaddle has not gained the same popularity as Google’s TensorFlow and Facebook’s PyTorch and Caffe2. The community around PaddlePaddle is still very thin at the moment.

The evolution of the Kirin series itself is revelatory of the development of AI capabilities in China. HiSilicon, Huawei’s chipset arm, worked with Cambricon Technology, a Chinese Intellectual Property (IP) core licensing vendor to develop NPU in Kirin 970, utilizing TSMC 10 nm chipset manufacturing technology. When Kirin 980 was launched in August 2018, Huawei claimed it was the world’s first AI SoC based on 7 nm chipset manufacturing technology, leapfrogging major consumer AI chipset vendors such as Qualcomm, MediaTek, Samsung, and Apple. ABI Research estimated that the market share of HiSilicon, or Huawei, in the global edge AI inference chipset market was around 25% in 2018.

Similar to Huawei’s surge in the Chinese smartphone market, Baidu has further expanded its influence in the Chinese smart home industry. Baidu claims that it is currently the largest voice control front-end device vendor in China, with an installed base of over 600 million. According to ABI Research’s Smart Home Systems (MD-HAS-111) Market Data, over 5 million voice control front-end devices were shipped in Asia Pacific alone in 2018, with China representing a significant share of that figure.

In addition, PaddlePaddle remains a very versatile framework, as it is capable of serving many different deep learning-based applications, with high scalability and performance. Baidu makes the right choice joining Microsoft’s Open Neural Network Exchange (ONNX) in the effort to expand its interoperability with other frameworks. Since PaddlePaddle is now supported by Huawei’s large installed base of premium and mid-tier smartphones, AI developers will be more motivated than before to develop their applications using the PaddePaddle framework.

For more information on various AI frameworks, please refer to ABI Research’s Benchmarking AI Frameworks: Productization, Market Rationalization, and Ecosystem Development (PT-2122) Technology Analysis Report.

Overall, engaging in this partnership is a good decision for both Huawei and Baidu. ABI Research strongly believes that this collaboration is going to play a key role in the future of the AI industry in China for two main reasons.

Firstly, the trade dispute between China and the United States has resulted in supply chain disruption for Chinese technology companies, particularly Huawei. Restriction on the access of critical technology means Huawei does not only lose access to hardware components, operating systems and middleware, but also potentially non-Chinese open-source innovations. By partnering with Chinese cloud AI companies, like Baidu, Alibaba and Tencent, Huawei can ensure it has a fail-safe strategy in place. While the restrictions on Huawei has eased in recent weeks, the threat still remains. In an ABI Insight published earlier this year, The American AI Initiative Highlights the Need for AI Governance on a Global Scale (IN-5408), ABI Research argues that until the industry is able to release a global AI governance framework that is acceptable and enforceable by all AI players, end users and regulators, it will be very challenging for all parties to evaluate all AI technology providers impartially. It will not be a surprising development if more and more Chinese cloud AI vendors are partnering or joining forces with local AI chipset and hardware companies.

Secondly, the partnership exploits the natural synergy between the two companies. Huawei is and will always be a captive vendor. While the company has a full stack AI solution, from the Kirin and Ascend AI SoC series to its proprietary machine learning framework MindSpore, they remain in-house technology and are not being sold to external clients. On the other hand, Baidu has global ambition and is keen to challenge the likes of Google and Microsoft. Baidu’s own chipset achievement has been lackluster as compared to those of Huawei and Google, but the company still has plenty of good AI technologies, such as PaddlePaddle and Apollo, its open-source autonomous driving initiative. The impact of the collaboration will probably be more significant to the Chinese market in the beginning, but Huawei’s global footprint will propel Baidu’s deep learning capabilities onto the global stage.