Huawei’s Ascend 910 Chip and MindSpore Framework Underscore its Ability to Forge Ahead with AI, Despite Escalating Tensions with the U.S.

R. Bhattacharyya

Summary Bullets:

• Huawei announced the commercial launch of the Ascend 910, an AI chip optimized for model training, as well as MindSpore, a training and inference framework.

• The announcements come during escalating tensions between the U.S. and China, underscoring that Huawei will forge ahead in emerging technologies, with or without the U.S.

In the realm of AI, Huawei is readying itself to go head to head with global leaders, including Google and Amazon. The company’s recent commercial launches of an AI chip optimized for model training, as well as a training and inference framework, come hot on the heels of the unveiling of its internally developed operating system for mobile devices. The timing of the announcements is opportune; they come during a period of escalating tensions between the U.S. and China, serving to underscore that Huawei will forge ahead in emerging technologies, with or without the U.S.

Announced last October, but only now available commercially, Huawei’s new Ascend 910 chip delivers 256 TeraFLOPS for half-precision floating point operations and 512 TeraOPS for integer precision calculations, and is designed for deployment in data centers. Its newly available MindSpore AI framework, designed to simplify model development and improve resource allocation, works for deployments in the cloud, at the edge, and on devices. Huawei notes that the two biggest challenges companies face when adopting AI are a cumbersome development process and the lack of processing power. With its recent product launches, Huawei is looking to address exactly these concerns.

But Huawei isn’t the only company racing to develop its own chip technology to enable faster and cheaper processing of AI models. Although U.S. tech vendors such as Qualcomm and Nvidia have largely dominated the chip space, other players are eager to make their mark on the lucrative AI processing market. Google has developed its own TPU, with the TPU v3 offering 420 teraflops, and TPU pods further accelerating processing speeds to over 100 petaflops (though still in beta). Amazon has announced that it has its own chip in the works, the AWS Inferentia, offering hundreds of tera operations per second (due out later this year). And Huawei’s compatriots are not to be left out: Baidu offers the Kunlun 818-300 for training and the Kunlun 808-100 for inference, and Alibaba launched its first chip last month, with plans to launch an AI optimized chip, the AliNPU in the coming months.

However, what makes Huawei unique is its desire to position itself as a one-stop shop for all things AI. Its vision is to offer a full stack of AI-focused solutions that spans chipsets, hardware, a developer platform, and pre-packaged solutions. The portfolio is designed to be ‘all-scenario’, meaning it enables a range of processing options, including in the cloud, on-premises, and at the edge.

Nonetheless, the company is in the midst of a nasty trade dispute with the U.S. Huawei and its affiliates are on the Entity List, which restricts U.S. companies from doing business with them. Huawei was recently granted a 90 day reprieve, but there is no way of knowing what is to come. The impact on its business, particularly in the area of mobile devices, will be severe. But as evidenced by recent actions, Huawei is preparing itself by putting in place a strategy that centers on greater self-reliance.

And despite recent trade disputes, Huawei still has a lot going in its favor. It operates in a country with huge government support and investment in AI technology. Its citizens don’t harbor the privacy concerns of many western countries, allowing for the adoption of a greater variety of AI use cases, all the while helping companies gain experience with AI. And the country is home to a multitude of AI startups, facilitating a vibrant ecosystem of partners, competitors, academics, and customers – all in need of AI processing and tools.


What do you think?

Please log in using one of these methods to post your comment: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

This site uses Akismet to reduce spam. Learn how your comment data is processed.