Huawei’s big move from component maker to AI service provider
Huawei made a major AI announcement today with its Ascend 910 data center and cloud targeted chipsets, as well as a new soon to be open-sourced framework called MindSpore. No one should be surprised about this announcement. Huawei has been positioning itself to be an AI powerhouse for the past 2-3 years. Its previous products such as its Ascend Nano and Tiny chips and its AI-embedded Kirin chips have been mostly targeted at devices like smartphones and gadgets. But it’s been eyeing a more competitive position, in the market for data center and cloud-based processing, spurred by Chinese government’s goal of being a worldwide leader in AI.
How does this announcement position Huawei?
The Ascend 910 is a high-end, 300+ watt system on a chip that’s certainly not going to make it into the phone or smart appliance (its lower-end Ascend 310 targeted at lower power mostly inference solutions may). Rather, it’s targeted squarely at the AI training market, where companies like Nvidia have staked a major claim, where Intel is pushing on the boundaries of its x86 architectures (but mostly for inference rather than training), and where specialty TPU chips from Google, Facebook, Microsoft, Intel, etc., want to dominate. When asked about pricing for this chip, Huawei wasn’t committal but said it would be competitive with higher-end devices, like those from Nvidia, currently running as much as $10K each. That’s clearly a high-end system strategy and will have some major impact on Nvidia specifically.
But beyond just offering a new powerful processor, Huawei is looking to alter the market in frameworks by competing directly with popular development environments like Google’s TensorFlow, even while saying it would indeed support TensorFlow. This is a major gamble for Huawei, as it may be difficult to convince developers to switch. Still the market is just emerging and newer developers may not put up much resistance if MindSpore does offer improved functionality (Intel has a somewhat similar strategy with its OpenVINO platform).
No more just a component vendor
In the past, Huawei has been seen primarily as a component supplier rather than a full-scale service supplier. The release of the Ascend 910 and MindSpore signal a different direction for Huawei. The 910 will not be offered as a commodity chip to other companies; instead Huawei will use it to bolster its own data center services. Add MindSpore to the mix a framework optimized for Huawei’s own chips and it’s clear the company sees its future not as a component supplier but as a systems house that designs complete systems for autonomous vehicles, smart cities, healthcare, etc. and offers back-end services to make solutions available to the market (e.g., its collaboration with Audi on autonomous cars).
Huawei will push its customers to adopt the full suite of Huawei tech chips and framework in order to create a more compelling lock-in. And it has its sights set on a worldwide market (U.S. permitting) rather than just the substantial but limited Chinese market. Huawei wants to shift the question from who has the best processors for the task (Huawei, Intel, Nvidia, and custom designs from Google’s TPU, Facebook, Microsoft) to who can offer a full ecosystem of services. And it has the weight to be able to build a substantial ecosystem of partners to do so (no AI company, no matter how big, can go it alone in this market).
It’s about more than technology
Today’s announcement must also be seen in light of the current geopolitical situation. The U.S. embargo has forced Huawei, and indeed many other Chinese companies, to seek full self-reliance from U.S. tech suppliers. That includes capabilities and components it needs for smartphones, 5G network gear, and to support its enterprise and data center aspirations. Sanctions could cut off access to required chipsets from Intel and others, restrict ARM licensing it needs to build its own chipsets (there is an a ARM core in the Ascend), restrict common AI frameworks (although many are now open sourced), and cut off the supply of tools it needs to design and manufacture its own chips. In the past, Huawei and other Chinese companies would have taken the path of least resistance and acquired existing technology in parallel to their own work. They now see their strategic path forward as accelerating their own technology to cope (for example, Huawei’s back-up plan to use its own Harmony OS for smartphones if Android is closed off to it). And the Chinese government’s call for the country to lead in AI only reinforces that path.
Huawei’s new AI offerings are precursors to many more technologies we can expect to see from the company as well as from other Chinese behemoths such as Baidu, Tencent, and Alibaba, who now are large enough and have enough resources to substantially impact new technology markets. They’ve stepped up their moves to be more self-reliant as a result of the current political climate. And I expect to see an acceleration of announcements coming from China, especially in AI, where the marketplace is just forming and many are jockeying for position. It remains to be seen if Huawei and/or others can build the full ecosystem needed to be a major player, but it’s clear they are pursuing this space with determination.