- Though unacknowledged, Huawei responded to growing global criticisms at its annual Connect Conference in Shanghai by introducing several new products in support of its full-stack, all-scenario AI portfolio.
- With several new solutions spanning AI training and algorithm execution hardware, cloud services, open source projects, and ecosystem investments, Huawei intends to build a vibrant, sizable, and influential ecosystem of partners.
Huawei may be facing a global and escalating chorus of scrutiny, criticism, and outright censure right now over whether or not enterprise buyers should trust the Chinese technology giant. But, here in Shanghai, China at Huawei’s Connect 2019 conference, the skies are blue, the temperature is temperate, and the trees appear ready to don their glorious autumnal colors at any moment. This was the sentiment – only slightly paraphrased – delivered as a response to these challenges by Huawei’s Deputy Chairman and Rotating CEO, Ken Hu, during the opening keynote on Wednesday.
Is the company in denial, or is it trying to ‘whistle past the graveyard’ in an attempt to at least appear unafraid? Perhaps, yes. But I’d like to posit an opposing view. Huawei, like China, understands the cyclical expanse that is human history. Perhaps, then, the company is taking the long view with regards to success and failure (either perceived or measured). I think that Huawei fully comprehends the difficulties it faces and is architecting its portfolio in such a way that, indeed, one day in the future international politics won’t matter.
This appears to be the case at least for its computing business; it was very much on display at this year’s conference, which was almost wholly centered on artificial intelligence (AI) capabilities delivered across its Ascend and Kunpeng processor hardware. The idea begins with the company’s business strategy for AI as described by Mr. Hu. Huawei intends to create open hardware and invest in open source software, creating from those a full-stack all-scenario AI solutions portfolio. Services built on this stack will be delivered via Huawei Cloud. Hardware running this stack will of course power Huawei Cloud, but will also be sold directly to the company’s partners. These partners will then build the applications (e.g., the actual business outcomes) that show up in the market.
Huawei, then, sees itself as a pure horizontal platform player for AI, one that will depend on a vibrant ecosystem of ISVs and hardware manufacturers that will purportedly seek to work with Huawei, owing to its hardware engineering expertise (chips, servers, and chassis), comprehensive software stack (AI frameworks, databases, etc.), and economies of scale (Huawei Cloud). A key here is that Huawei will not sell its processors directly. Rather, the vendor will offer those to customers via cloud services; it will, however, sell those to its partners as components with the promise of equipping those partners with fully integrated solutions.
Therefore, presuming Huawei can build its intended vibrant, sizable, and influential ecosystem of partners, one that extends beyond the boundaries of the APAC region, perhaps the vendor will be able to weather its current travails and rebuild its global position as a trusted partner, at least for AI. If enough globally influential partners build on top of Huawei’s AI hardware, software, and services, perhaps it won’t matter as much that the vendor has been locked out of numerous key markets (mobility, security, etc.). Time will certainly tell. But if this year’s Connect conference is any measure of future success, the company is well on its way. Here are a few key announcements from the show to illustrate.
- Huawei is investing heavily in its developer program, pledging to spend $1.5 billion dollars over the next five years, focusing on training, educational support, certifications, joint marketing, innovation funding, and co-development.
- The company introduced the Atlas 900, an AI training cluster made up of thousands of Ascend chips that takes aim squarely at very demanding scientific applications. Huawei contends that this hardware cluster (which will also be sold via Huawei Cloud) has beaten the ResNet-50 AI training performance standard by a full ten seconds.
- Other new hardware on display included the Atlas 800 AI training server and the Atlas 300 AI training card. The latter is purported to use Huawei’s MindSpore to nearly double the number of images processed per second compared to GPUs running TensorFlow.
- Huawei announced the availability of 43 Huawei Cloud services that are now powered by its Ascend 910 and Ascend 310 hardware. These specialize in AI inference, data training, image processing, knowledge extraction, and autonomous driving. Owing to economies of scale, Huawei is promising significantly lower prices for these cloud services, some as much as 70%.
- Similarly, the company announced that 69 of its Huawei Cloud services were now running atop its Kunpeng processors. Here too, the idea is to deliver economies of scale to Huawei’s partners.
- And lastly, Huawei announced several open source software (OSS) efforts spanning database (Huawei GaussDB) and operating system (Huawei EulerOS). Already under test with several customers, these products will be fully released to the broader developer community early in 2020.