今天的日经亚洲论坛Nikkei Asian Review 刊登Nina Xiang的丽文，说：

从依图到旷视，再到商汤科技，寒武纪，云知声，很多AI 技术公司已经上市或者正在排队上市。但是它们不是散户的菜。十年后可能誰也說不出它们的名字。

（1）巨亏不要紧，但是未来怎样赚钱必须清楚。它们有长期可持续的商业模式吗？

（2）高度依赖几大客户特别是政府。

（3）政府、银行、零售企业会把这些核心资料长期交给外部企业吗？

（4）没有护城河：海康威视和浙江大华科技等都能做。

（5）物流自动化和机器人送货可能有戏。

（6）招股说明书上的风险因素是真实的。

In their rush to go public, China's AI unicorns reveal a fatal flaw.

Investor should be wary, as risk factors listed in IPO prospectuses are real.

by Nina Xiang

December 1, 2020 05:00 JST.

Nina Xiang is the founder of China Money Network, a media platform tracking China's venture and tech sectors.

In 10 years no one will remember the names of China's artificial intelligence unicorns. While many aspects of the coming AI revolution remain unpredictable, one thing is clear: no AI company will emerge as a Big Tech brand.

While the internet era of the 2000s, and the mobile internet era of the 2010s, created the Chinese tech giants of today, such as Baidu, Alibaba Group Holding, and Tencent Holdings, collectively referred to as BAT, as well as Toutiao, Meituan-Dianping, Didi-Chuxing, together known as TMD, the AI era is unlikely to produce anything like that by comparison -- even if overly zealous investors have nursed over a dozen AI unicorns in China worth tens of billions in total.

That is partly because AI businesses are not consumer-facing. Because they are mostly providers of back end hardware and software to other businesses, or, more critically, to governments, AI business will not become giant platform companies servicing billions of users.

More importantly, as these highly valued startups rush to list publicly, their fatal flaw is laid bare: they simply do not -- and probably will not ever -- have a sustainable long-term business model. All those who have disclosed financials in an IPO prospectus, including Megvii, Shanghai Yitu Internet Technology, Cambricon Technologies and Beijing Unisound Information Technology, are deeply in the red to the scale of billions of yuan.

While AI businesses must invest heavily in research and are still in the early days of commercialization, suffering large losses is understandable. What is more concerning is their reliance on government contracts or a few big customers.

Take Megvii, where around 73% of its total revenue during the first half of 2019 came from city IoT solutions -- in other words, smart city initiatives paid for by government. For Yitu, the majority of its revenues also came from governments, with its top five customers accounting for 62% of total revenue in the first half of 2020. This type of concentration is probably true right across China's AI unicorn herd.

Whether AI companies can continue to gain ground in this to-government market is questionable. Compared to traditional surveillance camera makers and public security providers like Hikvision and Dahua Technology, AI unicorns like Megvii and SenseTime are not the ones with the deepest moats. The competitive edge enjoyed by most AI companies is also being eroded as the likes of Hikvision and Dahua improve their own AI algorithms to go with their rather comprehensive product suites.

Another core dilemma for China's AI unicorns is whether it makes sense for their big clients to outsource such a critical business function. Should a big bank contract out and hand over its most sensitive data to an outside AI company for a mission-critical initiative? Would it really be better for a big retailer to share its proprietary data with an industry outsider in exchange for algorithms that it could probably develop on its own?

In very few cases, the answer may be yes. But as the technological barrier to once inaccessible AI algorithms crumbles, the scale will tilt against AI companies, especially when the data is proprietary and in the hands of owners of traditional businesses.

This raises yet another challenge for AI unicorns. As the so-called intelligence "enablers" to various industries from financial services and health care, to retail and education, they need to develop expertise across vastly different sectors. If that is not hard enough, having to customize products to different industries makes it harder to achieve economies of scale. With all the AI unicorns seeking to expand in the same industry verticals, that will only further intensify competition.

This does not mean the coming AI revolution will not thoroughly transform our businesses and our lives, or that there will not be some sizable AI companies that will go on to become large and profitable businesses. Automated logistics and robotic delivery services are most likely to see such success stories.

But it also remains to be seen how big a role these AI unicorns -- born out of university labs amid a renewed wave of public fantasy following AlphaGo's landmark success in 2016 -- will play. Or, for how long -- if one believes such enabler roles to be transitional. What is more likely is that some of the vertical enabler AI companies will end up being acquired by their traditional business counterparts.

In the end, China's AI unicorns and their investors, clients, and end-users may all benefit from the third AI wave as AI makes our lives smarter. But at the same time, retail investors need to be extra careful buying these unicorns' shares, as all the risk factors listed in those IPO prospectus are very real.