China AI: A rude awakening

SenseTime, China’s largest artificial intelligence (AI) technology developer, has until recently breezed through funding rounds. A $600 million Series C led by Alibaba Group in April 2018 was followed two months later by a $650 million extension. Its valuation moved from $4.5 billion to $6 billion when SoftBank Vision Fund committed $1 billion later in the year.

Since then, nothing. This period of inactivity coincided with mounting concerns about the valuations achieved by unprofitable Chinese start-ups. Mubadala Investment reportedly pulled away from a joint round with Vision Fund for that reason. As of August, SenseTime was seeking a pre-IPO round – having reduced its target valuation from $10 billion to $8.5 billion – ahead of a dual listing in Hong Kong and Shanghai, according to Bloomberg.

Even with the prospect of a near-term liquidity event, some investors are reluctant to offer their support. “SenseTime’s business is basically three parts – software, hardware and customized projects. These projects are the major part but they are unstandardized and hard to scale up. The business model doesn’t justify the valuation,” one investor tells AVCJ.

SenseTime’s big picture vision is to "empower all industries with AI," using a proprietary deep learning platform that can be used in fields such as facial and image recognition, autonomous driving, and medical imaging. But its operating model is asset-light. The company establishes joint ventures with industry leaders, contributing intellectual property – algorithms – and relying on its partners to handle the technology landing.

These JVs include SenseNets, which was formed in 2015 with Shenzhen-listed NetPosa, one of the top 10 manufacturers of surveillance cameras. It was a logical tie-up intended to leverage SenseTime’s expertise in facial-recognition AI, but it went nowhere. The JV’s official WeChat account went silent in November 2018. Last year, it emerged that SenseNets had allowed personal information on 2.5 million people to leak into the public domain.

“Facial recognition in security is not only about the algorithm. It’s about the whole system, the integration of software and hardware. Running a reliable service is much more complicated than it might appear to be,” says Yang Gao, an assistant professor at Tsinghua University who specializes in computer vision and robotics.

Winners and losers

SenseTime still has supporters. They claim the company’s technology is far superior to that of incumbents in the surveillance camera space, such as Hikvision Digital Technology and Dahua Technology, and it will eat up their market share. However, those two companies are still growing while many other industry participants are sinking.

Hikvision generated revenue of RMB57.6 billion ($8.5 billion) and net profit of RMB12.4 billion in 2019, year-on-year gains of 15.7% and 9.4%, respectively. Duha’s revenue rose 10.5% to RMB26.1 billion its net profit was up 26% at RMB3.18 billion. These strong numbers continued through the first half of 2020.

SenseTime’s collaborator NetPosa, on the other hand, has been struggling. “The disposable income of local governments has decreased, security projects have been squeezed, the pace of government payments has slowed significantly, and the payment capabilities of downstream integrators have been heavily affected,” the company said in its annual report.

This is a well-known risk of doing business with local governments. In the three years ended 2018, NetPosa reported net profit of RMB353 million, RMB380 million and RMB314 million, respectively. Performance was buoyed by an increasing number of government contracts, with accounts receivable soaring to RMB2.78 billion in 2018, or 120% of total revenue.

Weak net cashflows in 2016-2018 – including a paltry RMB10 million in 2018 – indicate just how few of these contracts were being honored. When NetPosa finally admitted the extent of its bad debts, its balance sheet bore the scars. Revenue came to RMB353 million in 2019, down 84% year-on-year, and the company posted a net loss of RMB3.1 billion. In the first six months, revenue was down 81% at RMB66 million and the net loss was RMB274 million.

For companies looking to raise further private funding or positioning for an IPO, revenue growth can appease investors wary of mounting losses.

Megvii Technology, a direct competitor of SenseTime, built its business by supplying facial recognition software for smart phones. This accounted for 73% of the company’s RMB67.8 million in revenue for 2016. Two years later, the smart phone business had grown more than fivefold but its revenue share was 19%. City internet-of-things (IoT) solutions – chiefly security cameras – was responsible for 74% of Megvii’s RMB1.4 billion in revenue.

Megvii's net loss rose from RMB758.8 million in 2017 to RMB3.35 billion in 2018. For the six months ended in June 2019, net losses came to RMB5.2 billion.

The company received four rounds of private funding from the likes of Abu Dhabi Investment Authority (ADIA), China State-owned Venture Capital Fund, Ant Group, Foxconn Technology, BHR Partners, Qiming Venture Partners, and Sinovation Ventures. It achieved a valuation of $4 billion in May 2019 and filed to list in Hong Kong three months later.

The listing application expired in February, with Megvii yet to shake off the reputational damage caused by accusations from the US government of complicity in human rights abuses in China. Megvii and SenseTime were among eight Chinese AI start-ups blacklisted, which means they cannot source components from American suppliers.

Reasons to pass

These ethical concerns – the start-ups are said to have provided technology used in China’s repression, mass arbitrary detention, and surveillance against Uighurs, Kazakhs, and other members of Muslim minority groups – is a turn-off for many investors. “We are concerned about the security business because the data is sensitive. Our LPs prefer to keep a distance from such activities," says one investor.

However, the primary issue appears to be business model sustainability and a clear path to profitability. SenseTime must move quickly on this if its IPO ambitions are to be realized. Last year, the company underwent a round of internal restructuring that led to the creation of several new business lines, focusing on AI applications in healthcare, education, and semiconductors.

It makes strategic sense, but there are well-funded incumbents in each of these areas. For example, Wen Zhang, SenseTime’s former president, resigned only to reemerge last year as founder of Biren Technology. The AI chip manufacturer has already completed two rounds of funding, bagging RMB2 billion from GL Ventures, Qiming, IDG Capital, and others.

The global AI market is expected to expand rapidly. Grand View Research valued the industry at $39.9 billion last year and projected compound annual growth of 42.2% through 2027. Within China, rising demand for smart personal devices, city IoT solutions, and smart logistics services will likely filter through to AI start-ups.

But the numbers, regardless of the ambitions that underpin them, must add up. “In the end, AI is software and it has to be evaluated as software,” says another investor. “There can’t be some mysterious premium.”

• Tweet  
• Facebook  
• LinkedIn  
• Google plus  
• Save this article  
• Send to  

• Topics
• Greater China
• Technology
• Expansion
• artificial intelligence
• China
• Alibaba Group
• Softbank