Can SenseTime Maintain Its Edge?
Discover the Latest in Asia’s AI Industry! Subscribe to our weekly newsletter The Abacus for exclusive insights and updates. Join us on this exciting journey!
Ⅰ
Today let’s talk about SenseTime, one of China’s leading AI startups (the rest three is Megvill, Yitu, iFlytek), best known for computer vision technology that can power facial recognition software, right now losing its technology advantage.
SenseTime has always been in the turmoil of US-China politics. It’s famous for making software that can recognize faces, something you might have heard in use across China, like in city streets helping keep things safe and orderly. Then, in 2021, the U.S. Treasury designated it as part of the “Chinese military-industrial complex companies,” a move that subsequently delayed its IPO.
Recently, the research firm Grizzly chose to short sell SenseTime’s stock. Citing social media screenshots and lawsuits, they accused SenseTime of engaging in what’s called ‘revenue round-tripping’, which means that SenseTime allegedly provided funds to clients to purchase its own products, yet failed to deliver these products, thereby artificially inflating its revenue.
Short-sellers like Grizzly Research place ‘short’ bets against companies, profiting when those companies’ share prices fall. It’s worth noting that back in September, Grizzly targeted Pinduoduo (PDD), the parent company of Temu, for short-selling. Defying Grizzly’s predictions, Pinduoduo’s stock didn’t just hold steady, it surged, particularly after its Q3 reports last week. PDD’s market cap reached $195.9 bn, overtaking Alibaba as China’s top e-commerce company in value.
SenseTime has denied the report twice, stating that it is “without merit and contains unfounded allegations, misleading conclusions, and interpretations.”
Nevertheless, this situation has thrust SenseTime into the spotlight once again. It raises a crucial question for every AI startup in China: Can they achieve a position of sustainable, long-term profitability?
II
Now, let’s be real. Take a peek at SenseTime’s financials — it’s clear the company is still in the red. Here’s the scoop from their mid-2023 report: they’ve accumulated a loss of $3.4 bn, while their revenue stood at only $1.4 bn. Additionally, their gross profit margin experienced a significant drop, plunging by 30.6% to just 45.3%.
For revenue streams, both their smart city and smart car divisions are facing reductions in growth. In a notable move in August, the company initiated its second round of layoffs within just 12 months. This reduction impacted about 10% to 15% of the workforce in the Smart City and Business Group (SCG). Ouch.
SenseTime primarily serves enterprise clients, each requiring tailor-made solutions. In the era before LLMs became the hot topic in AI investment circles, the approach was pretty straightforward — develop different models based on individual client needs.
Take a look at these figures: In 2019, 2020, and the first half of 2021, SenseTime produced 1,152, 9,673, and 8,377 AI models, respectively. However, this massive output didn’t exactly translate into equivalent commercial success. In many cases, a model developed might only be reusable for three to five clients. What’s more, from 2019 to 2020, while the number of models SenseTime produced skyrocketed by 740%, its revenue growth lagged far behind at just about 14%.
SenseTime’s government-related business, which also accounts for a significant portion of its overall revenue, serving clients like subways and public security, faces a major challenge: most of these deals are one-off, meaning they don’t generate repeat business. The same situation applies to the other three startups.
I need to mention, even though facial recognition is big in China, it hasn’t revolutionized the industry as companies like Hikvision and Dahua offer similar services.
As one story comes to an end, SenseTime is betting its next growth point on autonomous driving and large language models.
In 2022, the company’s Smart Auto unit brought in about $40.6 million in revenue, marking a 59% increase from the previous year and accounting for roughly 8% of its total revenue. Its products are now used in 27 car models from brands like NIO and BYD. According to Bloomberg, SenseTime recently is even considering setting up separate auto units to attract additional funding.
In April this year, SenseTime launched a ChatGPT-style chatbot called ‘SenseChat1.0,’ and by July, they had updated it to version 2.0. Based on this, they introduced several generative AI products like AI painting, which passed the first batch of regulatory approvals. However, in this highly competitive field, SenseTime hasn’t shown any significantly superior AI capabilities.
After multiple rounds of fundraising, the path for startups like SenseTime is clear: commercialize, grow revenue, and then go public. These companies are constantly driven by the need to increase their revenues to support their valuations. This pursuit can impact their ability to make technological breakthroughs.
For startups like SenseTime, the path to success seems simple: raise funds, build a product, and go public. But the reality is more nuanced. These companies face constant pressure to generate revenue and justify their valuations. This focus on profits can sometimes come at the expense of groundbreaking innovation.
This is an old question from the last wave of the AI startup boom in China: will investors have the patience for LLM startups this time around, and will these startups focus on consumers or enterprises? Both of these factors are crucial in determining whether China can produce the next OpenAI.
Industry
- AMD Set to unveil new AI chip to challenge NVIDIA’s dominance.
- ByteDance offers to buy back $5 billion worth of investor shares as IPO prospects remain slim. The company’s valuation has fallen to $268 billion, down about 10% from a year ago when it repurchased $300 billion of stock. (Zou Xin / SCMP)
- San Diego-based self-driving trucking company TuSimple plans to shutter its US business, laying off ~150, or 75% of its US staff, and move its business to China. (Heather Someriville / WSJ)
- Apple tells component suppliers of its preference to source iPhone 16 batteries from Indian factories, as part of plans to diversify away from China (Qianer Liu, John Reed / Financial Times)
- Infingence-AI, a Shanghai-based AI chips startup, has received investment from Baidu, Tencent, and Sequoia China. The company focuses on the research and development of large-model inference chips.
- Minuscule AI startup raises $41 million to tap growth in Indian market, the largest such round by an early-stage AI contender in the country. (Saritha Rai / Business Standard)
- Meitu releases self-developed AI vision model MiracleVision 4.0, focusing on AI design and video. MiracleVision 4.0 is expected to be launched in its products in January 2024.
- Chinese legal AI startup PowerLaw AI raised $10 million in Pre-B round led by Blue Lake Capital, aiming to change the legal services market for SMEs.（Liu Peilin / Caixin）Application-layer AI projects rarely receive single-round funding of so much money.
Policy
- China’s Minister of commerce meets with Microsoft’s president, the statement said China hopes Microsoft will play a constructive role in co-operation on AI between two countries.
- Shanghai names AI firm SenseTime and Baidu’s Xiaodu among city’s 40 tech champions. （Xu Huiyun / Yicai Global)
- Korea passed a revision to the Public Official Election Act calling for a ban on political campaign videos using AI-generated deepfakes during the election season on Tuesday. (Kang Jae-eun / Yonhap)
