Technology Author:Yingwei Fu Editor:Luke Sheehan Nov 20, 2020 06:56 PM (GMT+8)

Now it seems everyone has a crush on semiconductors. So does the US-blacklisted Yitu, but for a special reason.

Semiconductor piece

Image credit“Public access”

A brief founding history of Yitu

Shanghai-based artificial intelligence (AI) company Yitu Technology (Yitu) applied for an IPO on the Shanghai Stock Exchange Star Market (Star Market) in early November. As one of the AI pioneers, Yitu has been nominated as one of 'China's Four AI Dragons' – it is a typical first-gen AI company in China, one with similar traits to the other three 'Dragons': good at general AI algorithm writing and specialized in computer vision (CV), product-focused and with a high valuation.

Founded in 2012, Yitu started from CV technology and it is one of the leading AI medical image analysis providers in the world. Following its foundation, the company had completed eight financing rounds by June 2020, pushing its valuation to USD 3.5 billion. Its financial backers include, but are not limited to VCs like Sequoia China, Yunfeng Capital, Hillhouse Capital, ZhenFund and Gaorong Capital and state-controlled industry guidance funds Zhangjiang Torch Fund and Shanghai Science and Technology Investment.

yitu funding hist.png.png

The AI industry has taken a roller coaster ride through the 2010s: from emergence to barbarous development with inflated bubbles, the industry has by now experienced a cooling stage. Both investors and AI startup founders are more realistic than daydreamers in the early days. Yitu walked through all development stages and the manic AI investment heat inevitably influenced the unicorn’s growth. Like other big AI companies, Yitu faces the same question from the market: how can an AI cash burner make money?

Yitu's business landscape: Things are getting harder

With Yitu we find a script that reads like those of other unicorn startups, starting with founders with a polished educational background and solid industry experience. Yitu was founded in 2012 by Leo Zhu (朱珑), who holds a Ph.D. in statistics from UCLA and worked as a fellow of MIT AI lab, and Chenxi Lin (林晨曦), who graduated from Shanghai Jiao Tong University and successively served Microsoft Research Asia and Alibaba Cloud. The two digital devotees teamed up and established Yitu in Shanghai.

Yitu's medical image analysis solutions took on new fame as coronavirus swiftly swept over China. Yitu joined the battle against COVID-19 by providing diagnosis-assistive solutions for radiologists. Beyond the medical image analysis, Yitu applies CV and other technologies to sectors like 'smart city,' 'smart community' and 'smart retail,' which are the main battlefields for AI firms. AI software was once the largest income source of Yitu and in 2017 software product sales were accounted for 55.9% of the total revenue.

The company started to transform when a growing number of AI companies clustered in fields like 'smart city' and 'smart retail.' As AI companies are typically categorized under the software sector, Yitu cannot be restrained by that definition with the ongoing transformation.

In the early days, Yitu provided CV products like facial recognition kits to customers. Later the company began to supply integrative solutions, which include implementation plans and software-preinstalled hardware sets. Scenario-based solutions drive revenue growth, but the company has shifted its focus from the application top to a more fundamental layer that is critical to AI developers: computing power. 'Computing power' (in Chinese "算力") is mentioned times in Yitu's prospectus, hinting the significance to the company.

yitu chip.png.png

Yitu released its original AI chip, QuestCore, in 2019. The chip itself does not generate any direct income for the company, but it is widely used on Yitu's products and services. QuestCore is a chip designed to process massive images and perform inference to enhance computing power. With the AI chip launch, Yitu's revenue structure soon changed. The once-dominant software business was squeezed while the integrated solutions (software and hardware combined) saw an increasing weight.

From serving terminal users of AI solutions to serving both terminal users and AI developers, Yitu defines itself as a computing power booster via providing QuestCore-supported solutions, which deviates from the first impression of Yitu – an AI software maker. Now, it is moving to a more fundamental layer of the industry.

The financials

Inside its prospectus, Yitu disclosed a shockingly huge uncovered loss of CNY 7.22 billion (USD 1.07 billion), which is caused by the recognition of preferred stocks. After rounds of financing, the inflated valuation is reflected in the fair value of preferred stocks: it led to a loss of CNY 982.74 million, 545.23 million, 2,618.97 million and 936.15 million in 2017, 2018, 2019 and the first half of 2020. For startups, a huge uncovered loss is expected when valuation swells.

Other than the inflated valuation, Yitu is indeed not making a profit. As mentioned previously, the company's income relies on integrated solutions rather than software solely. In general, matured software companies' gross margins fluctuate around 80%. Yitu's overall gross margin increased from 2017's 57.4% to 2019's 70.99%. Meanwhile, software sales decreased from 56% to 15% and integrated solutions sales revealed a strong growth when Yitu started to position itself as a computing power provider.

gross comparison.png.png

The change in business focus differentiates Yitu from Megvii, which heavily relies on IoT solutions. EqualOcean previously analyzed Megvii's business and found that the company was sandwiched by its business partners – camera/device sellers and solution buyers, who have strong bargaining power. At present, Yitu has the same problem as Megvii, as they both provide city-level IoT solutions, but with new positioning, the situation will alter with Yitu's new position and strategy.

The cost of sales grew from 2017's CNY 29.3 million to 2019's 258.8 million, which increased proportionally with revenue. It is worth mention that the cost-share on hardware expanded from 28.0% to 67.7% within the same period. Behind the hardware expenses, Yitu sold more integrated solutions (combo of software, devices and solutions): over 63.8% sales were generated from the combination of software and hardware (can be regarded as 'solution') in 2019, three times the percentage in 2017.

yitu_expenses_ration.png.png

From 42% to 55% in three and a half years, more expenses were spent on research and development. Yitu had over 837 R&D employees by the end of June 2020, accounting for 55.5% of total employees. The company attributed the reason for decreasing sales expenses to the pandemic and decreased marketing events.

However, compared with competitors, Yitu's sales expenses to revenue ratio of 41.7% spikes out. The AI chip maker Cambricon’s sales expenses to revenue in the first half of the year was 15.4%; IoT solution provider Megvii's sales expenses to revenue in the first half of 2019 was 15.1%; HikVision's sales expenses to revenue was 10.2% in its 3Q2020 report. Though with a decreasing expenses ratio, 41.7% is a red light for all investors.

Computing power is key to Yitu's future

Yitu admitted that it is not the dominant player in any field but faces intense competition from all directions. It is not the best AI chip maker nor the city-level IoT server with the largest market share in China. At the moment, Yitu's key clients are government and state-owned enterprises. But the landscape is expected to change when it eyes on computing power.

Embedded with self-designed chip QuestCore, Yitu's server is committed to providing strong computing power for AI training and inference with efficiency and lowered costs, especially the image processing capability. Competing with existed AI chip makers, Yitu's strengths and weaknesses are also obvious.

As an AI company with business in city-level IoT and medical image assistive diagnosis areas, Yitu is a user of its own Cloud computing equipment. From a product user to a product manager, the company understands AI training and inference requirements on equipment and environment. On the one hand, with a deep understanding of AI technology and as a pioneer in the area, Yitu is capable to be an infrastructural service provider for other AI companies – an AI firm that serves AI firms with computing power. On the other hand, as a latecomer in AI chip sector, Yitu still has multiple fierce competitions ahead: fighting for acquiring AI chip talents and the market.

AIsemiconductorEdge computingComputing powerAI chipYitu TechnologyMegvii