From SenseTime's 2021 annual report, we can see how production efficiency drives AI inclusion

Nearly four months ago, SenseTime completed its listing on December 30, 2021, becoming the world's largest IPO in the AI field.

The day before yesterday, SenseTime released its first annual report since its listing, which naturally attracted many eyes.

There are many business highlights in 2021:

Revenue accelerated, up 36.4% year-on-year to $4.7 billion

Gross margin of 69.7%

According to the market report of IDC and Sullivan, the Group has ranked first in the market share of computer vision applications in China for three consecutive years

When SenseTime's financial data was taken as the core information of major headlines, AI Nuggets found another noteworthy number in the annual report: the production efficiency of AI models.

In 2021, the number of AI model production will increase by 152%, and the efficiency of R&D human efficiency will increase by 72%.

What does this mean?

R&D and commercialization show a positive circular effect, and the AI universal benefit we expect has begun to appear.

AI production efficiency in the underlying logic of AI inclusion

Why is the improvement of AI production efficiency crucial to the accelerated landing of AI universal benefits?

The outbreak of the intelligent era is to make AI widely and large-scale applications, so that all walks of life can truly benefit from intelligent and digital transformation.

However, 80% of the intelligent needs of thousands of industries are long-tail scenarios, which are wide in scope, large in difference, small in volume, numerous in number, and diversified and fragmented in demand.

It is not easy to truly scale up massive fragmentation scenarios.

If the development process is cumbersome, the research and development process is difficult to reuse, and each project must be repeated to re-invest human and financial resources for development, so that the cost of comprehensive digital coverage remains high.

The essential reason behind this is that at this stage, AI is difficult to standardize and has low versatility.

For example, in the field of transportation, thousands of algorithms may be born in the future; for example, in the field of underground coal mines, there are many application scenarios, fine demand, and many production lines, and the whole line is very differentiated, and the production of a single AI algorithm will take dozens of engineers for months.

In other words, whoever can solve fragmentation to the greatest extent, improve production efficiency, and reduce marginal costs will get the key to open the door to AI scale.

If you want to build a tall building, you must first build a foundation, and AI infrastructure plays a vital role in the transformation of the real economy.

AI industrialization requires the underlying infrastructure, and only by solving the problems of full-link lifecycle management such as model development, deployment, management, and prediction can AI become digital productivity.

Only by continuing to focus on the construction of the underlying capabilities of artificial intelligence, generating innovative applications for scenarios in a systematic and industrialized way, and rapidly improving the efficiency of algorithm production can we break the boundaries of technology landing, and fundamentally realize the industrial production of AI models through the improvement of the underlying capabilities of the full stack, improve efficiency, and empower the industry on a large scale.

However, the heavy assets, talents, scientific research and risk characteristics of the AI underlying infrastructure make the threshold for the construction of the AI underlying infrastructure high.

The underlying hardware, training framework, algorithm model, and general technology all require continuous, high-intensity underlying investment in order to establish a weapon of technical cost and efficiency.

High technology, capital barriers, so that it has become a battlefield between AI giants, with this strength of the enterprise is not many, SenseTime is one.

This is exactly what Shang Tang has been doing since its birth.

SenseTime's AI device aims to make ai algorithm production leap from labor-intensive to industrial mass production.

The AI device is composed of three levels: computing power, platform and algorithm.

The computing power layer is based on Asia's largest AI supercomputing center (AIDC), integrating AI chips and AI sensors. SenseCore is the foundation of the AI business, and AIDC is the base of a large device. AIDC can provide large-scale elastic computing power to achieve complete training of 1,000 billion parametric models.

The platform layer is the link between the algorithm and the computing power layer, integrating the full-link AI development and batch application process of data preparation to model production, testing and deployment, and connecting the data platform, deep learning training framework, deep learning reasoning deployment engine, and model production platform.

Algorithm layer, algorithm toolbox produced in the model more than 22,000, open source framework OpenMMLab on GitHub has more than 37,000 stars, the highest in Asia.

How does the AI large device improve the efficiency of AI production?

By abstracting the algorithm models of different scenarios at the bottom level, thousands of modular suites are formed, and the general large model plus the subdivision optimization of a small sample single scene is not counted as a combined algorithm module, and the long-tail scene requirements are met more quickly and in batches.

They fundamentally improve the efficiency of AI research and development, reduce costs, let AI models get rid of manual production, gradually go up to the assembly line, and greatly reduce the cost of algorithm model production through large-scale production of AI algorithms, so as to achieve large-scale coverage of new scenarios at low marginal costs.

How does SenseTime do it?

As a leader in AI companies, innovation and R&D are the core weapons of SenseTime since its establishment.

60% of sense's fundraising at the time of listing is for research and development. Among them, most of them are invested in AI infrastructure SenseCore, 10% is used to expand the COMPUTING power of AIDC, 10% is used to strengthen the design capabilities of ARTIFICIAL intelligence chips and develop their own artificial intelligence chip solutions, 15% is used to improve the capabilities related to artificial intelligence models, and 25% will be invested in technical models, products and industry-university-research.

Scientific and technological research and development is the basis of product update iteration, technology-driven enterprises, innovative technology does not need to invest in research and development for many years, in order to maintain the leading position in the industry.

Its annual report also illustrates this point, in 2021, SenseTime will continue to maintain the industry-leading level of R & D investment, with annual R&D expenditure of 3.06 billion yuan, accounting for 65.1% of revenue. According to the financial report, SenseTime will increase its R&D personnel by 1,500 to a total of 4,200 in 2021, accounting for 70% of the total number of employees.

As of the end of 2021, SenseTime has a total of 11,494 global patent assets, an increase of 96% compared with the end of 2020, of which 78% are invention patents.

R&D investment accounts for more than 60% of revenue, and not all enterprises are persistent in adhering to innovative R&D.

Today's research and development is tomorrow's value.

R&D investment brings the support of business competitiveness and growth momentum. Continued R&D investment has brought great returns to SenseTime in both technical and commercial areas.

In fact, zooming in on the space-time dimension will find that AI universal benefits are not carried out by leaps and bounds, and the essence of innovation is technological evolution, step by step, step by step, or even half a step and a half step forward.

The first three industrial revolutions have spiraled up in optimization, correction and iteration, reducing production costs and improving production efficiency, and finally presenting subversive results.

If we look backwards from the final peak, we will find that SenseTime has always followed the essence of AI and industrial digitalization.

To achieve AI universal benefits, large-scale application of AI is needed, to achieve scale, it is necessary to solve the application of long-tail scenes, to solve the long-tail scene cycle is long and costly, it is necessary to improve production efficiency and reduce production costs.

From this perspective, SenseTime has had a long-term vision from the beginning, step by step to build the AI inclusive world in its heart.

Ai large devices are making AI light waves cover wider and wider through actual achievements, and gradually reach the critical point of scale.

It is reported that by the end of 2021, the final force of the 23 supercomputing clusters put into use is 1.17 billion billion floating-point calculations per second (1.17exaflops), the number of commercial models produced by SenseCore has reached more than 34,000, and the production efficiency of AI models has increased by 152% compared with the end of 2020.

At the same time, SenseTime's R&D personnel efficiency is improving year by year, and the average number of commercial models produced by R&D personnel per person per year in 2021 has increased by 72% and 13 times compared with 2020 and 2019, respectively.

152% of the model productivity, 72% of the efficiency of the improvement.

This number is quite beautiful even if you look at global AI companies.

Compared with the overall level of the industry, SenseTime's R&D efficiency is an order of magnitude higher, and correspondingly, in the context of the gradual maturity of AI in the second half, SenseTime's R&D cost is one order of magnitude lower than that of the industry.

SenseTime's gross profit margin continued to increase from 56.5% in 2018 to 69.7% in 2021, and the stable and high gross profit margin benefited from the scale effect of AI large devices on R&D efficiency.

Together, these constitute the foundation for SenseTime's LARGE-scale AI and ultimately the realization of AI universal benefits.

In the long run, lower model production costs and greater model production capacity allow SenseTime to empower the real economy more inclusively and more deeply, maintaining long-term commercial competitiveness and growth momentum.

Based on SenseTime's AI device as the core base, SenseTime provides generalized AIaaS (AI-as-a-Service) capabilities, which currently covers the application of AI in various vertical scenarios.

In 2022, SenseTime will further open SenseCore's capabilities to the market.

AIDC will open to the outside world to provide AI-as-a-Service intelligent computing services for industry, scientific research and urban management, comprehensively improve the productivity of AI, and accelerate the digital transformation of the entire industry.

Comprehensively solve the demand problem of long-tail applications, and achieve high-efficiency, low-cost and large-scale AI innovation and empowerment by significantly reducing the cost of artificial intelligence production factors.

Four major plates, open the door to AI universal benefits

On top of SenseCore's artificial intelligence infrastructure, SenseTime has developed in a balanced manner in the four major sectors of enterprises, cities, life, and automobiles, and continuously improved the closed-loop efficiency from data, AI model production, scene application, and commercialization.

From the four major plates of SenseTime, the shadow of AI universal benefit has been first seen.

Smart cities

The advancement of AI and other technologies has brought urban governance construction into a new stage, massive demand for fragmentation, garbage overflow, shared bicycles piled up, road damage, traffic accidents, fires and smoke, emergency exit obstacles... At the same time, facing a series of problems such as dense buildings, dense population, and traffic congestion, the efficiency of manpower supervision in the city is limited, the intensity of grass-roots work is large, and the quality of governance is difficult to meet expectations.

Urban governance urgently needs digital transformation to further expand the market for urban digitalization.

SenseTime is facing challenges and seizing opportunities. Smart city revenue increased by 56.6% year-on-year.

SenseTime City Ark, as the operating system of digital city operation, can not only manage public facilities such as fire hydrants, manhole covers, telephone poles and road signs, but also timely identify the timely detection and resolution of the above public events, as well as the impact analysis of natural disasters such as floods and typhoons and the tracking of the progress of follow-up disaster relief measures.

SenseTime's Ark City Open Platform contains more than 14,000 AI models, combined with the city's IT infrastructure, transforming raw urban data into operational insights, event alerts and management actions in real time.

SenseTime sinks to the most basic level, closely integrates with application scenarios, goes deep into the urban capillaries, and in the digital world created by SenseTime, forms a closed loop of intelligent governance such as urban safety hazards, urban sanitation, water channels, fire fighting facilities, etc., and online and offline integration makes the city truly become an intelligent body.

The transformation of urban management from manpower-intensive to human-computer interaction, from experience-oriented to data-driven, from passive disposal to active discovery, also allows SenseTime to further improve the penetration rate of the smart market.

It is reported that by the end of 2021, SenseTime has deployed city ark in 140 cities, an increase of 49% compared with the end of 2020, ranking first in the market share of computer vision software in China's smart cities.

By the end of 2021, the number of AI models carried by the City Ark will increase to 22,425, an increase of 156% compared to the end of 2020.

Smart Business

More than ten years of traffic competition tends to peak, the value of To B gold is beginning to show signs, under the tide of enterprise digital transformation, SenseTime's smart business has also achieved good results.

China's economy has reached a critical point, and a high-quality economy will be the absolute direction of economic development in the next 10-20 years. AND AI is the best key to opening this critical point.

Ai sweeps in, data exponentially increases, and the result is that the digital world and the physical world intersect more and more, and the boundaries become more and more blurred.

In 2021, SenseTime Smart Business's revenue increased by 31.8% year-on-year to RMB1,958 million. Behind this is SenseTime's AI device delivering AI energy to many industries.

In the field of industrial manufacturing, Foton Cummins engine production plant applies the Deep Spring Industrial Quality Inspection and Training Platform built on SenseCore AI large devices to achieve defect detection of key engine components.

Industrial scenarios often face many types of parts, models, and defects, and the Deep Spring platform provides solutions from multiple optical solution support, multi-component form support, and multiple quality inspection support, which can be accurate enough in the rate of missed detection and false detection.

At the same time, in the face of systematic projects such as AI quality inspection, SenseTime's deep spring platform can also perfectly integrate intelligent technology and production lines to improve the efficiency of production lines and change the iteration of the process from "months" to "weeks".

Nowadays, industrial enterprises have developed from the mass production of a single product to the flexible production trend of multiple products and small batches, and the deep spring platform provides industrial model training components, reasoning workflow scheduling components, report configuration components, low-code support flexible property inspections, to meet the high-quality quality inspection of multiple small batches.

After the application of the Deep Spring platform, the quality inspection efficiency of Foton Cummins Engine Plant has been greatly improved, and with the transformation of the factory to intelligent manufacturing, the competitiveness of enterprises has also been significantly enhanced.

In the field of transportation, SenseTime's AI technology is also shining.

SenseTime, China Railway Electrification Bureau Group Beijing-Shanghai High-speed Rail Vascular Company and China Railway Electrification Institute jointly built the "Starry Sky" 4C (Catenary Suspension Status Monitoring Device) intelligent analysis system, and successfully applied AI technology to the intelligent inspection of high-speed rail catenary network.

The Beijing-Shanghai high-speed railway is a major artery of railway transportation in the mainland, and in daily operation, there is a great demand for detection, and there are difficulties such as long manual analysis cycle and large differences in judgment standards. After the application of the "Starry Sky" system, only 2 technicians can complete the analysis of the same 3 million pictures in 10 days, which increases the efficiency by 20 times, shortens the time for the high-speed rail to complete a catenary routine inspection to just 4 days, increases the efficiency by more than 20 times, and greatly reduces the dependence of 4C detection on personnel experience.

Up to now, the "Starry Sky" system has detected more than 30,000 kilometers of high-speed rail nationwide, with more than 48 million pictures and more than 1.3 billion parts.

It is reported that the "Starry Sky" system not only improves the intelligent level and detection efficiency of catenary vasculars, but also fills the gap in the field of 4C intelligent detection and analysis in mainland China.

Smart cars

The smart driving track was hotter than ever last year.

In 2021, SenseTime's smart car segment continues to develop rapidly, showing a lot of potential.

As early as 2016, SenseTime and Honda jointly deepened the self-driving technology to accelerate the research and development process of smart cars. Accumulating momentum for 6 years, making SenseTime explosive.

Before the unveiling of the intelligent car platform "Absolute Shadow", it has established close ties with more than 30 domestic and foreign OEMs, and the total number of vehicles covered by fixed-point mass production projects (within the life cycle) exceeds 20 million.

After the official force, SenseAuto's shadow from point to surface, from SenseTime's most proud "visual perception technology" technology, from the three directions to create intelligent cars.

The Shadow Intelligent Driving Solution focuses on L2+ level advanced driver assistance based on different sensors to L4 level autonomous driving innovation.

The shadow intelligent cabin solution covers the active human-computer interaction experience of multiple scenarios from the user getting on the car to the car.

The cloud perception platform of the absolute shadow road improves the efficiency of urban and traffic comprehensive management through the integration of vehicle and road cloud.

With a three-way onslaught, SenseTime was able to cover the development of bicycle intelligence (including cabin and cabin) and vehicle-road collaborative intelligence, and open up the full-stack closed loop of the vehicle road cloud. It is reported that SenseTime has promoted the commercialization of the whole line in intelligent driving, intelligent cockpit, vehicle-road coordination, L4 level unmanned driving, and driverless minibuses.

As of December 31, 2021, SenseTime has served more than 40 car companies, and the cumulative number of intelligent driving and intelligent cockpit products has reached 23 million units, ranking first in the industry.

In the organizational structure upgrade of SenseTime in the first quarter of 2022, the intelligent car business group was officially established, and The Shadow rushed towards the direction of "the most influential AI empowerment platform in the automotive industry".

Smart living

SenseTime Smart Life has laid a solid foundation.

In 202, SenseTime also established the Digital Space Business Group, and the strategy of smart living was upgraded.

Under the trend of AI soft and hardware integration, SenseTime cooperated with the world's leading semiconductor companies to develop and deliver four AI sensors at the end of 2021 to achieve a breakthrough of 0 to 1, which can significantly improve the picture quality of mobile phone photos, and process the photos and videos taken by mobile phones with lower power consumption and stronger privacy protection, and successfully landed on the head mobile phone manufacturers.

SenseTime's AI+AR technology has landed in various scenes such as airports, museums, scenic spots, supermarkets, etc., AR navigation, AR tour, AR marketing, AR technology, and centimeter-level positioning, which have greatly enriched the user's scene experience, and SenseTime will integrate the sensory experience of virtual and real from APP to real life.

In the metaverse of the great fire, Shang Tang is a force that cannot be ignored.

SenseTime has carried out the strategic upgrade of "virtual and real integration, from soft to hard, from platform to ecology", under the strategic attention, SenseTime can integrate the advantages of AI perception, AR and MR technology, computing power and customer ecology to create an industry-leading meta-universe empowerment platform, providing a variety of key technology engines to empower various industries to build a digital space.

For example, SenseTime's SenseME Mercury smart mobile terminal platform enables the number of devices to lead the industry. SenseTime's SenseMARS Mars Mixed Reality Platform builds digital space applications and develops interactive immersive experiences that blend virtual and real.

Adhere to long-term doctrine and practice the mission of AI inclusiveness

AI universal access is a gradual path, and the gradual path is not easy to take. After the disenchantment of AI, the AI industry is becoming the path that few people insist on.

SenseTime believes that all the greatness in history began with crossing the river by feeling the stones, and this time, it is one of the few moments when China has run ahead of Europe and the United States, and it is necessary to fearlessly explore the infinite possibilities of AI commercialization.

SenseTime firmly believes that the transformation of artificial intelligence technology brings incremental value, and will also strive to promote the arrival of an intelligent era of inclusiveness and fairness.

As a leading enterprise, this is its responsibility and the mission it wishes.

Leifeng networkLeifeng networkLeifeng network