Text: Hao Xin
Edited by Wang Yisu
AI 1.0 companies want to rely on large models to fight a beautiful "turnaround", and it seems that they are still a breath away.
On April 23, iFLYTEK released its 2023 annual and 2024 Q1 performance reports. In 2023, the operating income will be 19.65 billion yuan, a year-on-year increase of 4.41%, and the net profit attributable to shareholders of listed companies will be 657 million yuan, a year-on-year increase of 17.12%.
However, it is worth digging into the fact that in 2023, iFLYTEK's net profit after deducting non-recurring gains and losses will only be 118 million yuan, a year-on-year decrease of 71.74%, and the company's cash outflow on the books will expand, and the net cash flow generated by its operating activities will be 350 million yuan, a year-on-year decrease of 44.55%.
In 2024, the situation worsens. According to the financial report, in Q1 2024, iFLYTEK's net loss was 300 million yuan, a year-on-year decrease of 418.99%, the loss after deducting extraordinary gains and losses expanded to 441 million yuan, and the net cash outflow from operating activities was 1.384 billion yuan, a year-on-year decrease of 17.04%.
The financial report pointed out that the decrease in iFLYTEK's net profit was mainly due to the investment in general artificial intelligence cognitive models, and the expanding investment, which eventually thinned the profits.
Huge sums of money have been spent, but it is difficult to achieve results in the short term, and the growth brought by AI is still fermenting, which is a common problem faced by some established companies, including iFLYTEK.
After the "darkest 2022" when revenue growth is close to stagnation and net profit jumps off a cliff, what impact has the year of the large model had on iFLYTEK?
The double-edged sword of large models, high investment & high return
Calculate the account for iFLYTEK, the large model is a double-edged sword, and high investment also means high returns.
From 2023 to the present, iFLYTEK's heavy investment is mainly concentrated in three directions, large models, domestic computing power bases and industrial support.
Like other domestic large models, iFLYTEK has also been struggling to catch up with GPT.
Looking back on the development of Xinghuo, from May last year to January this year, with GPT-4 as the goal, it has gone through a total of five iterations:
On May 6, 2023, Xinghuo V1.0 was released;
On June 9, 2023, the Spark V1.5 version was upgraded, and a breakthrough was made in open Q&A, with logical reasoning and mathematical skills upgraded, and multi-round dialogue ability upgraded.
On August 15, 2023, Xinghuo V2.0 was upgraded and released, breaking through the interaction capabilities of code and multimodal capabilities.
On October 24, 2023, Xinghuo V3.0 was released, surpassing ChatGPT in overall capabilities and GPT-4 in the medical field;
On January 30, 2024, Xinghuo V3.5 was released, surpassing GPT-4 Turbo in language understanding, math ability, and voice interaction ability, and at the same time, the Xinghuo voice model was released, surpassing the ability of Whisper's voice model.
iFLYTEK announced that on April 26, 2024, Xinghuo V3.5 will have three new features: long voice and long graphics, one-sentence reproduction of emotional expression, and intelligent twins platform.
On the whole, the internal Spark model has completed the "three milestones" task set by Liu Qingfeng, and externally, compared with the industry, the Spark model has not fallen behind and has entered the first echelon of the large model company. Although there is still a certain gap in the ability of large models compared with companies such as Alibaba, Baidu, and the dark side of the moon, iFLYTEK's advantages in the industry should not be underestimated.
iFLYTEK revealed that it will continue to improve the seven core capabilities of the iFLYTEK Xinghuo general base on the national computing platform "Feixing No. 1", which is expected to benchmark the current best level of GPT-4 turbo in the first half of this year and the current best level of GPT-4V and DALL-E3 in the second half of the year.
In terms of domestic computing power, in October 2023, iFLYTEK and Huawei cooperated to develop the Vanka domestic computing power platform "Feixing No. 1", which supports the training of trillion-parameter large models.
In terms of industrial support, iFLYTEK's large models and AI core technologies have been implemented in education, healthcare, smart city, operators, automobiles, finance and other industries. iFLYTEK revealed in its financial report that a total of 20 AI-related industry projects will be invested in R&D in 2023, of which 14 projects have been completed.
From the computing power required by AI to the layout of large models in the industry, iFLYTEK's "heavy" investment is doomed to high costs, which is also confirmed in the financial report.
In Q1 2024, iFLYTEK will invest about 300 million yuan in the research and development of large models, the independent and controllable core technologies and industrial chains, and the landing and expansion of the large model industry. Among them, R&D expenses were 842 million yuan, an increase of 126 million yuan compared with the same period last year.
After sorting out iFLYTEK's R&D investment expenses and the proportion of operating income from 2017 to 2023, Light Cone Intelligence found that in the past seven years, iFLYTEK's investment in R&D has continued to grow. After the arrival of the large model (2022-2023), the proportion of iFLYTEK's R&D expenses in operating income has increased steadily, and is gradually returning to the peak level of 2018.
(Intelligent Mapping of Light Cone)
Based on the Spark model benchmarked against GPT-4, iFLYTEK is further enhancing its ecological attributes. By the end of last year, iFLYTEK's open platform had opened 647 AI capabilities and large model technical capabilities, opened the assistant ecology to developer teams, built an API matrix, and opened the third-party plug-in market for all scenarios.
Driven by the ecology, rapid growth has been achieved. From the release of Xunfei Xinghuo on May 6, 2023 to the end of 2023, the iFLYTEK open platform has gathered more than 5.78 million developer teams, a year-on-year increase of 52%. Since the release of the Spark large model on May 6, more than 1.676 million developers have been added, an increase of 229% over the same period last year, of which the number of large model developers has exceeded 350,000.
Although the above-mentioned growth cannot be reflected in the newly released financial report, it can be clearly seen that iFLYTEK has put into the imagination space after the large model, and the first stage of efficiency improvement has begun, and it is not far from creating productivity and value.
In the era of large models, iFLYTEK has changed and remained unchanged
Since 2016, iFLYTEK has been betting heavily on deep learning, and has always attached great importance to the implementation of AI and applications, and has achieved the result of doubling its performance in five years.
Before the large model, iFLYTEK's performance fell into the predicament of growth, and at this critical moment, it grabbed the life-saving straw of the large model.
"On the basis of maintaining the continuous innovation and leadership of single-point core technology, we will continue to expand the breadth and depth of the science and technology tree", iFLYTEK once explained its idea of making a large model.
Rising to the level of the entire AI business, iFLYTEK formulated a "platform + track" strategy as early as 2017. In the era of large models, the strategy has been re-inscribed, the "platform" is the open platform of iFLYTEK, which provides developers with AI capabilities and one-stop artificial intelligence solutions to build an AI industry ecology;
By the end of 2023, iFLYTEK has built an AI ecosystem centered on core technologies such as large models, with cities + education + medical care + consumers + enterprises as the center.
From 2017 to 2022, the business in the education sector has been iFLYTEK's No. 1 cash cow business, and 2023 has become a key turning point. Judging from the results, in 2023, the proportion of open platform and consumer business in revenue will slightly exceed that of education business, becoming iFLYTEK's largest business, followed by smart city, operator, smart car, smart healthcare, and smart finance.
(Intelligent Mapping of Light Cone)
The open platform and consumer business is currently the third largest business segment of iFLYTEK, and it is also the fastest growing business segment among the emerging businesses of iFLYTEK, which is highly anticipated by iFLYTEK.
The business segment is divided into three parts: open platform, intelligent hardware, and mobile Internet products and services.
According to the financial report data, the revenue of open platform business increased by 41.61% year-on-year, which is the fastest growing business of iFLYTEK, the revenue of intelligent hardware business increased by 22.35% year-on-year, and the revenue of mobile Internet products and services increased by 17.08% year-on-year.
The open platform is currently the most active part of iFLYTEK, which not only gathers a huge number of developers and enterprise users, but also uses large models to revitalize some old businesses.
For example, in the industrial field, the Antelope industrial Internet platform launched an industrial model based on iFLYTEK Xinghuo, which can give professional suggestions and strategies for enterprise needs and accurately match resources, and the total number of platform users increased by 59% and the number of platform service enterprises increased by 88% in less than a month after the release.
In 2023, the GMV of consumer hardware will increase by 84% year-on-year, the sales of consumer hardware 618 will increase by 125% year-on-year, the sales of Double 11 will increase by 126% year-on-year, and the annual sales of iFLYTEK translators will increase by more than 300% year-on-year.
However, for iFLYTEK's old businesses (education, smart city, operators), it is showing signs of decline. In 2023, the revenue of education products and services will decline by 9.7% year-on-year, the revenue of some smart city businesses will decline by 28.75% year-on-year, and the revenue of carrier business will decline by 0.05%.
While the old business is facing sluggish growth, smart cars, smart healthcare, smart finance and other businesses have begun to emerge. In 2023, the revenue of the smart car business increased by 49.71% year-on-year, the revenue of the smart medical business increased by 15.41% year-on-year, and the revenue of the smart finance business increased by 23.01% year-on-year.
There are some similarities in the use of large models in the smart car, smart healthcare, and smart finance industries. Although the scenarios and requirements of these industries are clear, there are pain points such as huge amounts of data, high requirements for knowledge in specific fields, and cumbersome processes. After the text summary, generalization, analysis and reasoning capabilities of the large model are used, it can reduce the pressure of manual work, improve work efficiency, and assist decision-making to provide users with personalized services, so it has become the first batch of businesses to run out after the large model comes.
Looking for a new growth direction in the midst of changes, it is based on this that facing 2024, Liu Qingfeng has put forward the four annual goals of iFLYTEK's Xinghuo cognitive large model strategy: the Xinghuo large model is the leading benchmark GPT-4 in China, empowering 10 million intelligent hardware terminals and more than 100 million independent software users, ranking first in the education, medical care, finance, automobile, central state-owned enterprises, and urban industry markets, and gathering 1 million large model developers to create the industry's first ecology.
Liu Qingfeng said: "There are three major signs of the success of the Spark model, namely: the core technology first, the number of end users first, and the industrial ecology first, which is also the long-term goal of iFLYTEK. ”
On the whole, iFLYTEK's business is in a period of transition, and when the old and the new are alternated, if we can use the large model to break the old and establish the new, there is still a chance to rush again.
epilogue
In 2016, as if overnight, AI suddenly became hot, and a large number of startups led by CV computer vision became the protagonists in the hot entrepreneurial tide, and the valuation of the AI four tigers exceeded 10 billion, running to go public.
However, to this day, the AI Four Dragons are still mired in losses, but iFLYTEK has made a profit.
The reason why iFLYTEK was able to achieve profitability in the general losses of companies in the AI industry is mainly due to its earlier layout and stronger channels for the government and enterprise market in the early days, which is equivalent to the old giants in the CV industry, Hikvision and Dahua.
But at that time, when the AI Four Tigers were still staring at the to B channel and competing for the opportunity to bid for the project. In the wave of AI 1.0 in 2017, iFLYTEK felt that it had "had enough of the hardships of government and enterprise business", swept away the image of the previous enterprise-level business, and laid out an open platform for developers and intelligent hardware and App business for ordinary consumers in the wave of AI 1.0, creating a new growth curve.
The trajectory of history also coincided by chance at a certain moment,Baidu's heavy bet on autonomous driving has not succeeded,But the C-end business has grown unexpectedly,For example, Xiaodu、Baidu network disk and so on。
Nowadays, as an important milestone in the field of AI, large models are setting off a new wave of technological revolution and commercial applications. Whether to take the B-end route or the C-side, what kind of products and services to land, and what kind of business model to form are ultimately up to the market to decide.
In the last wave of deep learning, AI companies continued to lose money, on the one hand, due to excessive R&D investment and too many customized projects for revenue, which made it difficult to achieve a balance.
After the advent of the wave of large models, it will bring further efficiency improvements, allowing AI companies to see the possibility of paying for C-end products, and also allowing more lightweight deployment possibilities for toB services.
So, can the situation of AI companies continue to lose money?