2024/08/23 10:42
Large model, a battle that iFLYTEK cannot lose
2024/08/23 10:42
Author | Forest system
On the evening of August 21, iFLYTEK released its 2024 semi-annual report. Although the company achieved revenue of 9.325 billion yuan in the first half of the year, a year-on-year increase of 18.91%, the net profit loss reached 401 million yuan, a year-on-year decrease of 644.59%; After deducting non-profits, the net profit continued to lose 483 million yuan, a year-on-year decrease of 58.86%, compared with a loss of 304 million yuan in the same period last year, and the loss further expanded.
From its listing in 2008 to 2021, iFLYTEK has maintained a good record of double-digit revenue growth. This is the first time since its listing that it has handed over a half-year loss report card. iFLYTEK attributed the reason to the company's new investment of more than 650 million yuan in large-scale model research and development, independent and controllable core technology and controllable industrial chain, as well as the landing and expansion of large-scale model industry.
At the end of 2022, driven by ChatGPT, a wave of generative large-scale model research and development swept the world.
On the one hand, there are fast-growing user dividends and technical prospects, and on the other hand, there is huge cost pressure and update investment.
But if you want to say who feels the most real about this situation, iFLYTEK, as the "national team of the artificial intelligence industry", may have a lot to say.
On the one hand, it has really eaten the dividends of AI, not only taking advantage of the east wind of the AI large model to be sought after by the market, so that the stock price has doubled in the short term, and even after the large model empowers the traditional business, it has the posture of a "hexagonal warrior"; On the one hand, the financial statements began to lose more and more money, especially the newly developed AI large model business also explained what it means to "spend money like water".
Lei Jun once said, "Standing in the wind, pigs can fly."
Of course, Rebus was motivated by feelings, but for iFLYTEK, it is still difficult to gain a firm foothold in the tuyere.
From the perspective of time trajectory, fast is the intuitive feeling that the iFLYTEK Spark model gives to the industry.
In just over a year, the iFLYTEK Xinghuo model has been updated from version 4.0 to version 4.0.
Especially on June 27 this year, iFLYTEK released the Spark large model V4.0, which has greatly improved the model capabilities. The upgraded iFLYTEK Spark V4.0. According to the official claim, it has fully benchmarked GPT-4 Turbo in base capabilities, and 8 of the internationally renowned HumanEval, WinoGrande, GPQA and other 10 English evaluations and 2 Chinese evaluations such as C-Eval and CMMLU have surpassed GPT-4 Turbo.
"Benchmarking GPT-4", how does this sound so familiar.
A year ago, the battle of 100 models was in full swing, and the princes of all parties were sharpening their knives and guns, and at the same time they released heroic words, showing their ambitions.
iFLYTEK is one of them.
On May 6 last year, iFLYTEK released version 1.0 of the Spark model, when Liu Qingfeng said that the Spark model should benchmark ChatGPT in October, not only in Chinese, but also in English to reach the same level as ChatGPT, although GPT-4 came out in March last year.
This ambitious declaration could not resist the enthusiastic response of market sentiment, and the share price of iFLYTEK rose to the limit, and then continued to climb, hitting a record high, and the market value once exceeded the 170 billion yuan mark.
What is happening now? After a year, iFLYTEK released the iFLYTEK Spark model V4.0 in Beijing and its application in many fields such as healthcare, education, and commerce. It is understood that the model is based on the training of the country's first domestic Wanka computing power cluster "Feixing No. 1", and the model capabilities are fully benchmarked against GPT-4 Turbo, and surpass in text generation, language understanding, knowledge question and answer, logical reasoning, and mathematical ability.
What is particularly striking is that in the 12 mainstream test sets of large models at home and abroad, iFLYTEK Xinghuo V4.0 stood out in 8 of them, and the overall performance surpassed GPT-4 Turbo and other top international large models.
Liu Qingfeng further revealed that the number of downloads of the Xinghuo app has surged to 131 million, which not only attracted a large number of loyal users, but also gave birth to a series of popular app assistants. Powered by the Xinghuo model, the sales of some smart hardware products have achieved significant growth, with a year-on-year increase of more than 70%, and the average monthly use has exceeded 40 million times.
In 2024, iFLYTEK remains ambitious.
Chairman Liu Qingfeng and President Wu Xiaoru said at the 2023 annual meeting that in 2023, iFLYTEK has ranked among the first echelon of China's large model, and the goal in 2024 is to achieve "1 million ecosystems, 10 million hardware terminals, and 100 million software users".
There is no doubt that the super wave brought by OpenAI has brought all related industrial clusters to the top simultaneously, and iFLYTEK has occupied a dominant ecological niche in this wave and has become one of the few core targets in China with large-scale model technology.
However, looking at such a wide range of AI large model business layout this year, iFLYTEK has failed to enter the fast lane of rapid growth as expected, and the dilemma of increasing revenue but not increasing profits still plagues this star AI company.
In 2023, iFLYTEK's performance will be like a "roller coaster": from the first quarter to the fourth quarter, the net profit will be -57.8953 million yuan, 131 million yuan, 25.7901 million yuan, and 558 million yuan respectively. The non-net profit deducted for the whole year of 2023 decreased by 71.71% year-on-year, and the financial report for the first quarter of 2024 shows that although iFLYTEK's revenue increased by 26.3% year-on-year to 3.65 billion yuan, the loss has intensified - the loss attributable to the parent company is as high as 300 million yuan, and the non-profit deduction loss has reached 440 million yuan.
Especially after the release of this year's semi-annual report, iFLYTEK suffered its first loss after listing, and its stock price was hit hard.
Compared with the peak of market value after the release of the AI model iFLYTEK Xinghuo a year ago, the total market value of iFLYTEK has nearly shrunk by nearly 100 billion.
In this wave of AI large models, high investment has become the direct driver of losses in iFLYTEK's AI investment strategy. According to statistics, in 2023, iFLYTEK will invest more than 2 billion yuan in the research and development of the iFLYTEK Xinghuo cognitive model, and the total R&D investment will exceed 3.837 billion yuan. At the beginning of 2024, the company will invest about 300 million yuan in the research and development of large models, independent and controllable core technologies and controllable industrial chains.
To engage in large models, computing power is a big money. In 2023, iFLYTEK will purchase 1.7 billion yuan of server equipment, and the original balance on the books will be less than 300 million. iFLYTEK said that the increase in investment in computing power in the future will be smaller than that in research and development. Nevertheless, according to the depreciation period given in the annual report, the additional 1.73 billion yuan will be amortized at least 300 million yuan per year.
In this context, at the quarterly performance briefing this year, Liu Qingfeng reiterated the necessity of the company to increase R&D investment. "In the face of huge industrial opportunities, iFLYTEK will not hesitate to saturate investment" "Last year was the first year of the Spark model, and this year is the Spark of the prairie fire, and market investment is bound to increase." Obviously, continuous capital injection is still the main theme of iFLYTEK's 2024 strategy.
However, to be honest, with the blessing of the Spark model, many of iFLYTEK's businesses have begun to be linked to the strength of the model, which has given the market a lot of surprises, which can be seen from the financial report. Not only that, but on the free software application side, it has also been recognized by more and more users.
Since its full opening in September last year, the Xunfei Xinghuo APP has been downloaded 131 million times on the Android side, ranking among the best in the domestic large-scale model APP, and a large number of popular assistants loved by users such as copywriting masters, writing assistants, and PPT outline assistants have emerged to meet user needs in just need scenarios such as work, study, and life. On Apple's App Store, user evaluation and scoring are also ahead of domestic peers, and the scale of users ranks among the top in the industry.
Before the iFLYTEK Spark model came out, in 2021, Liu Qingfeng set the goal of "one billion users, 100 billion revenue, and trillion ecosystems" of iFLYTEK, "These tasks, we hope to achieve it in the next 5 years, during the '14th Five-Year Plan' period." "And the revenue in 2022 is less than 19 billion, and there is almost no growth.
In addition, iFLYTEK has been questioned for a long time, such as being good at asking for money (government subsidies) and not good at making money. Therefore, the iFLYTEK Spark model has been regarded as the savior of iFLYTEK since its birth.
According to a number of institutions, the large model will open a new era of growth. Goldman Sachs has predicted that generative AI will drive global GDP growth of 7% over the next 10 years, or nearly $7 trillion, and Gartner predicts that more than 80% of enterprises will use generative AI by 2026.
However, from the current point of view, it is really difficult to expect iFLYTEK's revenue target of 100 billion yuan to be achieved by relying on large models in the short term.
First of all, the large model is a protracted battle to burn money, judging from the financial report data, iFLYTEK's investment in the large model has continuously raised the company's R&D expenditure, and in the case of the company's main business is weak, iFLYTEK's financial pressure still exists. The market has also begun to question whether iFLYTEK's bet on the AI large model track and investing heavily in R&D funds is the right path.
Secondly, the problems of the current domestic large-scale model products are: first, the degree of technical intelligence is not enough; Second, the industry has already rolled up, and it has begun to roll up market share before an epoch-making breakthrough.
For example, compared with OpenAI, the "leader" of the large model track, iFLYTEK still has a big gap in technology.
Although iFLYTEK has repeatedly claimed that the iFLYTEK large model should surpass ChatGPT in terms of functionality, Chairman Liu Qingfeng also admitted that "there is still a gap between the domestic large model and GPT-4 in terms of complex knowledge reasoning, small sample rapid learning, ultra-long text processing, and cross-modal unified understanding". On the other hand, OpenAI is still continuing to develop and optimize ChatGPT. According to OpenAI CTO Mira Murati, GPT-5 could be released in late 2025 or early 2026, reaching the level of "PhD-level" intelligence in specific tasks.
It can be seen that as long as there is competition and you want to surpass your opponents, there is no end in sight to the investment in large models. And iFLYTEK's bet on the large model this time is undoubtedly a "big gamble". Under the melee of large model technology, iFLYTEK has come out of the predicament of profitability with the iFLYTEK Xinghuo large model, but there is still great uncertainty.
On the other hand, with the technical competition of large model manufacturers at home and abroad, the price war in the large model market is becoming more and more intense. Companies such as ByteDance, Alibaba Cloud, Baidu, and Tencent have joined the battle, and prices have dropped again and again. Some industry insiders even joked that the next follow-up manufacturer may have to pay money to reward customers for using large models!
In May this year, ByteDance's cloud platform Volcano Engine announced that its self-developed bean bag model was officially provided to the outside world, and the pricing was significantly lower than the mainstream level of the industry. Subsequently, a number of manufacturers such as Alibaba, Baidu, Tencent, and iFLYTEK followed suit to reduce prices.
Behind the price war is the survival anxiety caused by the reduction of computing power costs brought about by the accelerated evolution of technology and the involution of market competition. Vendors are competing to reduce the cost of using large-scale APIs in order to compete for market share.
In the early stages of a "cash-burning" business, this is not a good sign.
In addition, iFLYTEK is also facing severe challenges in the promotion of large models. Although "iFLYTEK Xinghuo" has made many breakthroughs in technology, how to transform these technologies into actual productivity and achieve commercialization is a complex and long process.
Who to ask for money and how to get more money is a must-answer question for so-called AI companies in the era of large models.
The chairman of iFLYTEK once said that there are three commercialization paths for the Spark model in 2024: the first is the empowerment of C-end hardware to maintain rapid growth; The second is to land in education, healthcare, automobile, finance, energy and operator tracks, that is, B-side service empowerment and large-scale model customization; Finally, it is the iFLYTEK open platform that sells APIs.
Since the official release of the Xinghuo model by iFLYTEK in May 2023, in summary, the monetization methods of the Xunfei Xinghuo model are divided into three ways: empowering existing businesses, API authorization and payment, and enterprise customized models, but only the C-end hardware under the existing business segment has obvious effects, of which the revenue of smart hardware reached 1.617 billion yuan, a year-on-year increase of 22.35%, and the smart car business achieved 695 million yuan, an increase of 49.71%. It is clear that the latter two models do not make money is the norm.
Earlier, there were rumors that many of Huawei's businesses were going to terminate cooperation with iFLYTEK. iFLYTEK refuted the rumors that the information was untrue, and the company maintained a good cooperative relationship with Huawei in various aspects such as Ascend computing power, intelligent terminals, HarmonyOS ecosystem, and sports and health.
For iFLYTEK, the all-out bet on the "iFLYTEK Spark Model" is undoubtedly a big gamble full of challenges and uncertainties.
*The copyright of the first image and accompanying picture of the article belongs to the copyright owner. If the copyright owner believes that his work is not suitable for everyone to browse or should not be used for free, please contact us in time, and the platform will correct it immediately.
This article is from Xinzhi self-media and does not represent the views and positions of Business Xinzhi.If there is any suspicion of infringement, please contact the administrator of the Business News Platform.Contact: system@shangyexinzhi.com