In the first half of 2024, IFLYTEK wrote in its interim report: "In the first half of 2024, the company invested over RMB 650 million in large model research and development, as well as in enhancing autonomy and controllability of core technologies and industrial chains, and in expanding the industrial applications of large models."

As a result, IFLYTEK's profits turned into losses in the first half of the year.

Recently, IFLYTEK released its latest financial results for 2024, showing that its revenue for the first three quarters increased by 17.7% year-on-year to RMB 14.85 billion; however, its net profit attributable to shareholders turned from profit to loss, declining by -445.91% compared to the same period in 2023 (RMB 99.3621 million), with a non-recurring loss of RMB 468 million.

The large model is aggressively extracting more profits from IFLYTEK's pocket, akin to a "sticky candy," which IFLYTEK is struggling to swallow.

The good news is that, compared to the significant losses in the first two quarters, the net profit in the third quarter stabilized. According to the financial report, IFLYTEK's net profit in the third quarter of this year was RMB 56.9614 million, representing a year-on-year increase of 120.87%, and a quarter-on-quarter turnaround from loss to profit.

Interestingly, Xueqiu indicates that the current number of IFLYTEK shareholders exceeds 350,000, essentially at an all-time high.

(Image source: Xueqiu)

Where there is controversy, there will be disagreements. How many people will continue to support IFLYTEK, and how many will choose to leave?

Taking the initiative to bear pressure: IFLYTEK's strong commitment to autonomy

In fact, IFLYTEK's profitability challenges can be traced back to the impacts of external restrictions faced in 2019 and 2022.

In October 2022, IFLYTEK once again faced extreme pressure.

Although it promptly implemented a series of countermeasures at the time, relying on the stable growth of sustainable base businesses to offset some of the impacts of delayed major projects, and achieving a modest increase in revenue and gross margin that year. However, the impact was difficult to eliminate in one go.

On the one hand, the process from supply chain to contract signing needed adjustment, and unexpected events also affected the pace of current order signing. At that time, more than 20 projects worth over RMB 3 billion were delayed in the fourth quarter, causing the annual revenue growth rate to plummet from 40.6% in 2021 to 2.8%.

On the other hand, experiencing external uncontrollable risks twice in three years stimulated IFLYTEK to plan for a longer-term future, strengthening its position as a "national team" in the AI industry and pushing autonomy and controllability to new heights.

In 2022, IFLYTEK's R&D expenses exceeded RMB 3 billion for the first time. That year, approximately RMB 800 million was newly invested in expanding cooperation platforms for businesses such as education and healthcare, developing new products, and enhancing autonomy and localization of core technologies.

With revenue growth running counter to investment growth, coupled with significant declines in investment income due to fluctuations in financial asset prices, IFLYTEK's net profit attributable to shareholders fell by -63.94% year-on-year to RMB 560 million that year – a manifestation of "taking the initiative to bear pressure."

In 2023, as external influences gradually waned, its profits further increased to RMB 657 million.

During this stage, its revenue scale continued to expand, demonstrating the reliability of its "national team" foundation.

"You can't step into the same river twice." IFLYTEK has faced external restrictions twice, solidifying its unwavering commitment to autonomy and controllability.

All in on large models: IFLYTEK means business

Large models represent a significant technological leap in the history of AI development. With the development potential of the large model track becoming a global consensus, IFLYTEK chose to go "All in."

In 2023, IFLYTEK invested over RMB 2 billion in the development of its iFLYTEK Spark cognitive large model, leading to a significant decline in its non-recurring net profit, down 71.74% year-on-year to RMB 118 million compared to 2022. This led to a rare period of losses in 2024.

Data shows that in the first and second quarters of this year, IFLYTEK's net profit attributable to shareholders and growth rates were: -RMB 300 million and -100 million, respectively; -418.99% and -176.21%, respectively.

Liu Qingfeng, Chairman of IFLYTEK, pointed out at the 2023 performance briefing that the expansion of losses in the first quarter was mainly due to "the company's continued increase in R&D investment, as well as increased market investment." Among them, R&D expenses related to the large model base increased by RMB 420 million, and marketing and promotion expenses related to the large model increased by RMB 120 million.

(Image source: IFLYTEK's official WeChat public account)

The total investment related to large models in the first half of the year was actually even greater, exceeding RMB 1.3 billion. Specifically, this included an increase of RMB 400 million in R&D investment for general large models, an increase of RMB 160 million in R&D investment for the engineering of large model training and inference platforms, an increase of RMB 120 million in investment for large model promotion and application, and an increase of over RMB 600 million in investment related to the productization of large models across various business groups (BGs) and business units (BUs).

If these investments and expenses are excluded, other expenses have actually decreased. Even if the costs related to the R&D and promotion of the Spark large model are deducted, IFLYTEK's non-recurring profit in the first half of the year increased year-on-year. This is the essence of "taking the initiative to bear pressure." R&D is like stockpiling food; spending it is normal, even if it temporarily makes the financial statements look unsightly.

In the third quarter, IFLYTEK's total operating expenses further increased by 17.33% year-on-year to RMB 15.517 billion. Among them, R&D expenses totaled RMB 3.0371 billion for the year, with development expenses increasing by 145.82% compared to the beginning of the period.

To fully support the development of large models, Liu Qingfeng also stated: "The company has sold all of its shares in Sanrenxing, Hanguoji, and SenseTime to focus on large models. Some assets that appreciated in the first quarter of last year are no longer available this year."

This is both determination and resilience. To avoid excessive deviations in profitability while pursuing extreme R&D, Liu Qingfeng explicitly stated at an August performance briefing that unprofitable small businesses would be eliminated, and strategic businesses would be expanded.

Jiang Tao, Vice President and Secretary of the Board, recently stated that over the past year, IFLYTEK has proactively abandoned some projects with difficult funding sources, following a strategic focus on the core competitiveness of the Spark large model.

In the first three quarters, with steady growth in business scale, its gross margin remained stable at 40.45%. In the third quarter, its revenue increased by 17.73% year-on-year to RMB 5.525 billion; concurrently, its gross profit grew by 16.68% to RMB 2.26 billion.

Meanwhile, IFLYTEK Spark's industrial ecosystem has further strengthened, with the number of new developers reaching 1.777 million in the first three quarters, and over 415,000 enterprise customers are using Spark for innovative application experiences. Data shows that the IFLYTEK Spark app has been downloaded over 190 million times on Android. According to Qimai data, "IFLYTEK Spark" ranks first among domestic general large model apps in the tools category.

Focusing on its core business may be one of the key factors driving IFLYTEK's profit stabilization and turnaround.

Pioneering the commercial path for large models

It is particularly noteworthy that despite heavy investments, IFLYTEK's cash flow recovery has been relatively healthy. In the third quarter, it achieved a net operating cash flow of RMB 715 million, representing a year-on-year increase of 100.09%.

In terms of collections, according to Jiang Tao, sales collections in the third quarter reached RMB 6 billion, an increase of 33% year-on-year, significantly outpacing the revenue growth rate for the same period. Following the seasonal pattern of cash flow, the fourth quarter will usher in a peak period for collections. If relevant policies are implemented effectively, the company is confident in its net operating cash flow for the fourth quarter.

This also implies that with the empowerment of large models, IFLYTEK's potential growth cannot be fully reflected by its current revenue and profit levels.

In fact, across the entire large model track, the establishment of a commercial model is crucial for many related companies to survive the long run.

Behind the quarterly turnaround to profitability, strengthened industrial ecosystem, and accelerated collection speed, lies the value proposition of IFLYTEK's AI large model commercialization feasibility.

Earlier, Liu Qingfeng mentioned three commercialization paths for the Spark large model in 2024: First, empowering IFLYTEK's C-end hardware to maintain rapid growth; second, implementing the model in education, healthcare, automotive, finance, energy, and telecom sectors favored by IFLYTEK; and third, empowering external developers through the IFLYTEK open platform to generate revenue.

Recently, Jiang Tao responded to the progress of these three paths. Firstly, the Spark large model is accelerating its large-scale deployment in existing scenarios such as consumer, education, healthcare, and automotive, with the closed-loop commercial value being initially validated.

According to data, its C-end hardware sales grew steadily in the first three quarters with the support of AI. Among them, learning machine sales increased by over 100%, with user satisfaction and recommendation value (NPS) consistently maintaining the industry's top position; consumer and open platform business revenue increased by 44%, and smart auto business revenue increased by 49%, effectively offsetting the decline in G-end business.

(Image source: IFLYTEK's official WeChat public account)

Secondly, focusing on large models, IFLYTEK is deeply engaged in strategic cooperation with central and state-owned enterprises in industries such as telecom, finance, energy, and transportation.

Currently, large models are accelerating their deployment in central and state-owned enterprises, and IFLYTEK stated that the "IFLYTEK Spark" large model has become the preferred choice for these enterprises.

The report shows that IFLYTEK has undertaken multiple national strategic tasks assigned by the National Development and Reform Commission, Ministry of Industry and Information Technology, Ministry of Science and Technology, and Chinese Academy of Sciences.

According to financial data cited in the financial report, this year, the number of awarded projects in China's large model market reached 653, with IFLYTEK successfully winning 38 projects, the most among market players in the first half of the year. In August alone, IFLYTEK won bids worth over RMB 150 million, ranking first in the industry.

For example, in collaboration with China Telecom, IFLYTEK created a large model for the marine energy sector for CNOOC. Together with China Mobile, it developed the "Kunlun Large Model" for the energy and chemicals sector for PetroChina. It also collaborated with China Evergreen to build the first green industry large model, the "Taishan Large Model," co-constructing the first AI headquarters for central enterprises. Its "national team" image in the B-end is prominent and distinctive.

(Image source: IFLYTEK's official WeChat public account)

The third path involves achieving large-scale promotion in the enterprise market through standardized products such as iFlyCode, procurement, meetings, documents, and translation.

Currently, the coding capabilities of the Spark large model, tested by Bank of Communications and JD.com, are leading in the industry. Since its launch, iFlyCode has been applied in over 100 institutions, including JD Cloud, iSoftStone, and Bank of Communications, bringing significant economic benefits and improving efficiency.

The initial success of the Spark large model's profit model has given IFLYTEK greater confidence to further refine its efforts.

The management team indicated that in the coming years, it will no longer blindly increase investments in computing power but will focus on achieving deep deployment of large models with limited investments.

It is expected that in the next two to three years, models with parameters ranging from tens of billions to hundreds of billions, or even lower, will be deeply applied in various fields, and the company has significant advantages in this regard. Especially in the fields of smart cockpits and robots, IFLYTEK's "voice interaction + backend large model" path will bring unique competitiveness.

"With limited investments, I believe we can successfully deploy large models,"

In two days, it will be October 24th, and IFLYTEK's Global 1024 Developer Festival will arrive. At the festival, it will once again upgrade the capabilities of the iFLYTEK Spark large model base, with improvements in mathematics, coding, and long-text capabilities. Additionally, it will unveil multi-modal visual interaction and ultra-realistic avatar interaction capabilities for the first time.

(Image source: 1024 Developer Festival official website)

New information brings new perspectives. Perhaps in the fourth quarter, the number of IFLYTEK shareholders will undergo significant changes.