HomeArticle

Returning to double-digit growth, iFlytek has shown remarkable resilience in the large model competition.

36氪财经2025-04-28 08:00
With the large-scale implementation of the SparkDesk large model in fields such as education, healthcare, and finance, combined with the collaborative advantages of domestic computing power and independent algorithms, the company's technological leadership and commercialization capabilities have been doubly verified.

Recently, iFlytek released its annual report for 2024. The financial report shows that last year, iFlytek's revenue reached 23.343 billion yuan, a year-on-year increase of 18.79%. This is iFlytek's return to double-digit growth after a two-year hiatus. During the same period, the net profit attributable to the parent company was 560 million yuan.

The first-quarter report released on the same day was also impressive. In the first quarter of 2025, the company recorded revenue of 4.658 billion yuan, a year-on-year increase of 27.74%. The net profit attributable to the parent company and the net profit after deducting non-recurring gains and losses increased by 35.68% and 48.29% year-on-year respectively, and the net operating cash flow increased by 48.54% year-on-year.

iFlytek had previously experienced an adjustment period of slow growth. Now, this impressive financial report marks that iFlytek's early investment in the AI track has entered the harvest period.

With the large-scale implementation of the Xinghuo large model in fields such as education, healthcare, and finance, combined with the synergistic advantages of domestic computing power and independent algorithms, the company's technological leadership and commercialization capabilities have been doubly verified.

01 Revenue Returns to Double-Digit Growth, Showing Profitability Resilience

From the latest financial report, it can be seen that in 2024, iFlytek achieved a net profit attributable to the parent company of 560 million yuan and a net profit after deducting non-recurring gains and losses of 188 million yuan, a year-on-year increase of 59.35%. The profitability of the main business is gradually recovering.

Meanwhile, the company's cash flow reached a record high. The net inflow of operating cash flow for the whole year was 2.495 billion yuan, more than six times the year-on-year increase. The financial report disclosed that the sharp increase in cash flow was mainly due to the increase in sales collections compared with the same period last year.

iFlytek attaches great importance to collection management. In 2024, it specially established a collection department to continuously strengthen the management of accounts receivable.

"This includes strengthening the regular collection work, forming a GBC business structure with better cash flow, and improving the timeliness of collections for G-side businesses through the selection of high-quality customers and government special bonds," introduced Wu Xiaoru, President of iFlytek, at the earnings conference.

Behind the reorganization of the collection work also includes the reorganization of the business structure, which is the foundation for the commercialization of large model companies. Technology alone cannot win the market; the right PMF (Product-Market Fit) can.

According to Wu Xiaoru, in terms of the business structure, iFlytek has established the strategy of "optimizing the C-side, strengthening the B-side, and selecting the G-side." "The proportion of C-side revenue continues to increase, and the sales collection rate is very high. For the B-side, high-quality industries with guaranteed cash flow are preferred, and for the G-side, cooperation is carried out in regions with better fiscal conditions."

After two years of precipitation, the results of the company's proactive adjustments have been smoothly reflected in the financial report. The company's various indicators initially show an enhancement of profitability resilience.

The financial report shows that the proportion of revenue from high-margin businesses (such as smart education and healthcare AI) has increased. Combined with the company's technological barriers in niche fields, it supports the overall gross profit margin to remain above 40%.

During the reporting period, the company's core track businesses maintained rapid growth. Data shows that the operating revenues of the consumer, education, automotive, and healthcare businesses increased by 27.58%, 29.94%, 42.16%, and 28.18% year-on-year respectively, and the revenue structure continued to be optimized.

The smart education business achieved revenue of 7.229 billion yuan, a year-on-year increase of 29.94%. Among them, the sales volume of AI learning machines increased by more than 100%. The automotive, healthcare, and enterprise AI solution businesses achieved revenues of 989 million yuan, 692 million yuan, and 643 million yuan respectively, with year-on-year increases of 42.16%, 28.18%, and 122.56% respectively.

02 Why Localization Is So Important

In essence, in the commercialization competition of large models, the key for iFlytek to lead lies in its full-stack independently controllable technological barriers.

In the past two years, iFlytek has built a full-stack AI technology system around "domestic computing power + independent algorithms," achieving full independent control of elements such as algorithms, computing power, and data.

Against the backdrop of geopolitical turmoil in recent years, "independent control" has become an important development direction for the technology industry chain. Especially for central and state-owned enterprises, the localization of computing power is a prerequisite for building competitive barriers.

The full-stack independently controllable technological strength not only ensures the company's leading position in technology but also provides a stable development foundation for the company in a complex geopolitical environment.

It is worth noting that iFlytek's full-stack independently controllable large model is the only general large model trained by a fully domestic computing power platform among the general large models open for national download in China.

Since its launch in May 2023, the Xinghuo large model has been continuously iterated and upgraded. In October 2024, it was upgraded to Xinghuo 4.0 Turbo, and in January 2025, the Xinghuo deep reasoning model X1 was launched.

On April 20, iFlytek's deep reasoning large model Xinghuo X1 was further upgraded, achieving significant improvements in general tasks such as mathematics, code, logical reasoning, text generation, language understanding, and knowledge answering.

With the model parameters one order of magnitude smaller than similar models in the industry, the performance of general tasks has reached the level of international top models such as OpenAI o1 and DeepSeek R1.

More importantly, with the support of fully domestic computing power, the full-version Xinghuo X1 can be deployed with only 4 cards (Huawei 910B), simplifying the private deployment process.

At the same time, Xinghuo X1 has also upgraded its model customization and optimization toolchain, supporting two model customization and optimization solutions: SFT (Supervised Fine-Tuning) and reinforcement learning. Only 16 cards (Huawei 910B) are needed to complete industry customization and optimization, effectively reducing the customization threshold for enterprise AI.

According to third-party data, in 2024, iFlytek's large model ranked first in both the number and amount of winning bids, covering multiple industries such as communications, finance, energy, education and science, government affairs, and healthcare.

The leading position of the AI core technology research and model training system with independent control of the core technology base and elements such as algorithms, computing power, and data has been further verified.

"Based on the current algorithms of pre-training and reinforcement learning and domestic computing power, iFlytek has the ability to make the base model the best in the industry," said Liu Qingfeng. The industry models trained based on the independently developed base model are better than those trained based on open-source models in terms of size flexibility, adaptation depth, and final performance.

03 The AI Harvest Season

At the earnings conference, Liu Qingfeng specifically emphasized the necessity of the self-developed large model base.

Not only from a technical perspective but also for the implementation of large model applications, he believes that continuously improving the original R & D ability with less computing power, smaller size, and less resource occupation is the key for Chinese large models to empower all industries.

"When training industry models, we took into account the needs of relevant fields such as education, healthcare, and justice. Therefore, when training the base model, we conducted higher-quality cleaning of the basic data resources."

Liu Qingfeng revealed that many central and state-owned enterprises found that open-source models had too many hallucinations and security vulnerabilities after pilot deployment at the beginning of the year and then turned to iFlytek for professional solutions.

"Building a base large model on a domestic computing power platform is of strategic significance to the country and also has unique application needs in the market," pointed out Liu Qingfeng.

"Whether it is central and state-owned enterprises or important industries such as security, the demand for an independently controllable base large model is very urgent. iFlytek is trained based on domestic computing power, and its reliability in all aspects is superior, making it more trusted by relevant customers and national authorities."

It can be seen that achieving full-stack independent control of large models has become extremely important under the current national strategic needs. This has become the key to iFlytek's differentiated competition in the G and B-side commercialization.

On the C-side, iFlytek also achieved high-speed growth last year. Its layout in the intelligent hardware field is mainly reflected in the rapid development and market penetration of products such as AI learning machines and intelligent voice recorders.

Data shows that the sales volume of AI learning machines increased by more than 100% in 2024, becoming an important growth point for the company's smart education business.

It is reported that the company launched the dual-screen translator 2.0 during the opening of the Osaka Expo. Liu Qingfeng revealed that iFlytek will launch a translation app and SaaS services. "The SaaS services of iFlytek Hearing have grown very fast. The performance almost doubled last year, and there was also a high growth rate in the first quarter."

It is foreseeable that in the future, with the further application of the Xinghuo large model in the intelligent hardware field, the company is expected to open up a new growth curve in this direction.

Relying on the in-depth penetration of scenarios such as education, healthcare, and smart cities, iFlytek has formed a closed loop of "technology R & D - scenario verification - large-scale implementation."

Moreover, the C-side and G & B-side businesses have a mutual promotion effect. The continuous in-depth exploration of these scenarios not only provides a large amount of industry data for large model training but also builds an irreplicable competitive barrier through the positive cycle of "data - model - product."

For example, as the core source of revenue, smart education products such as intelligent education software improve teaching efficiency through personalized learning programs and cover a large number of schools and research institutions across the country. In the healthcare field, iFlytek's Intelligent Medical Assistant passed the clinical comprehensive written examination of the National Practicing Physician Examination, and the practicality of the technology was verified by the authority, promoting the accelerated entry of healthcare AI products into hospitals.

At the earnings conference, iFlytek proposed to "emphasize productization" and increase the proportion of hardware - software integrated products in revenue in the future. "Productization" is different from customized services, with stronger replicability and more guaranteed gross profit margin and profit.

For any new technology, productization is a necessary prerequisite for achieving scale, and large models are no exception.

Although the capabilities brought by large models are powerful, for individuals and enterprises, it is actually unclear which specific needs large models can solve, and users need to explore on their own. Therefore, the lack of product certainty is a major pain point in the implementation process of large models.

iFlytek's emphasis on productization is precisely to solve this problem. Based on the specific functionality of hardware, the integration of hardware and software can atomize the complex capabilities of large models through the penetration of hardware to solve specific problems.

Liu Qingfeng said that in the future, there will be fewer projects that require addition for iFlytek, and more products that can achieve multiplication.

Currently, the large model industry has reached an awkward middle ground. Technologically, many industry players are hesitating whether to continue investing in underlying models. Commercially, it is mainly about cost reduction and efficiency improvement, and there are no killer products.

In such a situation, whoever can first find a niche market suitable for themselves and generate cash flow will stand more stably.

iFlytek was the first to seize the dividends of the localization of the industrial chain. With the large-scale implementation of the Xinghuo large model in fields such as education, healthcare, and finance, combined with the synergistic advantages of domestic computing power and independent algorithms, iFlytek's technological leadership and commercialization capabilities have been doubly verified. The stable cash flow will support the continuous realization of iFlytek's long-term value.