iFLYTEK boasts a strong government-sector (G-end) foundation, which has been its strength. However, in the AI era, this has also become its comfort zone, constraining its ability to generate revenue. To create commercial opportunities in the consumer (C-end) and business (B-end) markets, iFLYTEK needs to make substantial investments, suggesting that its long-standing profit miracle may now confront unprecedented challenges.

In the A-share market, iFLYTEK stands out as a unique entity.

Since initiating financial reporting in 2005, iFLYTEK has consistently reported profits for 20 years without incurring a loss.

However, maintaining this record has become increasingly difficult with the advancement of AI.

In the first half of the year, iFLYTEK's non-GAAP net loss reached 637 million yuan, marking a 75% year-on-year increase.

Over the past two years, iFLYTEK has managed to reverse losses through its fourth-quarter performance, although deficits in the first three quarters have continued to grow.

Part.01 AI as a Low-Impact Factor

AI has not significantly impacted iFLYTEK.

In terms of revenue, iFLYTEK's growth rate has been declining. In the second quarter, revenue reached 6.349 billion yuan, reflecting a slight year-on-year increase of 1.5%. Regarding gross profit, it stood at 40.4% in the second quarter, showing no significant change and even lower than before.

iFLYTEK operates in the "To B, To C, To G" markets, with the C-end primarily focused on education, including learning devices, translation devices, and AI glasses.

According to Runto data, in the first quarter of this year, iFLYTEK held a 18.5% sales share, ranking third. However, iFLYTEK's learning devices have a relatively high market share in the premium segment above 6,000 yuan.

The B-end business primarily involves integration with various industries, such as AI+education, AI+healthcare, and AI+finance. The G-end business entails cooperation with government departments, such as digital government and smart governance initiatives.

In fact, the distinction between the B-end and G-end is not very pronounced. Many of iFLYTEK's enterprise cooperation users are central state-owned enterprises, such as telecom operators, China Development Bank, PetroChina, and CNOOC, with a high degree of similarity in their cooperation styles.

These clients generally operate under a bidding system, and once cooperation is established, stability is high, and performance is less likely to fluctuate. However, these enterprises do not have stringent requirements for iFLYTEK's AI technological capabilities, which are more of a continuation of past projects.

In other words, iFLYTEK's core competitiveness lies in the stable relationships it has cultivated with governments and enterprises over the years, rather than being driven by technological capabilities.

According to its official website, iFLYTEK ranked first among Chinese general-purpose large model vendors in terms of the number of projects won and the total contract value for two consecutive years in 2024 and 2025. In the first half of 2026, iFLYTEK also led the industry in these metrics. Regardless of changes in external large model competition, iFLYTEK's orders remain relatively stable.

iFLYTEK receives government subsidies almost every year, with the amount generally trending upwards. In the first half of the year, government subsidies included in iFLYTEK's profit and loss amounted to 306 million yuan, compared to 636 million yuan in 2025, reflecting a 52% year-on-year increase.

On this basis, iFLYTEK also employs a key strategy to optimize profits—capitalizing R&D investments.

Capitalizing R&D investments is a concept contrasted with expensing R&D investments. When a project is in the development stage before commercial production and investments can generate benefits, R&D investments can be capitalized. However, if the project's prospects are uncertain, they are expensed.

The distinction is that expensed items are included in costs, while capitalized items are included in the balance sheet. iFLYTEK has consistently maintained a relatively high proportion of capitalized R&D investments. In the first half of the year, capitalized R&D expenditures were 1.227 billion yuan, equivalent to 48% of expensed R&D expenditures.

Having orders, subsidies, and optimizing expenses through accounting methods are key factors supporting iFLYTEK's steady growth and profitability.

In the first half of the year, iFLYTEK achieved revenue of 11.623 billion yuan, marking a 6.52% year-on-year increase. Although it reported a loss, iFLYTEK clearly stated in its semi-annual report that in the past two years, the company's semi-annual net profit attributable to shareholders has been negative, but it has turned positive for the full year, with annual operations continuing to improve. It remains confident in maintaining profitability.

So, if we set aside the stable bidding business in the B-end and G-end markets and focus solely on revenue generation capabilities, how is iFLYTEK faring?

Part.02 Generational Gap of Spark

iFLYTEK does not clearly distinguish the specific performance of the three ends in its financial reports.

However, some indicators can be gleaned from changes in certain businesses.

Revenue from the open platform business was 3.705 billion yuan, reflecting a 36% year-on-year increase. Revenue from the smart hardware business was 732 million yuan, marking a 16% year-on-year decrease. Revenue from enterprise AI solutions was 226 million yuan, reflecting a 49% year-on-year decrease.

The open platform business is driven by both the "API economy" and the "large model Token economy." This is currently the most popular business model for large models. API interfaces and Token consumption are experiencing rapid growth. In the same industry, Zhipu saw a 400% increase in the first half of the year, while Minimax saw a 283% increase. iFLYTEK's growth is largely influenced by industry trends, and although its growth rate is not as high as Zhipu and Minimax, its large base indicates relatively good development.

The decline in smart hardware is related to the limited size of the education market, which has little potential for further growth. The significant decline in enterprise AI solutions may be related to project fluctuations and also serves as indirect evidence of iFLYTEK's AI capabilities.

iFLYTEK's Spark model already exhibits a generational gap compared to current mainstream models.

According to financial reports, the Spark X2-300B version upgraded on June 30 is generally equivalent to GLM-5.1.

Zhipu's GLM-5.1 is an open-source model released by Zhipu on April 8, 2026, and Zhipu has now reached version 5.3. This indicates a nearly three-month gap between the two. Given the current frequency of large model updates and releases, three months already represents a generational gap.

Previously, iFLYTEK was one of the first companies to launch a large model, but as technology has evolved, iFLYTEK has gradually fallen behind.

This is, of course, related to the level of investment. In the first half of the year, iFLYTEK's R&D expenses were 2.535 billion yuan, reflecting a 22.6% year-on-year increase, compared to only a 7% increase last year, indicating a relatively slow growth rate.

The insistence on full domesticization is also a key factor. iFLYTEK has consistently focused on training and inference based on fully domestic computing power.

On September 7, iFLYTEK released its latest Spark X2.5 large model, which was fully trained and inferred based on fully domestic computing power.

At present, there is a certain gap in fully domestic computing power. Liu Qingfeng once provided a quantitative statement: the training efficiency gap between domestic computing power and H200 is fivefold. This has clearly impacted iFLYTEK.

Liu Qingfeng once said, "In this wave of large model enthusiasm, due to the limitations of domestic computing power on ultra-long context, iFLYTEK appears to be lacking in this wave of market hotspots focused on code and intelligent agents."

But this is a burden iFLYTEK has to bear. Due to its close and stable cooperative relationships with government departments and central state-owned enterprises, iFLYTEK should naturally provide more reliable security guarantees while also more strongly supporting domestic computing power enterprises.

Currently, the leading domestic large models are DeepSeek, Kimi, Zhipu GLM, and Minimax. Other large models generally need to conduct business based on their own platform and ecological advantages, such as Alibaba leveraging Alibaba Cloud and DingTalk ecosystems; ByteDance leveraging Douyin and Feishu.

iFLYTEK does not have such an ecosystem and finds it difficult to break through in the competition of large models.

Based on its current position and trends, the education market has a limited scale, and large models are lagging. iFLYTEK is unlikely to have sufficient revenue generation capabilities in the C-end market and will continue to rely on stable output from central state-owned enterprises in the B-end and the G-end.

But iFLYTEK does not hope for this situation and has higher expectations for C-end revenue generation capabilities.

Part.03 Bets for the Second Half of the Year

iFLYTEK's current strategy is to "strengthen the C-end, deepen the B-end, and selectively focus on the G-end."

Specifically, for G-end projects, a more cautious business strategy will be adopted, concentrating operations in the C-end and B-end business areas with higher returns and strong cash flow, and promoting a shift in business structure from project-driven to product and user-driven.

iFLYTEK is placing its bets on the second half of the year.

Liu Qingfeng also provided a timeline: a phased version of the new-generation general-purpose large model will be launched at the end of August; at the "1024 Developer Day" on October 24, a new main large model based on fully domestic computing power will be released, aiming to enter the first tier in China for code capabilities and Token cost-effectiveness.

He also revealed that R&D investment in the second half of the year is expected to be more than double that of the first half, with about 50% allocated to large models. In other words, R&D investment in the second half of the year will exceed 6 billion yuan, with a direct investment of 3 billion yuan in large models.

To this end, iFLYTEK revealed plans to establish Spark as an independent company, implement flatter management, and introduce external financing to promote large model business.

This is similar to Kuaishou's approach, allowing large models to achieve faster development through independent operations.

Under the circumstances of large-scale investment, a more flexible and efficient organizational structure is needed to maximize the value of funds.

But this is undoubtedly still a high-stakes gamble, at the cost of iFLYTEK's operational stability. In the first half of the year, iFLYTEK's operating cash outflow was 945 million yuan, reflecting a 22% year-on-year increase.

iFLYTEK has consistently followed a seasonal curve of "losses in the first half of the year and profits for the full year": education and government projects are planned in the first half of the year and accepted in the second half, with R&D and labor costs evenly allocated throughout the year. In 2025, the company's net profit attributable to shareholders was 839 million yuan, and net profit after non-recurring items was 264 million yuan. This curve relies on concentrated project acceptance in the second half of the year, especially in the fourth quarter.

When the revenue structure shifts from "selling projects" to "selling products," can the seasonal dividends of project-based business withstand the continuous investment in the product side? With R&D investment exceeding 3 billion yuan in the first half of the year and set to double in the second half, relying on government subsidies and project acceptance to generate profits is becoming increasingly insufficient.

This means that iFLYTEK's sustained profitability miracle is likely to face unprecedented challenges this year.

But in reality, even with increased investment in C-end and B-end businesses, iFLYTEK will find it difficult to quickly eliminate the generational gap when facing more dynamic competitors like DeepSeek, Kimi, Zhipu, and Minimax. Even if the generational gap is eliminated, its competitiveness in price wars with AI intelligent agents like WorkBuddy, Doubao Work, and Feishu remains uncertain. This is a difficult path.

Having a solid foundation is iFLYTEK's strength, but in the AI era, it has also become iFLYTEK's comfort zone, limiting its revenue generation capabilities.