Over the past year, iFLYTEK's inherent AI capabilities have been repeatedly interpreted and explored by the market.

The super wave brought by OpenAI has synchronized the ascent of all related industrial clusters, and iFLYTEK has occupied an advantageous niche in this wave, becoming one of the few core targets with large model technology in China's A-share market.

While everyone is focusing more on model iteration, few observe another hidden trend in iFLYTEK's financial reports: the positive impact of large model upgrades on iFLYTEK's various businesses, which is beginning to emerge in the financial reports.

On July 6, iFLYTEK released its 2024 semi-annual performance forecast. According to the forecast, the company's operating revenue and gross margin are expected to increase by 15% to 20% in the first half of 2024, with the gross margin growth rate higher than the revenue growth rate.

With the growth of both revenue and gross margin, how should we understand iFLYTEK's value in the era of large models?

/ 01 / High-quality growth behind proactive adjustments

Based on iFLYTEK's 2023 performance and this performance forecast, the company's operating revenue is expected to reach 9.02 billion to 9.41 billion yuan, and its gross margin is expected to reach 3.62 billion to 3.776 billion yuan in the first half of 2024.

Upon closer examination of this performance forecast, it is not difficult to find that iFLYTEK's profitability is improving.

In terms of profits, iFLYTEK's net profit attributable to shareholders was a loss of 300 million yuan in the first quarter of 2024, with a deduction of non-recurring gains and losses resulting in a loss of 440 million yuan. In the second quarter of 2024, the net profit attributable to shareholders is expected to be a loss of 160 million to 80 million yuan, with the loss amount narrowing sequentially despite continued investments in large models. Regarding the deduction of non-recurring gains and losses, the second quarter of 2024 is expected to be a loss of 60 million yuan to a profit of 20 million yuan, a significant improvement from the 440 million yuan loss in the first quarter of 2024.

In terms of gross margin, the company's gross margin growth rate was 14.7% in the first quarter, lower than the 26.2% revenue growth rate. In the second quarter, the gross margin growth rate exceeded the revenue growth rate. Based on the forecast performance, iFLYTEK's gross margin in the second quarter is expected to increase by at least 2 percentage points sequentially.

Various signs indicate that iFLYTEK is entering a round of high-quality growth, which is attributable to the company's proactive adjustment of its revenue structure.

Starting last year, iFLYTEK actively adjusted its business structure, shifting its business model from a focus on G, B, and C segments to a stronger emphasis on the C-end.

In 2023, iFLYTEK's business structure was further improved. In particular, the GMV of C-end hardware empowered by iFLYTEK Spark grew by over 84%, and the revenue of the open platform increased by over 40%. The open platform and consumer business contributed 6.19 billion yuan in revenue, a year-on-year increase of 33.4%, becoming the company's largest business segment.

The expansion of ToC business and the narrowing of ToG business have played a positive role in improving iFLYTEK's performance.

On the one hand, compared to G-end business, C-end business grows faster with stronger certainty in revenue growth. This trend continued into the first quarter. In the first quarter, revenue from the consumer and open platform business grew by 63%, and revenue from the learning machine business grew by 99%, driving the company's performance growth.

On the other hand, considering the current financial pressure on the G-end, appropriately abandoning some G-end business is also conducive to improving cash flow.

This has already been reflected in last year's annual report and this half-year report forecast.

As of the end of 2023, the company's net operating cash flow was 350 million yuan, with over 1.5 billion yuan in the fourth quarter alone. In the first half of this year, iFLYTEK's total sales collection amounted to approximately 9 billion yuan, an increase of approximately 1.49 billion yuan over the same period last year.

It can be seen that the explosion of C-end business has supported the success of iFLYTEK's business adjustment, which is particularly valuable today when large models struggle to find viable applications.

/ 02 / The dividend of large model technology is spilling over

The rise of C-end business is inseparable from the continuous iteration of large models.

On June 27, iFLYTEK released Spark Large Model V4.0, which has significantly improved its capabilities. In terms of base capabilities, Spark V4.0 has fully benchmarked GPT-4 Turbo, surpassing it in 8 out of 10 English evaluations (such as HumanEval, WinoGrande, GPQA) and 2 out of 2 Chinese evaluations (C-Eval, CMMLU).

The upgraded Spark V4.0 has also driven the capability upgrade of hardware products. Taking learning machines as an example, based on the improved language understanding, multi-turn interaction, and image-text recognition capabilities of Spark V4.0, the newly upgraded iFLYTEK AI learning machine can achieve ultra-humanoid personalized auxiliary learning functions, realizing AI one-on-one question-answering tutoring.

The upgrade of hardware products has directly translated into sales growth. After the release of the Spark Large Model, products such as iFLYTEK's learning machines, smart office pads, translators, and smart hearing aids have all experienced high-quality growth. During this year's "618" e-commerce event, iFLYTEK's smart hardware sales increased by 70% year-on-year.

In addition to driving hardware sales growth, model iterations have given rise to new opportunities in the software field.

Concurrent with the release of Spark V4.0, the Spark APP and desktop version were upgraded, pioneering the launch of "Personal Space" to create an AI assistant that understands you better. In Personal Space, users can upload various materials related to their work, study, life, health, etc., forming an exclusive knowledge base for each individual. Combined with user personas, the large model generates more personalized content.

Thanks to its strong model capabilities, the Spark APP has maintained its position in the first tier of domestic general large model tools apps, despite far less market investment compared to other leading large model vendors. According to Qimai data, the download volume of the Spark APP on the Android platform has exceeded 142 million times, ranking first among domestic general large model tools apps.

Benefiting from its strong performance in both hardware and software, iFLYTEK is one of the few Chinese large model vendors to achieve a software and hardware layout. Currently, the Spark Large Model has integrated the entire iFLYTEK C-end software and hardware product ecosystem. This means that data interoperability and operational linkage between software and hardware enable files from iFLYTEK's smart office pads and smart recorders to be synchronized to the Spark Personal Space with one click.

The significance of constructing an integrated software and hardware ecosystem lies in solving the problem of large model application scenarios. From a development trend perspective, software and hardware integration has become an important trend in the large model industry, with hardware becoming the best entry point for large model applications.

Not long ago, Apple's cooperation with OpenAI was the best proof. At this year's WWDC conference, Apple officially announced its cooperation with OpenAI. At the same time, Apple AI-enhanced all native applications on Apple systems, such as cameras, calendars, browsers, and email.

In China, almost all model vendors are focusing on models and applications, with a severe lack of hardware layout. In the long run, this will constitute a differentiated advantage for iFLYTEK. Especially during the large-scale deployment of large models on hardware, iFLYTEK's software and hardware integration advantage will be more fully reflected.

/ 03 / Leading the last mile

Since developing large models, iFLYTEK has always placed utmost importance on application deployment. With overall capabilities already benchmarking GPT-4 Turbo, iFLYTEK is more concerned about how to deploy large models for the last mile.

Not long ago, Liu Qingfeng provided iFLYTEK's answer: building an enterprise intelligence platform to create exclusive assistants for each position, enabling AI to be truly utilized by enterprises and function as a workforce.

Specifically, Spark 4.0 possesses the capabilities of an intelligence body. When encountering professional scenarios in various industries, such as R&D experiments, offices, and industries, iFLYTEK can customize AI for each industry. The intelligence body only needs to access internal enterprise data, without spending too much computing power and time learning, to become an industry expert and company steward.

Currently, iFLYTEK is leading the competition in the last mile of large model deployment.

According to the "China Large Model Tender Project Monitoring Report," iFLYTEK led the industry in the number of large model-related tender projects won in the second quarter, with a total disclosed tender project amount of 43.774 million yuan. June saw the highest number of wins, leading other vendors.

Not only does iFLYTEK lead in quantity and scale, but it has also secured head customers in many niche segments such as operators and finance.

For example, in the operator sector, iFLYTEK jointly developed 5G AI business stenography and a new mobile app with China Mobile, which was officially launched this year. In the financial sector, iFLYTEK collaborated with the People's Insurance Company of China (PICC) to develop "PICC Smart Lingxi - A Large Model for Exclusive Q&A," the first billion-parameter large model in the insurance industry. In the automotive sector, iFLYTEK's partners include both domestic brands such as Chery, NIO, GAC Motor, and FAW Group, as well as international automotive brands like Volkswagen.

There are two reasons why iFLYTEK can lead the last mile of large model deployment:

First, before deploying in enterprises, iFLYTEK created typical application cases such as the Spark Business Opportunity Assistant and Spark Bid Evaluation Assistant to set examples for enterprise applications. Among them, the Spark Bid Evaluation Assistant achieves a 98% human-machine consistency rate in intelligent bid evaluation results and an over 80% abnormal bid detection rate, significantly improving bid evaluation efficiency and reducing procurement costs.

Second, iFLYTEK adheres to an independent and controllable technology roadmap. Liu Qingfeng introduced that "the Spark 4.0 model is trained based on our domestic computing power platform, with every line of code and every piece of data written and cleaned by ourselves." This means that the Spark 4.0 AI large model is an authentic "Made in China" product with fully independent and controllable technology.

Amid increasingly intense geopolitical tensions, model security is being raised to a higher level of importance. On June 25th, OpenAI sent an email to users announcing that it would terminate API services to China from July 9th this year.

In May this year, the U.S. House Foreign Affairs Committee successfully passed a bill aimed at strictly controlling AI technology exports. The bill advocates strengthening the power of the U.S. Department of Commerce to more easily implement export controls on AI models and further restrict cooperation between the United States and foreign entities in developing AI systems that may threaten national security.

It can be predicted that to reduce the uncertainty brought about by geopolitics, more and more domestic enterprise customers will choose large model service providers with independent and controllable technology. iFLYTEK will become the biggest beneficiary of this wave.

From integrating the C-end software and hardware product ecosystem to leading the last mile of large model B-end territory, we can relatively clearly see the future outline of iFLYTEK. As Matty Stepanek said, "Although the future sounds far away, it has already begun."