Source: Houchangcun, Cover Image: Doubao

In recent years, AI has become a high-frequency buzzword in public discussions. If 2023 can be called the "Year of AI Technology Frenzy," then 2024 can be regarded as the "Year of Deepening and Expanding AI Applications."

As 2024 draws to a close, with the rapid advancement of cutting-edge technologies such as large models, humanoid robots, and intelligent connected vehicles, AI is leading a new wave of technological and industrial revolution at an unprecedented speed and scale. Within the industry, anxiety about super apps is growing. How much money does it take to run large models? When will super apps emerge? These are questions that every practitioner in the AI industry wants to know.

01

The First Half of AI Technology Frenzy

Before 2023, AI technology underwent years of steady development and accumulation. From the introduction of the concept of AI in the 1950s, to the golden era of research in the 1960s, to the new advancements in expert systems and artificial neural networks in the 1980s, AI experienced a resurgence. Then, in 2012, with the advent of the big data era in the 21st century, breakthroughs in deep learning algorithms occurred.

At the end of 2022, ChatGPT emerged, sparking global attention and discussion.

Whether in image recognition, speech recognition, or natural language processing, the accuracy and efficiency of AI models have been significantly improved.

Moreover, AI technology has been widely applied in various fields such as autonomous driving, medical diagnosis, and financial risk control. The enthusiasm for exploring AI in the tech world continues to grow, with numerous enterprises entering the AI field, making the first half of the AI competition fiercely intense.

Today, AI is a red ocean. Besides established players like Baidu's ERNIE Bot, Alibaba's Tongyi Qianwen, ByteDance's Doubao, and SenseTime's RISE, newcomers like Zhipu AI, BaiChuan Intelligence, Zero-One Everything, Dark Side of the Moon, Minimax, and Stepwise Stars are also breaking into the scene. Various "large-scale deep learning models" are competing on the same stage, creating an unprecedented "Hundred Models War" that is intensifying. However, capital investment and corporate losses resemble a bottomless "black hole."

SenseTime, known as the "leader of the AI Four Little Dragons," has been losing money for years since its IPO in 2020. Data shows that from 2018 to 2023, SenseTime's cumulative net loss amounted to RMB 50 billion.

Kunlun Tech, which began deploying large models in 2023, reported a cumulative revenue of RMB 3.828 billion in the first three quarters of 2024, but a net loss of RMB 627 million.

Fourth Paradigm, which mainly provides platform-centric AI solutions, listed on the Hong Kong Stock Exchange on September 28, 2023, and has yet to achieve profitability. In the two and a half years from 2022 to the first half of 2024, it accumulated a loss of RMB 2.743 billion, with a loss of up to RMB 1.802 billion in 2021.

It is understood that OpenAI, the well-known developer of ChatGPT, also faces long-term huge losses.

OpenAI has raised approximately RMB 138.5 billion in funding (based on disclosed data). However, financial reports show that OpenAI has invested heavily in AI technology development and operations, with an estimated financial deficit of USD 5 billion this year and cumulative losses of USD 44 billion from 2023 to 2028.

With industry giants facing such challenges, the survival dilemma of other AI enterprises is even more evident. Many enterprises are caught between the difficulty of applying AI in practical scenarios and monetizing it, relying solely on financing to survive and urgently seeking new avenues.

It can be seen that although AI technology has radically changed people's lifestyles, work methods, and ways of thinking, the AI era is still undergoing a transition from technological frenzy to the emergence of AI super apps.

02

The Value of AI Super Apps

What are AI super apps? When will super apps appear? What is the value of AI super apps? These questions deserve in-depth consideration by all AI practitioners.

Houchangcun believes that super apps in the AI era will definitely emerge, but the timing is not yet right.

Secondly, before the advent of super apps, there must first be super useful apps. Only through rapid iterations can revolutionary applications be produced.

Baidu founder Robin Li once said in an interview with the media that models themselves do not generate direct value. Only by developing various applications on top of models and finding the so-called PMF in various scenarios can true value be generated.

AI applications can be mainly classified into the following categories:

1. Embodied AI applications. These applications usually have a physical form, can perceive, interact, and actively enter the "real world," such as autonomous vehicles and humanoid robots.

2. AI software: These applications mainly exist in software form, utilizing large language models, deep learning, and other technologies to provide users with functions such as office assistance, graphic design, video generation, and intelligent services in finance, education, and other fields. AI assistants are currently the fastest-implemented AI super apps and a direct product of the "Hundred Models War." ByteDance's Doubao, Baidu's Comate, Tencent's Yuanbao, iFLYTEK's Spark, Kuaishou's Keling AI, and Dark Side of the Moon's KimiChat are currently the main AI assistants in China, with distinct advantages in differentiation.

3. AI consumer electronics: These applications usually involve upgrading traditional electronic devices with AI, integrating with the metaverse, or developing new demands based on new technologies, such as AI PCs, AI mobile phones, XR devices, and brain-computer interface technology.

Large models have not only sparked an AI technological revolution but have also reshaped our production and lifestyle to a certain extent, enhancing productivity.

AI large models not only have high user stickiness and a wide coverage of users but also have a huge user base. It is understood that Baidu's ERNIE Bot has reached 430 million users, with over 1.5 billion daily calls to the ERNIE large model.

It is evident that AI super apps built on basic large models will generate immeasurable practical value, empowering various industries.

In the second half of the AI era, when the capabilities of basic models are ready, we will hopefully witness the "iPhone moment" for AI applications.

03

Where Will the Future Wind Blow in the Second Half of the AI Era?

From the frenzy of technological "stacking" to the transformation towards super apps, many AI enterprises are seeking their own paths to breakthrough.

In the "first half" of AI, after more than half a century of algorithmic advancements, GPUs replaced CPUs as the carriers of AI computing power, significantly expanding chip demand and giving rise to the "Hundred Models War." As the second half arrives and the generative AI revolution unfolds, where will AI go from here?

Nvidia founder Jen-Hsun Huang believes that the next wave of AI is "embodied AI," which refers to entities with AI capabilities that can truly understand, reason, and interact with the physical world. It can be a program, a system, or a robot, highly automated, intelligent, and integrated, analogous to the difference between general intelligence and mechanical intelligence.

Traditional robots, such as industrial robots, service robots, and special robots, have relatively mature technologies. For example, in the industrial field, there are numerous commercial cases of mobile handling, automated logistics, industrial manufacturing, and automatic power grid inspection. In the service sector, there are applications like household sweeping robots and hotel automatic delivery robots.

However, general-purpose robots, represented by Tesla's Optimus, are no longer limited to specific application fields, posing higher requirements on the algorithms and computing power of large models.

On the other hand, hardware innovation is the unsung hero behind AI large models. From CPUs to GPUs, to dedicated AI chips, from SSDs to NVMe, from distributed training to Chiplet technology, and to the construction of a software and hardware co-ecosystem, the synergy between hardware and AI large models provides a solid foundation for continuous innovation in AI technology and the expansion of application scenarios, paving the way for the next step in AI development and reshaping the future.

04

Conclusion

Although the road ahead is not smooth, and AI still faces ethical dilemmas, algorithmic bias, responsibility attribution, and energy consumption issues, it cannot be denied that as the core driving force behind a new wave of industrial transformation, AI has been widely applied in many fields and is steadily becoming the core force driving the fourth great technological revolution in human history. This revolution is not only underway but its profound influence and enduring momentum for change will undoubtedly create a new powerful engine, sparking a new wave of industries globally and shaping a new era that is smarter, more efficient, and full of endless possibilities.