Introduction: With conversations, searches, and life services becoming fully AI-driven, how close is the 'Age of Intelligence?'

In 2026, the AI industry’s popularity persists.

Recently, the AI application sector in the A-share market has performed outstandingly, with several AI application stocks such as Kunlun Tech, BlueFocus, and iFLYTEK hitting their daily limits consecutively, becoming the market's focal point.

Earlier, the highly anticipated Minimax officially listed on the Hong Kong Stock Exchange, with its stock price soaring and market value exceeding 100 billion yuan.

MiniMax boasts a series of AI-native applications, including Hailuo AI, Xingye, and Talkie.

These signs are interpreted as precursors to an impending explosion in AI applications, prompting tech giants to gear up for an imminent AI application battle.

Alibaba has diversified its offerings, launching multiple consumer-end applications such as Qianwen, Lingguang, and Ant Afu. ByteDance has invested heavily to secure the Spring Festival Gala sponsorship for Doubao, which already leads in monthly active users (MAUs).

Tencent, while appearing low-key, quietly appointed former OpenAI core researcher Yao Shunyu as Chief AI Scientist in December last year, unifying leadership over the Large Language Model Department and Infrastructure Team.

Internet giants view AI as a disruptive super-entrance, while workers anxiously ask on Zhihu: 'Will AI replace humans?'

How close are we to an AI-driven world transformation?

I. Gatsby's World: The AI Application Battle Looms

Over the past year, the prosperity of the AI industry is undeniable.

The iteration speed of large models is accelerating.

According to Caijing Magazine, from January 1 to July 25, 2025, within 206 days, 11 tech companies from China and the US involved in model competition released or iterated at least 29 versions of large models, averaging one new model every 7.1 days.

Large models have shifted from monthly to weekly updates.

User demand has also exploded.

Doubao's MAUs have surpassed 200 million, extending its reach to lower-tier markets. Qianwen APP saw its MAUs rise to 30 million within 23 days of launch, with user growth nearly tenfold in just over a month.

Doubao mobile phones sold 30,000 units in one day, while Rokid Glasses sold 40,000 units in five days, with AI phones and AI glasses gaining popularity simultaneously.

Frost & Sullivan data predicts that by 2030, China's AI market size is expected to further increase to 993 billion yuan, with a compound annual growth rate of 35.5% from 2024 to 2030.

Major companies are increasing their AI spending.

According to LatePost, in 2025, Baidu plans to spend 30 to 50 billion yuan on AI; Tencent, 70 to 100 billion yuan.

Alibaba's management indicated at an earnings call that its previously proposed 380 billion yuan investment plan may be conservative, as demand currently exceeds supply. If market needs remain unmet, further investment increases may be considered.

AI companies are accelerating their listings.

Zhipu AI and MiniMax have successively listed on the Hong Kong Stock Exchange, providing relief to investors and excitement for industry analysts with two new case studies.

AI chip, server, and other supply chain companies are also following suit, with a flurry of IPO bells.

According to The Paper, in Shanghai alone, five AI companies have gone public in the past month.

Salaries for AI talent are soaring.

Million-yuan annual salaries are common, with some companies offering 'no upper limit' compensation packages at job fairs.

DingTalk directly posted recruitment ads at the gates of Tsinghua University, Peking University, and other top universities. Tencent, to poach talent, offered double salaries to ByteDance's AI researchers.

Zhang Di, known as the 'Father of Kling,' recently left Kuaishou and was reported to join Bilibili as head of technology.

According to the '2025 Mid-Year Talent Supply and Demand Insight Report' released by Liepin Big Data Research Institute, among the top 20 fastest-growing job positions in the first half of 2025, AI-related roles account for 60%, all with growth rates exceeding 30%.

The scene appears glamorous and booming, reminiscent of Gatsby's world.

However, behind this prosperity, some realities remain harsh.

Profitability challenges persist.

Prospectuses show that from 2022 to the first half of 2025, Zhipu AI accumulated losses exceeding 6.2 billion yuan; from 2023 to the first three quarters of 2025, MiniMax accumulated losses exceeding 8.7 billion yuan.

Major companies treat AI as a traffic acquisition tool, burning money for users: Qianwen APP launched Asia's largest database of primary and secondary school exam questions, freely accessible to users. Ant Afu quickly surpassed 30 million MAUs but announced a permanent non-commercialization policy.

It's not that companies want to be charitable, but Chinese consumers, spoiled by free internet services, have become picky. Baidu's ERNIE Bot initially followed ChatGPT's lead in charging users but ended up 'rising early yet arriving late.'

AI empowerment has not altered the internet competition landscape.

With major companies' efforts, nearly all internet entrances now feature embedded AI. Except for Pinduoduo, WeChat, Douyin, Taobao, and Baidu have all transformed into 'AI+.'

Additionally, Meituan launched the AI life assistant 'Xiaomei,' fully integrating it into core services like food delivery, travel, and hotel bookings on the Meituan App.

Didi recently announced a strategic partnership with Zhipu AI to explore the implementation of travel agent scenarios.

Ctrip also announced that its travel platform has evolved to the 'Intelligent Engine 3.0' stage, covering multiple dimensions such as AI-driven service enhancement, AI traffic allocation, AI intelligent shopping guides, and AI content generation.

In short, AI has intervened in nearly all internet services.

But if all applications incorporate AI, does that mean the competitive landscape remains unchanged as if no AI existed?

In a recent AI debate, investment master Michael Burry introduced a theory called the 'Buffett Escalator':

In the late 1960s, Warren Buffett ran a department store. When a competitor across the street installed an escalator to attract foot traffic, he felt pressured to follow suit.

However, this investment yielded no real benefits: the store's profit margins did not achieve long-term growth, nor did operational costs optimize, ultimately leaving both sides in their original market equilibrium.

In the future, AI may become like Buffett's escalator—not necessarily used, but indispensable.

This may explain why Alibaba launched Qianwen APP after Kuake.

Tencent and ByteDance, despite owning super-apps like WeChat and Douyin, still launched independent AI entrances, likely for the same reason.

It is generally believed that independent AI entrances can disrupt the original traffic distribution mechanism by changing interaction methods (e.g., Doubao mobile phones). However, with all internet giants now offering independent AI entrances, homogeneous competition has resurfaced, leaving no one daring to relax.

Qianwen APP maintains high-intensity promotion, with subway billboards and Bilibili recommendation zones as its battlegrounds. Doubao plans to leverage the Spring Festival Gala's popularity to maintain its lead.

The 2026 AI application battle may be as fierce as the 2025 food delivery war.

II. Are 'Shovel Sellers' More Profitable Than 'Miners?'

A crucial prerequisite for the AI application explosion is the rapid reduction in large model usage costs.

As early as 2024, the large model industry had already engaged in a price war, with DeepSeek-V2's input price dropping to 1 yuan per million tokens. Baidu Smart Cloud announced free usage for several ERNIE series models, while Alibaba Cloud's Qwen-Long saw a significant price reduction of up to 97%.

According to Dolphin Research, by the end of 2025, the unit cost of AI inference had dropped by over 90% in the past two years.

Large models serve as the infrastructure for AI applications, and the premise for any application explosion is that this infrastructure is affordable enough.

Much like how the short video (short video) industry rose during the mobile internet era, it owed much to the 'fee reduction and speed increase' initiatives by the three major telecom operators.

A 2014 survey report revealed that due to high data fees at the time, mobile users typically used mobile data for no more than three hours daily. Additionally, many users actively turned off mobile data when not in use to prevent apps from consuming background data.

By 2019, mobile internet data fees had dropped by over 90% compared to 2014, enabling apps like Douyin and Kuaishou to gradually become national favorites.

However, one key difference is that while data fees dropped significantly, the combined daily profits of the three major telecom operators still exceeded 400 million yuan. In contrast, large model companies, after engaging in price wars, have fallen into sustained losses.

The reason may lie in the fact that while telecom operators can build base stations once and for all, with marginal costs for data usage nearly zero, the inference costs for large models are almost linearly related to user usage: each user request consumes computing resources (e.g., GPUs/TPUs), and the inference cost for processing one token remains basically fixed. The more users and requests, the higher the total cost.

Thus, we see an addictive increase in demand for computational power from large models.

Zhipu's computational service fees in the first half of 2025 reached 1.1 billion yuan, accounting for 71.8% of its R&D expenditure; MiniMax spent $140 million (approximately 987 million yuan) on R&D computational power in the first nine months of 2025, accounting for 79%.

Other vendors may differ in proportions but are unlikely to vary significantly.

The result is that money is flowing to computational power service providers.

NVIDIA's gross margin nears 75%, with its market value increasingly distancing itself from the runner-up. Alibaba Cloud Intelligence Group reported revenue of 39.824 billion yuan in the third quarter of 2025, a year-on-year increase of 34.50%.

In contrast, the self-sustaining capability of large models as standalone projects is pitifully weak.

Zhipu AI's prospectus shows its total revenue in 2024 was 312 million yuan, ranking second among China's top language large model vendors and first among independent large model vendors.

Major companies can afford to lose money, but startups rely solely on investor funding. Tianyancha APP reveals that MiniMax completed seven rounds of financing before going public, with each round exceeding $250 million.

Zhipu AI is even more extreme, having completed 18 rounds of financing before its listing.

Similar situations occur in the new energy vehicle (NEV) sector.

According to LatePost, in the first 11 months of 2025, the average selling price of domestic NEVs was 169,000 yuan, reverting to 2022 levels. 'Selling one, losing one' has become the norm for many automakers.

In the same year, CATL's net profit exceeded the combined total of Seres, BYD, Geely, and Great Wall Motor.

Similarly, the enthusiasm of NEV players relies on investor support.

According to Auto-First's incomplete statistics, in 2025, Chinese automotive companies (including parts suppliers) completed over 100 billion yuan in equity financing.

The similarity between large models and NEVs lies in the fact that while few players are currently making substantial profits, there is widespread belief that both sectors have promising futures.

Indeed, glimmers of hope are visible on the horizon. The NEV industry is on the cusp of breakthroughs in solid-state batteries and high-level autonomous driving, drawing significant attention from the Ministry of Industry and Information Technology.

Meanwhile, the AI industry is poised for an application explosion, a consensus among major companies. Revenue growth covering model costs may no longer be far off.

For numerous startups, a brighter tomorrow may be on the horizon. However, whether dawn can break into reality depends not on the grandeur of visions but on who can survive today.

After all, in the long river of technological revolutions, only those who endure can define the future.

Disclaimer: This article is based on legally disclosed company information and publicly available data, offering commentary. However, the author does not guarantee the completeness or timeliness of this information.

Additionally: The stock market involves risks; enter with caution. This article does not constitute investment advice; investment decisions require personal discretion.