Image source: Generated by Unbounded AI

Use one keyword to summarize the AI ​​industry in 2024, and it is none other than burning money.

According to statistics, the total advertising expenditure for domestic AI native applications in 2024 will be 2.45 billion yuan.

From crazy purchases of GPUs to crazy investment in streaming, burning hundreds of millions of them every month, this AI money-burning war is comparable to the thousands of group wars and taxi wars in the Internet era.

But when it comes to user retention and commercial revenue, many AI companies may be silent. Even though it is as powerful as Byte, its Doubao AI investment has become the second in the world through traffic. It has recently been revealed that user retention is less than 30%, and users use it for no more than 10 minutes a day. The test of the advertising model also caused users to complain, "It feels itch without advertising." ".

On the contrary, some products in the non-chatbot mainstream field, such as the AI ​​love input method, WPS’s AI office software, and Baidu’s library business, have recently made headlines due to their commercial performance: "lovekey keyboard" "Cumulative revenue is nearly 30 million; wps AI Q3 commercial revenue is 700 million yuan; Baidu Wenku AI paying users exceed 40 million. Based on the 99 yuan/year subscription fee, the annual revenue should be in the order of 4 to 5 billion yuan. This number should be the envy of many AI practitioners.

If Doubao is the king of AI growth in 2024, some products (such as Wenku) that have been ignored by many AI lists may have actually become the invisible revenue leaders in the AI ​​industry.

How long can we burn money?

The war of big models in the past two years has swept almost all investment institutions, from VCs and large company venture capital to national industry funds and bosses with free wealth. It can be said that the easiest to get on the market Financing has been almost divided up.

Ironically, after two years of financing war, what frequently made headlines was not the product features that made people exclaim “Wow”, but the staggering marketing costs and investor anxiety. For a while, the investment circle even There is a popular joke, "Last year I invested in AI, this year I will invest in my resume." There are also rumors that an investor in a celebrity AI startup is eager to resell his equity.

To be honest, AI itself is a money-burning game. Some organizations predict that the cost of large model training will be close to US$5-10 billion in the next two years. Giants such as Meta, Google, and Microsoft may increase the cost of large model development to the order of US$50 billion.

After burning the GPU, you still have to burn the users. According to AppGrowing statistics, as of November last year, the total investment amount of ten domestic AI native applications was 1.5 billion yuan. The star chatbot products Kimi and Doubao were the most popular, with 540 million yuan and 400 million yuan invested respectively.

KIMI bean bag delivery cost comparison chart

Looking over a longer period of time, the bean bag launch is more intense. From April to May last year, Doubao's advertising amount was expected to be more than 15 million, but starting from June, Doubao launched multiple rounds of large-scale advertising, with the monthly advertising amount reaching 100 million.

However, as powerful as Byte, it’s time to settle the score.. According to foreign media reports, OpenAI, the company behind ChatGPT, may have total operating costs of US$8.5 billion last year, or face a huge loss of US$5 billion. Doubao benchmarks against ChatGPT. Even if the cost is not as exaggerated as ChatGPT, it must be an astronomical figure.

A clearer signal is that 36kr recently disclosed that Byte management judged that AI dialogue products may be just the "intermediate state" of AI products, so it raised Jimeng's product priority and tried to use new products. Path to create “Douyin” in the AI ​​era. The implication is fascinating.

When investors run out of patience and landlords have no food left, where will the money come from next? There may not be another player like OpenAI, which is burning money and throwing out Wang Zhao1 in exchange for US$6.6 billion in financing.

When burning money cannot create a future, how to quickly implement and break through the closed business loop has become a must-answer question for every industry player.

Why can you make money?

Both Doubao and KIMI are essentially chatbot (AI conversation) products pioneered by ChatGPT. Even ChatGPT, the originator, currently has no clear profit prospects.

If you look at users voting with their feet, emotional AI applications may be the real need. I recently read an article that the monthly revenue of several domestic leading AI love input methods has reached tens of millions.

The AI ​​Love Input Method actually helps users generate online replies to ambiguous partners and lovers. Paste the other person's comments into the dialog box, select the desired reply style, and the reply will be automatically generated and sent in one click.

Many products have also added the functions of "reply rewriting", "opening line generation", "invitation speech generation" and "confession speech generation", closing the "sultry" scene.

Productivity tools may be another AI-demanding scenario. Reports from Microsoft and LinkedIn found that 75% of the world's "knowledge workers" (people sitting in front of computers) use AI applications, and improving work efficiency is their primary need. Among them, PPT, which is most commonly used by workers, is also the scene with the highest monetization efficiency, attracting a number of new and old players such as domestic wps, AIPPT, and Baidu Wenku to enter the game.

Last year, the smart PPT market also experienced an interesting reversal. Yuehu big data showed that Wenku actually won 80% of the PK with the old office software WPS.

Comparing the difference between Wenku and wps, it may be the definition of "service".

For AI products, what is a service?

It is to help people complete tasks, not to output answers.

It is end-to-end delivery, not single link optimization.

From this perspective, Wenku seems to have considered all aspects of a "practical AI". After all, it includes searching for templates, generating PPT, changing words, changing pictures, changing the number of pages, and generating speech drafts... …finally deliver a usable PPT.

The most important thing is toWith AI capabilities, the collection is completed in one window. WPS only provides PPT templates.

High market share brings high ROI, which seems to be very significant. According to technology V Pan Luan revealed in a podcast, the revenue of Baidu’s entire library product line is similar to that of WPS, and the net profit is twice that of WPS. The authenticity of the data is open to question, but it generally reflects the trend comparison.

No wonder some big AI Vs say that Baidu Wenku is said to be the number one AI product in China in terms of revenue.

What kind of AI products have a commercial future?

In the past year, too much limelight has been given to various conversational applications following ChatGPT, from KIMI to Beanbao... These gimmicks have obscured the real question we should be discussing: what kind of AI is it? Does the product have a commercial future?

From Geek Park’s recent interview with the person in charge of Baidu Library, we can see an idea: don’t do links, do services.

What does it mean?

Behind AI conversational search is to accurately determine user needs and complete end-to-end tasks. And user needs are by no means limited to searching for templates and information:

Searching for a topic, "Analysis of the Growth Trend of New Energy Vehicles in China", is not essentially about searching for information or templates, but about writing a deliverable and of a certain quality. Standard industry research report PPT. Search for a paper, "The role of epigenetic regulation of autophagy in acute kidney injury caused by sepsis." What I want is not to download the data, but to sort out, modify and output my own report. When searching for a product, what users need is not product information, but direct ordering and purchase after comprehensive comparison.

Wenku AI takes advantage of its advantages in professional literature search and uses one entrance to solve the entire content creation process from searching for answers, outputting outlines, writing articles, to editing, polishing, revising, and finalizing.

There are two interesting comparisons. The recently popular computer version of Doubao seems to integrate an AI browser, article editor and AI generation function. However, the actual user creation process still requires frequent switching between many web pages and chat windows.

Perplexity, the foreign AI search king, also quietly tried to connect to e-commerce services a while ago. Users can place orders directly after searching for products - isn't this also a kind of all-in-one AI pragmatism?

Breaking the chatbot model dominated by ChatGPT is beginning to show results. Robin Li said at Baidu’s 2024 Q3 financial report that Wenku’s Q3 subscription revenue increased by 23%. In the same period, the investment cost was only 1/4 of Doubao. The person in charge of Wenku calculated in a public interview that after the AI ​​transformation, Wenku has gone from a sales volume of 2 to 3 billion to a market of 10 billion.

In 2025, the elimination round of AI applications has begun. Many industry insiders predict that the window for large-scale investment in AI application tracks will close in the first half of 2025. Those companies that cannot get users to pay the bills and cannot settle the accounts may not be able to avoid the fate of playing cards.

And those who remainThose who come are the long-term players who truly have a commercial future and can harvest the sweetest “fruits”.