"This is obviously false. Large models are an aggressive offensive and a necessary defensive strategy. I haven't heard of any reduction in model priority."

"The boss clearly stated that we will never abandon the table."

In response to rumors that "Baidu is likely to abandon general-purpose large models," two Baidu employees told "City Vision."

The rumors stemmed from a post stating, "Baidu is likely to abandon research and development of general-purpose large models (another case of getting up early but arriving late?). Li Yanhong recently clarified with the internal team that Baidu's main strategy is now focused on applications, and Baidu is exhausted at the model level."

After briefly gaining attention, the post eventually forced Baidu to respond. On September 9, Zhang Quanwen, head of the marketing department of ERNIE Bot, posted on WeChat Moments, "The so-called 'abandonment of general-purpose large model research and development' is purely a rumor! ERNIE Bot has just completed a comprehensive functional upgrade. We will continue to increase investment in research and development in the field of general-purpose large models."

Afterward, an informed source close to Baidu revealed to "City Vision" that this sparked widespread discussion within the company.

A Baidu employee candidly told "City Vision" that such rumors were not surprising: "In the SuperCLUE overall rankings (August 2024), ERNIE-4.0-Turbo-8K's score has fallen behind many other models. If we shift to application areas, the opportunities will be even slimmer."

In the past six months, market focus on AI large models has shifted from the technology itself to the practical implementation of intelligent applications, and AI expectations have also started to adjust. Therefore, making a choice between investing in basic large models and implementing AI applications has become an unavoidable challenge for all large model players.

01 The Pathway to the Birth of "Rumors" from an ROI Perspective

"Or put it this way, training the next generation of models is not Baidu's top priority," said the poster, implying that "abandonment" is not accurate.

His previous judgment on Baidu's large model direction was based on the following logic:

GPT-4 hasn't been caught up with yet, and a model at the GPT-5 level could cost up to $3 billion (rough estimate, with some fluctuation, which is almost Baidu's annual net profit). Moreover, you need to know how to do it, and consider whether the ROI (return on investment) will be positive. A failure would result in a huge waste of resources and opportunity costs, which is already a very real problem. A company of Baidu's size cannot sustain the dream of models anymore.

From an ROI perspective, focusing solely on the output side, i.e., the monetization of large models, Baidu is indeed facing certain commercialization pressures at present.

According to the second-quarter financial report, Baidu's largest source of revenue, online advertising, generated 19.2 billion yuan in revenue, a year-on-year decline of 2%. The reason for this performance in advertising is that Baidu's push for AI innovation in its search business has created considerable internal pressure. Currently, 18% of Baidu's search results are provided by generative AI, which has not yet been monetized.

An industry insider told "City Vision": "Baidu's investment in AI technology is paving the way for the future, but currently, there is no suitable monetization model for AI-generated content, which has affected advertisers' decision-making to some extent."

In terms of cloud services, Baidu Cloud generated 5.1 billion yuan in revenue for the quarter, of which 9% came from external customers' demand for large models and generative AI-related services. According to calculations, generative AI contributed approximately 459 million yuan to Baidu Cloud's revenue during this quarter, compared to 324 million yuan and 274 million yuan in the previous two quarters, respectively.

Although generative AI revenue is showing a growth trend, its contribution to overall performance is still limited and not yet sufficient to drive Baidu Cloud's overall growth curve.

The commercialization pressure on large companies like Baidu in large model monetization also comes from the price war initiated by domestic cloud vendors, either actively or passively.

Recently, Tencent launched its new large language model "Hunyuan Turbo," further intensifying the price war by offering a 50% price reduction. Prior to that, Baidu had also been continuously reducing the inference cost of its models to cope with market price pressures.

In May this year, Baidu Intelligent Cloud announced that its pre-installed services for the ERNIE-Speed, ERNIE-Lite, and ERNIE-Tiny series models would be freely accessible to customers; during the World Artificial Intelligence Conference (WAIC) in July, Baidu Intelligent Cloud announced significant price reductions for its flagship models ERNIE 4.0 and ERNIE 3.5, with ERNIE 4.0 Turbo fully open to enterprise customers at prices as low as 0.03 yuan per thousand tokens for input and 0.06 yuan per thousand tokens for output.

With fierce competition in the domestic large model market, the price war has led to a sharp compression in the cost of invoking large models, and the industry has fallen into the so-called "negative gross margin era." Both open-source and closed-source models are currently facing the same dilemma - direct profitability in the large model business is difficult to achieve.

According to Caijing Magazine, "The elimination round of the large model market has already begun." An executive in charge of large model business at a Chinese cloud vendor analyzed that the negative gross margin of large model invocations means that, in the short term, the more invocations there are, the greater the losses for cloud vendors.

Taking these factors into account, "It is somewhat unrealistic to achieve a positive ROI solely through the current level of commercial development of large models. Therefore, it is entirely possible that Baidu will redefine its strategy," said a Tencent AI algorithm expert to "City Vision."

02 Baidu's AI Path is 'Both And'"

Although Baidu denied the rumors of "abandoning research and development of general-purpose large models," it did not deny the priority given to application development. Compared to other major companies, Baidu attaches great importance to AI application development.

Throughout the CEO's speeches, he consistently emphasizes that "convoluting large models is meaningless, and there are greater opportunities in convoluting applications." At the recent WAIC, Li Yanhong even stopped emphasizing whether models should be open-source or closed-source, saying, "None of that matters. Basic models lacking application support are worthless, whether open-source or closed-source."

An industry insider close to Baidu revealed to "City Vision": "Li Yanhong's intention is to encourage entrepreneurs to develop more AI applications based on the ERNIE large model rather than waste resources on too many basic large models."

"The idea is right, but the problem is how to implement it. Baidu itself is engaged in basic large models, so why should others listen to you?" He added further.

This issue has also been discussed internally at Baidu, with some employees expressing their hope that Robin (Li Yanhong) would refrain from making statements in public that are prone to ridicule. "Only Baidu's large models are not a waste; other companies doing the same are wasting resources. As an industry leader, this attitude is indeed lacking in vision," one employee said.

From the current market perspective, Baidu's ideal goal in large models and applications is actually a two-pronged approach.

In the area of applications, which has always been emphasized, Baidu has launched three AI development tools: AgentBuilder for intelligent agents, AppBuilder for AI-native applications, and ModelBuilder for model customization.

Intelligent agents are the most promising direction for AI applications at Baidu. At the 2024 Baidu AI Developer Conference, Li Yanhong revealed that 10,000 Baidu customers have successfully launched intelligent agents, covering more than 30 industries such as education and training, real estate and home furnishing, machinery and equipment, and business services.

In a recent internal sharing session, Li Yanhong also reiterated the importance of intelligent agents, arguing that their low threshold is a direct approach to application development.

However, there is no consensus within the company on the direction of intelligent agents. According to a source close to Baidu, some employees privately expressed dissatisfaction during the Baidu AI Developer Conference.

One employee said that the so-called "intelligent agents" showcased on the ERNIE Intelligent Agent Platform are mostly toys, with only a handful of projects used for demonstration having any practical value. These "toys" have little real value, even when distributed through search engines. Truly meaningful intelligent agents should provide high-quality services or have private domain data that is not available on the public network. To call these "toys" AI-native applications would be self-deception.

Li Yanhong may also recognize the shortcomings of intelligent agents. "Many people are optimistic about the development direction of intelligent agents, but so far, intelligent agents are not a consensus. Not many companies, like Baidu, regard intelligent agents as the most important strategy and development direction for large models," he said.

Apart from the product attributes of intelligent agents, external developers also maintain a certain degree of caution towards Baidu in AI application development.

Judging from the currently public cases, most users of Baidu's AI development platform are merchants providing lightweight applications to their consumers, while enterprise-level applications are relatively scarce. Although Li Yanhong has repeatedly promised that Baidu will not involve itself in application development, leaving enough space for entrepreneurs, the common practice of "pixel-level copying" in the mobile internet era has made developers more vigilant in the era of large models.

The founder of a vertical model startup once told Caixin that developing on the Baidu platform means exposing algorithms and data to Baidu. On the other hand, Baidu may not be able to provide sufficient interface support for mid-level developers, which means that developers' customer data may not be easily deployed on the Baidu platform.

03 Strategic Shift Amid the Bubble of Large Models

Now, after the hype has subsided, more people are starting to examine the actual benefits brought about by generative AI.

In a report released at the end of June, Goldman Sachs titled "Generative AI: High Costs, Low Returns," senior strategist Allison Nathan raised a crucial question: Are we investing too much in AI while reaping too few benefits?

The Goldman Sachs report mentions that generative AI technology is considered transformative for companies, industries, and society, prompting many large companies to plan to invest $1 trillion in AI-related endeavors such as data centers, chips, and power grids over the next few years.

This expenditure has been verified. A CoutueAI report also mentions that infrastructure construction is expected to cost approximately $1.2 trillion by 2030, primarily for the procurement of approximately 25 million GPUs and other related expenses.

To achieve a positive ROI, with a 25% return rate, the expected revenue needs to reach $600 billion. Adding the $1.2 trillion investment, AI's total revenue must reach $1.8 trillion by 2030 to achieve a break-even point. However, according to Goldman Sachs, so far, these huge investments have not yielded significant results beyond slightly improving developers' work efficiency. Even NVIDIA, which has benefited the most, has seen its share price decline.

Of course, this is just a macro-level discussion. If we focus on specific large companies, the actual problem they face is that general-purpose large models are gradually evolving into a "super money-burning" game that has become a "rich man's game" accessible only to giants. Companies without sufficient funds will have to withdraw from the competition.

According to information publicly disclosed by the National Internet Information Office, since the first batch of registered AI large models in China was announced on August 31 last year, a total of 188 large models have been registered nationwide in just one year.

From a final outcome perspective, as underfunded players are eliminated or exhaust their resources in competition, only a handful of truly competitive large model companies will ultimately dominate the market.

"Essentially, basic large models are like CPUs, both are general-purpose technologies. As long as a few companies become strong enough, they can satisfy most market demands. Competition for general-purpose commodities is like this - the winner takes all, and only a few vendors will survive and thrive through market competition," said a computer industry insider.

This "money-burning war" has played out many times among internet giants. Taking the price war for large model invocations among cloud vendors as an example, cloud vendors are gambling that as the price of large model invocations drops by 90%, the number of invocations will grow exponentially over the next 1-2 years. In this gamble, a certain number of model vendors will inevitably collapse due to the price war, while survivors will take over the market and clean up the mess.

Obviously, for large companies to win the "money war" in basic large models, apart from having sufficient financial support in the short term, their long-term self-sustaining capability is crucial, and AI applications are the key direction to achieve this goal.

In fact, a closer look at the moves made by major vendors this year reveals that apart from their continued investment in basic large models, they are also constantly emphasizing the development of AI applications. AI applications are not only the future blood-making tool but also one of the important ways to ground large models and create real value.

Therefore, "Baidu's adoption of a dual-track strategy (simultaneously advancing both large models and AI applications) is a prudent approach to cope with the potential bubble in large models," said a source close to Baidu to "City Vision."

"The only problem is that the direction for the implementation of AI applications is still unclear. Many projects are still in the proof-of-concept stage and are still a considerable distance from true commercial applications."