AI Boosts Advertising Conversion and Reshapes the Cloud Market (Baidu 3Q23 Conference Call Summary)

The following are the key points of the 3Q23 earnings conference call for $Barnes(B.US)aidu.US. For a review of the earnings report, please refer to "Can AI Support Baidu's 'New Story'?".

1. Management Report

1) Ernie Bot

On August 31st, it was approved for public access, and on October 17th, Baidu World Conference launched Ernie Bot 4.0. We believe that Ernie Bot 4.0 has reached the level of GPT-4 and has begun to build an ecosystem.

(1) To C Applications

In this quarter, we started testing new features that directly generate answers to search queries, supplementing traditional search functions. Starting from November 1st, Ernie Bot 4.0 will be open to the public with a subscription fee of $8 per month.

(2) To B Applications

We launched GBI, which simplifies data analysis through natural language interaction and promotes fast decision-making in business operations.

Comate (AI coding assistant) and Ruliu (enterprise IM collaboration) focus on improving productivity. Many companies in traditional industries and public sectors have started using them and have shown strong interest.

We authorize companies to use Ernie Bot to build their own AI solutions through APIs, which will also promote the development of the Ernie Bot. Currently, more than 10,000 companies use Ernie Bot through APIs every month.

In early November, external queries increased by more than 50% compared to the same period in October. We are also actively attracting developers to connect their information and services with Ernie Bot through plugins.

2. Mobile Advertising

(1) Advertising revenue grew by 5%, benefiting from the recovery of the medical and tourism industries. This quarter, we continued to use Gen AI to improve conversion efficiency.

(2) Starting in September, advertising clients can use our new marketing platform, which supports natural language input and multi-round conversations, helping advertisers express their needs more comprehensively and enabling us to develop more effective marketing campaigns and strategies for them.

(3) We launched AI Digital Person to help businesses with live streaming, reducing barriers and costs.

(4) Outlook: Future advertising growth will exceed GDP growth. AI-native marketing products may bring more opportunities than traditional search.

3. Cloud Intelligence

(1) In the third quarter, non-GAAP operating profit of Cloud Intelligence remained positive. More and more companies are using the Ernie Bot API to develop their own native AI applications and solutions.

(2) By quickly identifying and handling GPU failures through technology, training costs have been significantly reduced, and currently 98% of training time is effective training.

(3) In the third quarter, there was a 2% YoY decline, mainly due to weak demand for smart transportation projects. In the fourth quarter, AI demand is expected to rebound, achieving positive growth. Moreover, last year's smart transportation projects had a low base.

4. Autonomous Driving

(1) The goal remains unchanged, which is to achieve UE breakeven for Robotaxi in different regions within a few years, and then achieve operational profitability. To this end, we are strategically focusing our resources on key areas. Wuhan is still our largest operating area, and we believe it is also the largest region globally to provide autonomous ride-hailing services, currently covering a population of approximately 2.7 million. In the third quarter, the proportion of fully autonomous orders in Wuhan exceeded 40% in dispatches, higher than the 35% in the second quarter.

Apollo Go started operations and continued to expand in late August. In the third quarter, Apollo Go had a volume of 821,000 orders, a YoY growth of 73%. As of the end of the third quarter, the cumulative volume has exceeded 4.1 million.

II. Analyst Q&A

Q1: Why is the advertising growth rate slower than the industry average? Outlook for the fourth quarter? Contribution of e-commerce advertising? Impact of AI on advertising prospects?

A: The main reason is the weakness in e-commerce advertising, which currently accounts for 10% of total advertising. We are building our own native e-commerce business and continuously improving the Baidu shopping experience, and currently, the native e-commerce business is growing strongly.

The impact of AI on marketing is evident as we are rebuilding the entire advertising platform. AI has shown its effectiveness in ad targeting and bidding. We expect these measures to bring in hundreds of millions of yuan in incremental revenue in the fourth quarter. In the future, we expect advertising growth to outpace GDP growth.

Q2: Guidance for cloud revenue growth in 2024?

A: Excluding ACE in the third quarter, the growth is still relatively robust. Cloud revenue is expected to recover in the fourth quarter, and this trend will continue. The main factors contributing to this are:

New opportunities brought by generative AI and large language models, which, although currently not generating significant revenue, are growing rapidly.

Customers in the online education and technology sectors are increasing their spending on cloud services.

Q3: Progress of Ernie Bot 4.0? Feedback on fees?

A: Compared to 3.5 and other LLMs in the market, the replies generated by Ernie Bot 4.0 are more organized and clear in terms of excel coding. Many companies have already started testing 4.0.

We are developing a chat box-type product that is expected to improve the conversion effectiveness of advertisements for advertisers, continuously enhance ROI, and also help drive our payment model from CPC to CPS.For different customer needs, we also provide different services. Some customers prefer to train their own models but are limited by GPU resources. We can help customers fine-tune the models of Ernie Bot and use them to meet the customized needs of various scenarios.

Q4: What is the revenue growth rate of Baidu Core in 2024? How to provide services to more advertisers?

A: The AI-enabled advertising platform has received positive feedback from customers. In the past few quarters, we have invested a lot of effort in using artificial intelligence and LLM to reshape the application system, and now we have an integrated marketing platform.

Advertisers can use it to generate creative ad materials and improve conversion rates.

Advertisers can engage in multi-round conversations to better understand their intentions and formulate campaign strategies, achieving higher ROI.

In addition, it greatly reduces the time required for advertisers to plan campaigns. Even experienced advertising managers used to spend hours developing ad strategies. Now this process only takes a few minutes.

Currently, we have already migrated thousands of advertisers to the new advertising marketing platform. Although this number is relatively small, the growth rate is very fast.

In addition, we will continue to use artificial intelligence to improve the bidding system. It has been observed that the conversion rate and click-through rate of ads have increased. On average, advertisers using this capability achieved high single-digit growth in 3Q ad conversion rates.

Q5: How do you view the competitive landscape of cloud services and the competition between internet companies and telecom operators? What advantages does Baidu have?

A: Traditional cloud services are slowing down, and AI is reshaping the cloud service and market competition landscape.

Currently, we see a rapid increase in interest from cloud service customers who want to leverage Baidu's AI technology to improve productivity.

Advantages of Baidu: (1) A unique four-layer AI architecture, each layer can be flexibly adjusted, and the different layers are compatible with each other, resulting in high efficiency in training and inference. (2) Currently, 98% of our training time is effective training, and we have sufficient GPU computing power for developing LLM.

Q6: What is the future development strategy for large-scale models? How will industry competition evolve?

A: In the next stage, we will adopt an application-driven approach, where AI-native applications will tell us what needs to be improved. However, currently, there are very few AI applications, so most of the API calls are from our internal applications.

Currently, there are 10,000 companies using Ernie Bot, and the API costs are calculated on a monthly basis. Compared to the March version of Ernie Bot, the marginal cost of the current version has decreased by 98%, and the QPS has increased by 5 times on the same computing power. The further cost reduction strengthens our advantage in large-scale models.

We believe that in the future, there will only be a few large-scale models in the market, and Baidu is one of them. More and more companies will use leading foundational large-scale models to create AI applications instead of building their own large-scale models, which will result in millions of early-stage AI-native applications.Q7: How does AI investment affect costs? What is the core OPM in the next few years?

A: Currently, our AI investment is mainly used for generative AI and large models. The investment in computing power is mainly reflected in our capital expenditure. In the past few quarters, we have invested a large amount of chips and resources to train LLM. Because depreciation is spread over multiple years, the impact of AI investment on profit margin is limited.

Meanwhile, our original mobile ecosystem continues to improve profit margins and generate significant cash flow.

In terms of cloud business, non-GAAP operating profit of Intelligent Cloud continues to be positive, and it will maintain a certain level of profitability in the future.

The new opportunities under AI and LLM will also bring incremental profits. Currently, we have concentrated resources and shifted towards AI-related businesses.

Q8: How does the further restriction of chip exports by the United States affect AI development?

A: The short-term impact of restricting chip exports on Baidu is limited because we have a large reserve of chips, which can help us continuously improve Ernie Bot in the next 1-2 years and support a large number of native AI applications.

However, in the long run, restricting the export of the most advanced chips will inevitably affect the pace of AI development in China, and we are also looking for alternative solutions.

Some of our peers sell computing power to start-ups and then invest in these start-ups to train basic models. Instead of doing this, we are trying to optimize all aspects from the infrastructure layer to the framework layer, and then to the model layer and the APP layer. By investing in the development of this end-to-end systematic approach, we can complete the training and inference process more efficiently and economically with the same computing power.

Risk Disclosure and Statement of this article: Dolphin Research Disclaimer and General Disclosure