Baidu's AI Struggles and the Future of Software Profit Margins

On August 23 Beijing time, Baidu Inc. released its second-quarter financial report on Thursday, showing a slight decline in revenue. Bloomberg pointed out that this financial report shows that Baidu has difficulty monetizing through artificial intelligence (AI), and the transition from search advertising to AI in its revenue-generating business has encountered difficulties.

Baidu's second-quarter revenue was 33.9 billion yuan, a year-on-year decline of 0.4%, falling short of the average analyst estimate of 34.1 billion yuan. By the close of US stocks on Thursday, Baidu's share price fell by 4.40%.

AI monetization difficulty

Bloomberg said that the poor performance highlights the challenges Baidu faces in converting its leadership position in the generative AI field into considerable income. With the help of advertising and cloud services, Baidu's Wenxin large language model has gradually begun to bring additional sales, but it has also been involved in an AI price war with companies such as Alibaba Group and Tencent Holdings. Baidu may need several years to comfortably reduce its reliance on advertising.

Baidu's business seems to be at a crossroads, Tianhao Capital analyst Hou Xiaotian wrote in a report before the release of Baidu's financial report, Its AI plan has not yet produced the expected results, and AI has not yet become Baidu's growth driver.

Baidu founder Li Yanhong has high hopes for building a Chinese version of ChatGPT, but he is facing a tough battle with other large technology companies and emerging startups. IDC estimates that last year, in China's $250 million generative AI market, Baidu's share was about one-fifth.

However, this leadership position is being rapidly eroded. Data from mobile application analysis company SensorTower shows that ByteDance's chatbot Doubao had more downloads than Baidu's Wenxin Yiyan last year and had more monthly active users on iOS in China.

Bloomberg Industry Research analysts Robert Lea and Jasmine Lyu said in a research report: Baidu's prospects still face huge challenges. As Tencent and Alibaba continue to narrow the gap, Baidu's AI business is expected to continue to lose money in the next three years. We expect Baidu's main cash-generating search engine business to continue to come under pressure under the increasingly fierce competition in the short-video field.

The deteriorating AI price war is likely to cause Baidu to lose more market share this year and prevent it from using its technical expertise to generate revenue and reverse its loss-making AI business. We expect Baidu's adjusted net profit to decline by 5%-10% this year. They said in the report.

AI leads to lower software profit margin

In addition to Baidu, US tech giants are also investing heavily in generative AI. Data from research firm Synergy Research Group shows that Amazon, Microsoft, and Google spent more than $48 billion in the second quarter, mainly on data centers.

Large tech companies' spending on generative AI is expected to reach a staggering $1 trillion. This is because they increasingly believe that generative AI will become the next major technological wave after cloud computing.

But how much return can such a huge AI investment bring? This week, analysts at Royal Bank of Canada Capital Markets gave a preliminary answer, not that high.

The result of generative AI is that the long-term gross profit margin of software will structurally decrease. They said in the research report. According to Royal Bank of Canada Capital Markets' data, when software shifts from on-premises (enterprises running software on their own computers) to cloud (running remotely on rented servers), the gross profit margin drops from 90% to 75%.

These analysts estimate that the shift from cloud computing to generative AI will further reduce the gross profit margin of software to about 60%.

In the software industry, the gross profit margin is usually around 90%. This sounds like a lot, which is why this industry is so attractive to investors and why the valuations of software companies are so high. The upfront cost of developing new software is high. But once it is developed, the cost of making a new version and distributing it to customers is almost zero. So the more software a company sells, the greater the profit.

High cost

So why might the profit of the software business decline in the upcoming AI era?

Generative AI has high development costs and high operating costs. AI models need to be trained, which includes buying extremely expensive GPUs from NVIDIA and then putting these AI chips into servers. These servers need special cooling and networking within large data centers. These facilities consume a lot of electricity and have high costs and require expensive upgrades.

This doesn't include the data cost for AI model training. Large tech companies and startups are trying to avoid paying most of these costs, but collecting and cleaning this data is still expensive.

Once the AI model is trained, they still need to run. This is the inference step, that is, the model will process new data or requests and infer useful information from them. This step also requires expensive chips and is a continuous expense.

This is different from the previous on-premises software business. Previously, almost every new sale was 100% profit. But whenever an AI customer uses a generative AI service, the provider incurs a large amount of cost. For example, industry analyst Dylan Patel estimated last year that ChatGPT's daily operating cost is $700,000.

Revenue compensates for the decline in profit margin

However, analysts at Royal Bank of Canada Capital Markets are not completely pessimistic.

They expect that generative AI will become so revolutionary that customers will spend more on new AI software. They estimate that this will make future software revenue twice or even three times the current level.

Analysts also explained that as the software market expands, even if the profit margin is lower, it may still create more profit dollars.

When the profit margin declines, profit dollars is a metric used by management and analysts to measure a company's absolute profit. For example, if a company has $100 million in revenue and a 10% profit margin, then its absolute profit is $10 million. If this company's revenue increases to $300 million, but the profit margin drops to 8%, then its profit is still $24 million, more than the previous profit.

A lthough we expect generative AI to bring pressure on the profit margin, we believe that in the post-generative AI era, the long-term gross profit will be higher. Royal Bank of Canada Capital Markets' analysts concluded.

However, there is an important assumption here, that is, whether generative AI can bring huge revenue growth. Otherwise, these huge AI investments may have quite bad economic benefits.