Xiaomi can't afford the big AI model

Source: Alphabet List, Author: Zhao Jinjie, Editor: Yan Fei

While the car-making business is still under pressure, Xiaomi has plunged into another money-burning game.

Following the announcement in March this year that it is exploring AI large models, in the first-quarter earnings conference call on the evening of May 24, **Xiaomi President Lu Weibing said that the company has formally established an AI laboratory large model team in April. There are more than 1,200 related personnel in the field. **

However, Xiaomi does not want to be the next OpenAI.

Lu Weibing said that **Xiaomi will not study a general-purpose large language model. **In contrast, in order to catch up with OpenAI, a number of domestic technology manufacturers such as Baidu and Ali have released self-developed general-purpose large models.

**However, Xiaomi's consumer electronics business, which researches and develops blood transfusions for cutting-edge technologies, is still slowing down in the downward channel, and has not yet ushered in a rebound moment. **

In the first quarter of this year, Xiaomi’s revenue was 59.5 billion yuan, a year-on-year decrease of 18.9%; its adjusted net profit was 3.2 billion yuan, a year-on-year increase of 13.1%.

Specific to each business segment, smartphone revenue decreased by 23.6% year-on-year to 35 billion yuan; IoT (Internet of Things) and consumer product revenue decreased by 13.6% year-on-year to 16.8 billion yuan; Internet service revenue decreased by 1.2% year-on-year to 7 billion yuan; others Revenue decreased by 36.9% year-on-year to 600 million yuan.

At a time when the income of the four major businesses has declined, **Xiaomi can still achieve a year-on-year increase in net profit, mainly due to the strategy of reducing costs and increasing efficiency in the past quarter. **

In the first quarter, Xiaomi reduced its staff by 134 and greatly optimized costs and expenses. Financial report data show that Xiaomi’s cost of sales decreased by 21% year-on-year to 47.9 billion yuan; sales and promotion expenses decreased by 21.9% year-on-year to 4.1 billion yuan; administrative expenses decreased by 8.8% year-on-year to 1.1 billion yuan.

Lu Weibing said that by reducing costs and improving efficiency, we will continue to see further room for growth this year.

However, as the business growth engines have slowed down one after another, how Xiaomi, dominated by cost reduction and efficiency increase, can occupy a place in the competition of large-scale models that are extremely expensive to burn, will become another problem for Lei Jun besides car manufacturing.

01

The mobile phone business, the core of Xiaomi’s revenue growth engine, accounted for 58.8% of revenue in the first quarter, but its sales continued to decline, down 21.1% year-on-year to 30.4 million units.

Xiaomi is not the only mobile phone manufacturer that has experienced a decline in sales. The global smartphone market continues to be under pressure. According to Canalys data, in the first quarter of 2023, global smartphone shipments fell by 13.3% year-on-year, the lowest first-quarter shipments since 2014.

In mainland China, smartphone shipments fell 11% year-on-year to 67.6 million units, the lowest first quarter since 2013.

** This is the fifth consecutive quarter that the smartphone market has experienced double-digit declines in shipments. At present, domestic consumer demand is picking up and improving, but it has not yet brought a significant boost to the smartphone market. **

Canalys research analyst Zhong Xiaolei believes that although the total retail sales of social consumer goods in the first quarter increased by 5.8% year-on-year, communication consumer goods still declined, and more users maintained a rational attitude towards electronic consumer goods.

** Overseas revenue, which once supported half of Xiaomi's revenue, also suffered a decline in the first quarter. ** In the last quarter, Xiaomi’s overseas revenue was 27 billion yuan, a year-on-year decrease of 28%, and its revenue share fell to 45.4%.

Among them, Xiaomi's overseas Internet revenue achieved a growth of 16.4%. The bigger reason for the decline in revenue is obviously the unfavorable development of the hardware business.

Lu Weibing said that in the first quarter of 2023, India's smartphone market shipments fell by 20% year-on-year. India and other overseas markets accounted for the bulk of Xiaomi's 8 million fewer smartphone sales compared with the same period last year.

Xiaomi smartphones can no longer be sold, and the sales of IoT products, which were originally regarded as new growth poles, have also begun to come under pressure. **

The Canalys report shows that in the first quarter, the total shipments of desktops and notebooks in the global PC market fell by 33% to 54 million units, a double-digit decline for four consecutive quarters. Specific to the mainland China market, PC shipments fell 24% year-on-year to 8.9 million units.

Reflected in Xiaomi's financial report, its revenue including smart TVs and laptops will decrease by 24.2% from 6.2 billion yuan in the first quarter of 2022 to 4.7 billion yuan in the first quarter of 2023.

"Since the first quarter, China's overall consumer spending has been relatively strong, but it has been mainly concentrated in areas such as tourism and hotel accommodation, and has not been used to purchase new computers." Canalys analyst Xu Ying said, "At present, the Chinese economy has entered the recovery phase, but the PC market has yet to show clear signs of improvement."

02

In order to compete for limited replacement users in the stock market stage, top mobile phone manufacturers are considering AI large models as a new means of differentiated competition. **

In April, it was revealed that Samsung is considering changing the default search engine of its mobile phones and tablets from Google to Microsoft's new Bing, so as to bring a new experience to Samsung mobile phone users with the help of GPT-4's large-scale interactive experience.

Just as Samsung was still judging whether to change the door, Google's countermeasures also came. At the Google I/O conference in May, Google released four new-generation large language models PaLM 2 in one breath, which are Gecko, Otter, Bison and Unicorn according to the size of the parameters. (Unicorn).

The smallest "Gecko" model can be adapted to run on a mobile phone. According to Google CEO Pichai, not only is it fast enough, but it can also be used without an Internet connection.

After the release, Nvidia artificial intelligence scientist Jim Fan praised "Gecko" as "the most important highlight" in the PaLM 2 model, "the productivity improvement on the small screen is much more than on the large screen."

Apple also hopes to show AI technology to the outside world with the help of hardware carriers. In the first-quarter earnings call, Apple CEO Cook said, "We believe that artificial intelligence is a huge change, and we will continue to weave it into our products on a very deliberate basis."

A report from the "Wall Street Journal" mentioned that Apple has banned employees from using large-scale model products including ChatGPT and Copilot, and is starting to develop Siri's new AI function. The project is code-named Bobcat, and the new technology framework is called Siri Natural Language Generation. (Siri natural language generation) function.

Before many mobile phone manufacturers have shown their attention to the AI model, at the fourth quarter 2022 earnings conference call in March, when Lu Weibing responded to the current hot AI model, he only said that Xiaomi is "using multiple technology routes to explore in parallel." The landing of the AI model."

When Samsung, Apple, etc. stepped up the development or application of large-scale model products, Xiaomi also began to accelerate its layout, and formally established an AI large-scale model team in April.

Lu Weibing said that Xiaomi already has a very clear goal and strategy for AI large models: first, unlike OpenAI, Xiaomi will not study general-purpose large language models; second, the thinking point of Xiaomi's development of large models is how to integrate them with specific businesses Get up and turn it into business results.

** For Xiaomi, which is still in the process of reducing costs and increasing efficiency, Lu Weibing also hopes to further improve internal efficiency with the help of large models. **

03

The huge amount of money spent on developing a general-purpose large model may be a major reason for Xiaomi to bypass the self-developed large model.

According to OpenAI's recently released operating data, its losses doubled last year to around US$540 million, which was mainly used to support the operation of ChatGPT and poach people from Google.

In the past few years, OpenAI has burned billions of dollars in financing in order to train the GPT general-purpose large model. But to win the big model competition, the money is far from enough. According to the prediction of OpenAI CEO Sam Altman, OpenAI may need to try to raise as much as 100 billion US dollars in the next few years to develop sufficiently advanced general AI while maintaining the normal operation of the company.

**In addition to avoiding the development of general-purpose large models to reduce R&D costs, Xiaomi does not even rule out giving up its self-research plan. **Lu Weibing mentioned that Xiaomi may also invest in external projects, lay out large-scale model tracks, or introduce third-party large-scale model products to integrate into the Xiaomi product system.

More importantly, ** Compared with large-scale model projects that still have many uncertainties, the phased results of Xiaomi's car manufacturing are already in sight. **

Lu Weibing said that Xiaomi's car manufacturing business will proceed according to the original plan. The first car will be rolled off the assembly line and mass-produced in 2024, and the schedule will meet or slightly exceed expectations. In 2023, Xiaomi will also conduct winter and summer tests as planned.

But the more mass production is imminent, the more challenges and doubts Xiaomi faces.

Compared with when Xiaomi announced the car building in March 2021, the competition in the new energy vehicle industry has intensified, and the window of opportunity left for Lei Jun to lead Xiaomi to break through is getting smaller and smaller.

The price/performance ratio has become an important weapon to help Xiaomi rise rapidly in the past ten years. But the expected "young people's first car" may be difficult to produce.

According to "Later" reports, the starting price of Xiaomi's first car, which will be mass-produced in the first quarter of 2024, is set in the range of 260,000 to 300,000 yuan, and the starting price of the high-end version is more than 350,000 yuan.

If the above revelations come true, then Xiaomi will not only have to face the competition from first-line new energy car brands such as Xiaopeng, Ideal, and Tesla in the most competitive range of 250,000 to 300,000 yuan, but also fight a price war with Tesla. , and after lowering the starting price of the Model 3 to about 200,000 yuan, we must also beware of the dimensionality reduction blow from Musk.

Especially considering that there are very few profitable companies in the field of new energy vehicles, the Xiaomi car at that time may not be able to skip the stage of selling cars at a loss.

After investing another 1.1 billion yuan in smart electric vehicles in the first quarter, Xiaomi’s publicly disclosed research and development expenses on the car project have reached 4.2 billion yuan. The intensity of continuing to burn money will only increase.

In the face of the bottomless investment in car manufacturing, Xiaomi may not have too many resources to squander on large models.