Mobile phone manufacturers have released AI big models one after another： Is the battle of industry reshuffle started？

The wind of the AI ​ ​ big model has been swaying in the technology industry for a year, and now it has finally blown into the smart phone industry.

At the developer conference held on November 1, 2023, vivo released the self-developed general AI large model matrix-blue heart large model. It is understood that the blue heart model contains five products, covering all kinds of usage scenarios. In fact, not only vivo, but also smart phone manufacturers such as Huawei, Glory and Xiaomi have officially announced or launched their own AI model products, and some have even implanted related products into smart phones.

Considering that the large model can greatly improve the efficiency of users’ work and life, with more and more smartphone manufacturers adding related technologies, consumers’ acceptance of this function is gradually increasing, and the large model may become the key factor to determine the sales of smart phones.

In the battle of AI big model, mobile phone manufacturers were collectively late for half a year.

Because the big model is related to the next era, it is normal for smartphone manufacturers to increase their business, but it is worth reflecting that compared with technology companies, smartphone manufacturers are about half a year late in laying out the big model.

At the beginning of 2023, with the explosion of ChatGPT, many technology companies launched large model related products. For example, on March 16th, 2023, Baidu released "ERNIE Bot". A month later, Ali also launched "Tongyi Qianwen". In May 2023, the data disclosed by China Institute of Science and Technology Information showed that 79 large models had been released in China, which was called "White Model War".

It can be found that most products related to large models in China were released in the first half of 2023. On the other hand, smartphone manufacturers were extremely calm in the wave of big model release in the first half of the year.

According to PChome, in the mobile phone industry, Huawei was the first manufacturer to release large model products. On July 7, 2023, Pangu Big Model 3.0 was launched. However, it is worth noting that at the beginning of the release, the Pangu model was not deeply connected with Huawei’s mobile phone. "No poetry, only work", focusing on government affairs, finance, manufacturing and other industries.

Lu Weibing, partner and president of Xiaomi Group, said, "We won’t make a general model like OpenAI.". Xiaomi will actively embrace it (big model) and think about how to do it. We have several firm views-whoever owns the user, who owns the data and who owns the user’s usage scenario in the future may have a better advantage in the new round. "

Obviously, smartphone manufacturers realize that although they belong to the same big model track, the big model of mobile phone and the big model of web page have very different usage scenarios. The former needs to retrieve the data in the mobile phone. If cloud computing is used to provide services completely, then the user data security will not be guaranteed. Therefore, the large model of mobile phone also needs some computing power on the end side.

In this regard, Zhao Ming, CEO of Glory, said that "individuals do have some needs for artificial intelligence or AI, but because it involves data privacy protection, information needs to be isolated to a certain extent, so it is impossible for us to share all our stored contents and photos with the network".

The upstream computing power is mature, and mobile phone manufacturers are rushing to the beach.

Obviously, the reason why smartphone manufacturers didn’t launch AI big model related products in the first time may not be because the technical ability is not enough, but because they need to wait for the end-side computing technology of the upstream supply chain to mature considering the privacy and security needs of consumers.

In this regard, Zhao Ming said that the end-to-end big model "has more and more processing power, and the 1 billion parameters can’t be used. It’s a toy, 5 billion, 6 billion and 7 billion, which is quite good to use, and it may be 10 billion and 20 billion in the future".

On October 25th, 2023, Qualcomm released a brand-new third-generation Snapdragon 8 mobile platform. One of the highlights of this product is that the speed of Hexagon NPU is increased by 98%, the energy efficiency is also increased by 40%, and it supports models with more than 10 billion parameters.

Coincidentally, on November 6th, MediaTek released the flagship 5G generative AI mobile chip of Tianji 9300, which is a major selling point, namely, it supports the terminal to run large AI language models with parameters of 1 billion, 7 billion and 13 billion, with a maximum of 33 billion, and supports leading mainstream AI models such as Android, Meta Llama 2, Baidu ERNIE Bot Model and Hundred Rivers Intelligent Hundred Rivers Model.

It can be said that it is precisely because the flagship mobile platforms of Qualcomm, MediaTek and other enterprises can provide abundant local computing power that the large model products of major smartphone manufacturers have the possibility of landing.

After the brand-new flagship mobile platform came out, many mobile phone manufacturers began to act. At the Snapdragon Summit, Glory announced that the Glory Magic6 series will be equipped with the third-generation Snapdragon 8 mobile platform, supporting the 7 billion-parameter end-to-end AI model. Coincidentally, on October 26th, 2023, when Xiaomi released Xiaomi’s mobile phone equipped with the third-generation Snapdragon 8 mobile platform, it also announced that Xiaomi’s surging OS and Xiao AI’s input assistant would be introduced into the AI big model to support AI to create texts with one-click access.

However, it is worth noting that, because the mobile phone AI model requires high local computing power, perhaps only the mobile phone equipped with the latest flagship processor can support the ultimate AI model function. PChome learned from MediaTek that the demand for memory for large language models has soared, and the large model with 13 billion parameters needs at least 13GB of memory. In addition, the mobile phone with 6 GB of APP and 4GB of Android OS cannot run the large model with this parameter.

In order to reduce the memory pressure of smart phones, Tianji 9300 supports memory hardware compression technology, which can reduce the memory occupation of large models by 61%. But even so, a mobile phone with a minimum memory of 16GB is needed to run the 13 billion parameter model.

Because the AI ​ ​ big model has high requirements for end-side computing power and memory, it also means that it is difficult for old smartphones to experience related functions next. Take Xiaomi 澎湃 OS as an example, its first batch only adapts to Xiaomi 14, MIX Fold2, MIX Fold2, Xiaomi 13 and Redmi K60, and none of them are old flagships.

The landing of the big model will trigger a wave of mobile phone industry changes.

It is difficult for old smartphones to experience the related functions of AI big model, which is certainly a loss for consumers, but it may be a new opportunity for smartphone manufacturers. Because this may inject new vitality into the smartphone industry that is sluggish during the day.

Since 2017, with the completion of the sinking of smartphones, coupled with factors such as economic pressure and overcapacity, most consumers are no longer keen on changing machines, and the dividend in the smartphone market is gradually disappearing. According to IDC data, in 2022, China’s smartphone market shipped 286 million units, down 13.2% year-on-year, the biggest drop ever.

In this context, many smartphone manufacturers are actively exploring new technologies that can effectively promote consumers to change their phones. For example, in the past two years, many smartphone manufacturers are stepping up the layout of folding screen phones.

From the market point of view, folding screen mobile phones are indeed growing at a high speed. Canalys data shows that in 2022, the global shipment of folding screen mobile phones was 14.2 million units, much higher than the previous year’s 8 million units, with a significant increase. However, it is worth noting that the folding screen mobile phone market can only be regarded as a drop in the bucket compared with the overall smart phone industry.

According to statistics, in 2022, global smartphone shipments were 1.193 billion units. From this point of view, the penetration rate of global folding screen mobile phones in the same period was only 1.19%. This is mainly because the folding screen mobile phone is expensive and has a poor experience, so it is difficult to satisfy all consumers.

On the other hand, the big model does not change the shape of the folding screen mobile phone, and it can significantly improve the work and life efficiency of consumers without too high hardware cost. For example, in the "Little V Assistant" dialog box of OriginOS 4, users can use fuzzy semantics to search for photos, documents or notes in their mobile phones. It can be said that after the big model is empowered, the mobile phone can really be called a "smart phone".

Because the big model is very attractive, and it is difficult for old products to experience related functions, with the gradual landing of terminals equipped with this technology, it may prompt the smart phone industry to usher in a wave of machine change.

In this regard, at Qualcomm Snapdragon Summit, Xiaomi President Lu Weibing said that it is estimated that the sales volume of mobile phones in China market will reach 268 million units this year, and the sales volume will return to more than 280 million next year. It is estimated that the global smartphone industry will reach the bottom by the end of this year, and there will be a 5% growth in 2024.

All in all, it is not accidental that many smartphone manufacturers have launched AI big model products in the second half of 2023, especially after September, but because they have considered the data security needs of consumers, so they need abundant computing power at the peer side.

Unlike the technology company’s AI big model product, which is just a cloud function, for smartphone manufacturers, the AI big model is not only a brand-new technology, but also likely to trigger consumers to change machines and reorder industries, so it can be called the "shuffle war" of the next era.

This may also mean that smartphone manufacturers that have not laid out AI big model related technologies today will face tremendous downward pressure in the future.