news

the large model gives internet companies an opportunity to join the fruit chain.

recently, latepost broke the news that byte is exploring the combination of large models and hardware. it will not only develop ai hardware itself, but also cooperate with external hardware companies. among them, the heads of relevant departments are promotingdoubao big model cooperates with smart assistants from other mobile phone manufacturers, similar to the combination of openai and apple.

as early as april this year, byte began to launch an internal ai mobile phone research and development project, exploring large-scale model software solutions based on mobile phones, providing them for reference by mobile phone manufacturers, and emphasized that it would not end up making and selling mobile phones by itself.

after successively reaching cooperation with domestic mobile phone manufacturers oppo, vivo, honor, xiaomi, etc. in the field of large models, the latest mobile phone manufacturer to be breached by byte is samsung. in june, samsung officially announced that some of its voice assistant bixby’s generative ai capabilities are based on the bean bag model of byte volcano engine, similar to the combination of apple’s siri and openai gpt-4o.

as the last mobile phone brand among mainstream mobile phone manufacturers to launch ai capabilities, coupled with policy and regulatory factors, it also needs to find a national version of the iphone to cooperate with domestic large models. this has been given to byte and other internet companies with large model self-research capabilities. a historic opportunity to join the fruit chain.

on the iphone 16 series officially released in the early morning of september 10th, beijing time, apple stated that the launch time of apple intelligence in china will be subject to regulatory approval.

according to the official plan, after apple’s smart phone is launched on iphone and other devices in october, it will temporarily only support american english. in december, it will expand to localized english in australia, canada, new zealand, south africa and the united kingdom, and will support chinese, french, japanese, spanish and many more languages.

during the second quarter earnings call, apple ceo cook said in response to the specific opening time that apple has cooperated very constructively with regulatory agencies in china and other places. "our goal is to take action as quickly as possible. we you have to understand the regulatory requirements before you can commit to what to do and set a timetable accordingly.”

who will grab the tickets for the ai ​​fruit chain? after entering 2024, relevant rumors continue.when cook visited china in march this year, jiemian finance news broke the news that apple had selected baidu to provide ai functions for this year’s iphone 16, mac system and ios 18. subsequently, some media revealed that alibaba and baichuan intelligence were also considered by apple.

at that time, baidu responded to alphabet (id: wujicaijing) that the two parties were in the negotiation stage.as of now, no further news has been released from the parties.

the mystery surrounding the progress of ai cooperation is also related to apple’s usual confidentiality system. the outside world is basically unable to obtain specific information about the cooperation between apple and suppliers. "whether the cooperation is smooth or there are problems, apple will not take the initiative to make public comments." said major general ding, an industrial economic observer and founder of ding technology.

how strict is apple's supply chain control? an example from apple's car-making project may illustrate one thing or two. in january 2021, in response to rumors that kia was expected to oem apple cars, hyundai motor executives stated that negotiations with apple were still in the early stages.

bloomberg later reported that apple chose to terminate the cooperation because it was dissatisfied with hyundai motor's leaks to the outside world, which led to widespread publicity about apple car.

but such secrecy may sometimes help the fruit chain. for example, the pending list of large model suppliers leaves room for latecomers such as byte to compete for apple's favor.

it is conceivable that an invisible melee is taking place among various internet companies around the competition for suppliers of large models of iphone 16 chinese version.

A

terminal mobile phones that come with a series of high-frequency applications naturally have many scenarios for the implementation of large ai models.what mobile phone manufacturers have to do is not to go to great lengths to create a large model app for users out of thin air. instead, they have chosen to integrate the current generative ai into various existing applications.

the ai ​​application strategy for smartphones coincides with some of the views of zhu xiaohu, managing partner of jinshajiang venture capital. compared with reconstructing new scenarios, in zhu xiaohu’s view, large models are more beneficial to existing players in each industry. “players who already have usage scenarios can easily add it to their products through large models such as chatgpt. the functions of artificial intelligence, such as the current intelligent customer service.”

especially considering that each large model manufacturer is trapped in the industry problem of finding its own pmf (product market fit) to complete the commercialization closed loop, mobile phone manufacturers with their own application scenarios and willingness to pay have become large model manufacturers. competing for core customers.

before chatgpt was released, byte was once trapped in commercialization and chose to suspend the development of large models.later latepost reported that june 2020OpenAIafter the release of gpt-3, byte trained a large generative language model with billions of parameters. due to the limited parameter scale, the model’s generation ability was average. at that time, byte could not see its commercialization possibility. “roi ( the return on investment (roi) cannot be calculated." this exploration was just a scratch.

idc disclosed in the "ai mobile phone white paper" released that it is expected that in 2024, ai mobile phone shipments in the chinese market will reach 37 million units, and by 2027, this number will increase to 150 million units. the market share will exceed 50%.

this also means that domestic internet companies that want to sell large-model software solutions to mobile phone manufacturers will have a potential market worth at least tens of billions of yuan.

after openai reached a cooperation with apple, the outside world has been rumored that the two parties are more of a resource exchange, and neither party has charged a licensing fee. cook did not make any further comments at the second-quarter financial report. he only mentioned, "as for the monetization model, i don't want to get involved in the terms of the commercial agreement because they are confidential to both parties."

even if domestic internet companies also continue the openai strategy and adopt a resource replacement cooperation method with apple, the profit space left for internet companies will still exist, that is, selling cloud services.

the arrival of the wave of large ai models has further highlighted the importance of cloud services in major internet companies. in order to promote more large-model services to their own clouds, each company is willing to use price wars to clear the way. alibaba cloud, byte volcano engine, etc. all hope to use price reductions to pursue the network effect and scale effect of cloud computing, so as to achieve more customers using the cloud. the more supply chain procurement costs, amortized r&d costs and resource idle costs are, the lower the positive cycle effect will be.

it is worth mentioning that due to the extremely high shipment volume of iphones and strict standards for suppliers, in addition to bringing in large orders, the shortlisted fruit chain will also be widely recognized by the industry.

the apple symbol, bound to high-end products, is like a walking golden signboard in the supply chain field. after breaking through the iphone, these internet companies may be the first to have a stepping stone to more terminal hardware, such as becoming large model suppliers for foreign smart car brands.

B

compared with the undercurrent of domestic large model manufacturers competing for ai fruit chain tickets,apple also desperately needs a catalyst to reignite demand in the chinese market.looking around, apple's smart phone is undoubtedly the most effective accelerant in its hands.

in the second quarter financial report, apple's greater china revenue was us$14.7 billion, a year-on-year decrease of 7%, which was lower than analysts' expectations of us$15.3 billion, becoming the only region where apple's revenue fell year-on-year in the second quarter. to make matters worse, in the chinese market, this is apple's fourth consecutive quarter of year-on-year revenue decline.

along with the decline in revenue, there is also iphone sales in the chinese market.canalys data shows that in the second quarter of 2024, mainland china's smartphone market achieved further recovery after experiencing an inflection point in the previous quarter, with shipments increasing by 10% year-on-year and returning to the level of 70 million units.

however, apple bucked the trend and fell, with its market share falling by 2% year-on-year and its ranking falling to sixth place. as a result, it was the first time in history that a domestic mobile phone manufacturer occupied the top five positions in the chinese mainland market.

apple smart is expected to inject new growth momentum into the iphone.wedbush analyst dan ives predicted that starting with iphone 16, the chinese market will grow again.

in order to prepare for apple smartphones, the iphone 16 series has undergone a key upgrade. driven by the new a18 and a18 pro chipsets, apple breaks the tradition of reusing old chipsets in low-end models, and the iphone 16 series comes standard with 8gb of memory to facilitate the smooth operation of large models.

the combination of ai and mobile phones is indeed expected to bring about a subversive experience at the interactive level. "now if i want to run a process within the company, i may click seven times. if i switch to ai hardware, it can be completed in one step through voice interaction, which is equivalent to save time on the six intermediate steps. " according to jiang yi, founding partner of hengye capital, the interface changes brought about by ai hardware will eventually bring about efficiency improvements on the user side.

however, the hope that apple’s intelligence will drive the iphone’s chinese market to grow again has been poured cold water on by the uncertainty of the release date of the national bank version of ai.

for apple, another important factor in stimulating iphone sales to return to growth is thatif the slowdown in hardware growth continues, it will inevitably affect the growth of apple's software services.

in the middle of the year, apple dared to put pressure on tencent and byte, requiring the two internet companies to take measures to prevent mini program operators from directing users to external payment systems to avoid the 30% apple tax. otherwise, the app store will reject the confidence behind the subsequent updates on wechat and douyin lies in the strong distribution channels created by their large-scale iphone hardware.

moreover, once ai is integrated into the iphone, it may also contribute additional software revenue to apple. on-device ai in smartphones (and pcs) will bring at least three profit models to device manufacturers: ai as function, ai as service, and ai as interface. "for apple, joining the ai ​​army is not just to catch up with industry trends, but the opportunity and potential to increase revenue that it brings cannot be underestimated." canalys analysts pointed out.

C

joining fruit chain is also a double-edged sword for internet companies. while enjoying apple's endorsement, these internet companies also need to bear more stringent technology iteration and cost management risks.

multiple production of the same parts is a common method used by apple to control risks and reduce costs.on the one hand, if one manufacturer has problems with its supply, other companies can replenish it in time, which improves the stability of apple's supply chain; on the other hand, because apple strictly controls the supplier's production equipment, the supplier's production line needs to be supplied to apple separately. to prevent being kicked out of the fruit chain, suppliers will compete with each other, giving apple ample room for negotiation.

after using oled screen for the first time on iphone x in 2017,samsungit was once the exclusive supplier of oled screens for apple, but this obviously does not comply with the rules of apple's supply chain. after that, apple began to intend to introduce more oled suppliers, and lg display and boe gradually joined.

with apple technology certification and just one year after joining the fruit chain, boe has made rapid progress in the oled market. by 2021, a cinno research report shows that of the 668 million global amoled smartphone panel shipments, boe will account for 9%, ranking second in the world, second only to samsung's 72%, but higher than lg display's 7% . the price boe pays is to maintain high investment in technology research and development at all times to prevent samsung from widening the gap.

now,in the field of large models, domestic internet companies are crossing the river by trying openai.the pressure to catch up technologically continues unabated. while domestic large model manufacturers are still struggling to pursue gpt-4o, openai released the o1 model in mid-september, once again pointing out the direction of technological evolution in the field of large models.

jiang daxin, the founder of step star, regards the emergence of the o1 model as a new direction for scaling laws of model performance. in addition to bigger is better, the rl reinforcement learning route has become another iterative new paradigm of current large models.

especially in the context of the current slowdown in the evolution of large model technology, with the help of rl reinforcement learning, "openai o1 has raised the upper limit of ai." yang zhilin, founder of dark side of the moon kimi, commented.

openai once again allows the outside world to see the path to continuously improving the upper limit of ai capabilities. even in the view of baidu founder robin li, the gap between large models may become wider in the future. if you want to continue to stay on the poker table, the model needs to be continuously and rapidly iterated, updated and upgraded, and this iteration and upgrade process may require several or more than ten years of continuous investment.

the end point of all the above efforts is to make ai capabilities easier to use and more powerful.although ai+ mobile phones have become the focus of current large-model applications, it is undeniable that this is also a track where consumer expectations are severely overdrawn.

so far, ai features have not become a major driving force for consumers to switch to new phones. samsung has heavily promoted the ai ​​capabilities of its latest phones, but after the initial sales of the devices, ai has not become a major selling point for carriers, according to a baystreet research survey of u.s. carrier stores.

"ai has not yet sparked a strong response among the public," said cliff maldonado, principal analyst at baystreet research.

now, changes in demand for the ai-powered iphone are about to become another touchstone to test the market’s interest in large models.

references:

"byte tries hardware again: exploring ai headphones, glasses and other products, linked with large bean bag models" latepost

"questmobile2024 ai agent application insight semi-annual report" questmobile

"interpretation of apple press conference strategy: new iphones and ecosystem optimization open a new chapter in the future of ai" canalys

"robin li's internal speech discusses the three major cognitive misunderstandings of large models: the gap between models will widen in the future, and they are still far from ideal" sina technology

"zhang hongjiang: my eight observations and opinions on ai and large models" deep web