In the past two years, everyone can clearly feel that the era of Internet companies has passed.
In the past, the internet industry has never lacked myths of savage growth. The sharing economy, new retail, sinking market, social e-commerce, short ** and other outlets have blown up one after another new software and new platforms.
It's just how lively it is when it rises, and how tragic it is when it's silent.
The news of large-scale layoffs of thousands of people has brought an atmosphere of panic, and the "Internet employment battle royale" has been staged in various places.
At the same time, according to official data released by the Ministry of Industry and Information Technology of the People's Republic of China, in 2022, the total revenue of Internet and related service enterprises above designated size in China reached 146 trillion yuan, down 11%, the first year-on-year decline in the past decade.
* The era is over, and the age of black iron has arrived.
Where is the road?How to get on the road?These have become life-and-death problems that all Internet companies must face.
No one can escape the Internet winter
In addition to layoffs, there is another high-frequency word in the Internet industry in recent years: business contraction and adjustment.
In particular, those exploratory and marginal businesses have suffered a catastrophe.
For example, Tencent has shut down a number of well-known products including Little Goose Pinpin, Phantom Core, QQ Video and Audio, and Kandian ExpressByte directly cut the investment department at the beginning of the year, and not long ago, it also contracted its game business for a light year.
Another intuitive manifestation of the cold weather is that the financial report data and market value of most Internet companies are not the same as they were at their peak.
Not only did the new and old giants collectively "stall", but the rest of the small and medium-sized Internet companies also failed to escape. Mushroom Street and 51 Credit Card are now worth more than 90% from their peak valuationsThe changes in companies such as Daily Youxian and WeWork have also poured cold water on some emerging Internet business models.
Compared with the decline in data, the "death list" of Internet companies in recent years is even more shocking.
For e-commerce platforms alone, according to incomplete statistics from Lianshang.com, there will be at least 89 closures in 2022, involving social e-commerce, shopping guide e-commerce, community**, cross-border e-commerce and other subdivisions, some of which are still popular with capital and backed by giants, such as eBay, Fengqu Haitao, Shihuituan, Little Goose, etc.
The vast majority of these platforms have a "lifespan" of only 2-3 years, and some even have a few platforms that have been shut down hastily after only a few months of being online.
Seeing such statistics, no one has the happiness of the rest of their lives, and some are more sad about the death of rabbits and foxes. No one is sure if they will be the next to be on the "death list".
The disappeared SF "pro-son" Fengqu Haitao.
The reason for this is that most of today's Internet companies are confined to a practical predicamentThe tuyere disappears, the flow peaks, and the hardcore innovation is no more. In particular, the lack of hard-core scientific and technological innovation has a particularly serious impact. The official media also spoke: Don't just worry about a few bundles of cabbage, the sea of stars of scientific and technological innovation is even more exciting.
No matter how you look at it, the current Internet industry can be described as a bleak.
However, if history is any guide, it will be found that the cold winter is also a good time to find new increments and break the involution.
At a technology conference earlier this year, Robin Li recalled the last Internet winter: "Just two months after I started my business, the Internet bubble burst, and the world evaporated 8 trillion yuan in market value. At that time, many people were shaken by difficulties and chose to give up and choose to leave. Finally, it is those who have persevered through the cold winter that have made the first decade of Internet development. ”
As he said, the companies that have survived the cold winter have shined in the era of PC Internet and mobile Internet, such as Microsoft, Google, Pinduoduo, and Tik Tok, and have fought their way out of the encirclement through technological innovation, and have been the hegemon of one party so far.
In addition to past successful cases, the importance of technological innovation to Internet companies can be supported by many negative cases this year.
In March this year, WeChat and QQ were down for a period of time and were notified as first-class accidents by the Ministry of Industry and Information TechnologyAli's **, Xianyu, Aliyun disk, DingTalk, almost every crash has been on the Weibo hot search;Last month's Didi app was even more exaggerated, crashing all night, and some employees reported that even the intranet collapsed.
Relevant reports pointed out that based on the hot search lists that triggered a number of ** reports or appeared on major social **, there were about 9 app crashes last year, and there have been 14 this year.
Every crash not only makes countless users complain, but also directly brings huge economic losses. For example, Didi's system collapsed, and it lost more than 10 million orders and more than 400 million transactionsVipshop was down for 12 hours and lost more than 100 million yuan.
What's more serious is that there is a problem with the cloud product. Last month, Alibaba Cloud experienced two service failures, which not only affected almost all of Alibaba's apps, but also affected customers' business operations. In addition to these two incidents, Alibaba Cloud has had 5 accidents in the past 6 years, especially the large-scale service interruption in Hong Kong Region C in December last year. ”
Because of the frequent collapses in the past two years, some netizens also attributed the reason to layoffs, and commented: It turned out to be "reducing costs and increasing efficiency", but the result was "reducing costs and increasing laughter".
It can be seen that the ability of scientific and technological innovation has always determined the survival and development of Internet enterprises.
The next question is: what kind of technological innovation do they need?
Different eras have different priorities for innovation. In the AI era, intelligence is undoubtedly the key to scientific and technological innovation, especially the intelligent upgrade represented by large models.
Huawei, which is in the limelight this year, has put forward the "All Intelligence Strategy".Alibaba announced the launch of "AI-driven" as a major strategic focusMa Huateng called the large model "a once-in-a-century opportunity similar to the industrial revolution that invented electricity".Other Internet companies have also set off a boom in AI and large models - according to incomplete statistics, more than 130 AI large models have been launched in China so far.
Behind the excitement is the consensus that most Internet companies have automatically reached: all industries are worth doing again with large models, and all products need to be comprehensively upgraded with large models.
Technological innovation in the era of AI
Burning money is just the easiest first step
The current "100-model war" in full swing clearly indicates the focus of Internet companies' next scientific and technological innovation.
If you can do it, you will have a chance to win, and if you can't do it, you will have a high probability of going out.
This is by no means alarmist. The reason is simple, this is a real intelligent upgrade that can greatly improve efficiency and experience.
For example, based on the Pangu model, HUAWEI CLOUD has launched a number of industry-oriented joint innovation models and solutions.
The Pangu Mine model can cover more than 1,000 subdivided scenarios under the business processes of coal mining, excavation, machinery, transportation, and communication, so that more coal miners can work on the groundThe Pangu Railway model can accurately identify 67 types of wagons and more than 430 kinds of faults running on the live network, and the fault-free** screening rate is as high as 95%.
In addition, various industries such as education, entertainment, social networking, e-commerce, office, and medical care all have places for large models.
In the field of education, large models can teach students according to their aptitude, and realize personalized analysis and guidance and guided learning XiDo live e-commerce, generate real and vivid copywriting with one click, and then seamlessly connect to the digital human for live broadcast;For patients, the large model can help them manage their personal health, establish medical history information, and form an AI private doctor ......
However, the opportunities are great, and the corresponding challenges are even greater.
Taking computing power as an example, this is the "hydropower and coal" in the intelligent era, and the blowout development of the large model industry, coupled with the continuous increase in US export controls and other objective reasons, many Internet companies have to face the dilemma of shortage of computing power. This has also led to the outbreak of the computing power leasing industry, and this year's computing power leasing has increased several times year-on-year.
Unfortunately, with the exception of a few powerful giants, with the scientific and technological innovation capabilities of most Internet companies, it is destined to be difficult to cope with these shocks. If you attack rashly, I'm afraid it will only end in embarrassment.
First of all, training large models is a business with high technical threshold and high cost.
A large model with hundreds of billions of parameters often requires tens of thousands of A100 computing power, and the price of an A100 chip is as high as 1Around $50,000. Therefore, it is rumored that the training cost of GPT-4 is as high as $1 billion.
Spending money on hardware is just the beginning, and there are many challenges to face, including data preparation, hyperparameter tuning, model design and debugging, training time, and model evaluation and iteration, all of which require a lot of experience, time, and resources.
What awaits companies is likely to be a long, uncertain cycle.
Then there are data security and infrastructure concerns.
Data is the core asset and source of competitive advantage for Internet enterprises. However, as a system based on artificial intelligence technology, the large model data side has a series of security issues, including storage security, privacy protection, and malicious data leakage.
In March of this year, shortly after Samsung Electronics introduced the ChatGPT service within the enterprise, three confidential data leaks occurred. Some employees entered confidential information related to semiconductor production** and internal meeting information into the ChatGPT port, resulting in these sensitive materials being uploaded to a U.S. server, which is likely to have been leaked.
This also leads to another difficulty: the combination and co-creation of industry and large models.
The large model can not only chat and answer questions, but can only go further if it is combined with the industry, landed in the scene, and turned into actual productivity.
This requires that it not only have the versatility and generalization of the general large model, but also have professional knowledge and industry skills, so as to provide users with professional services. Therefore, the large models used by enterprises have extremely high requirements for reliability, must be controllable, traceable, and correctable, and must be repeatedly and fully tested before they can be launched.
In short, the cruel and unavoidable practical factors of burning money, brains, and time have become obstacles to the intelligent upgrading of Internet companies and large models.
Many Internet companies will have to face the risk of being eliminated, but they are in a dilemma.
The cold is not gone, but there is no need to despair
Despite the difficult road ahead, Internet companies that are at the forefront of the times still have the opportunity to strengthen their technological innovation capabilities and go through the cycle.
However, there is a premise: everyone needs to "recognize the situation".
One is that external difficulties will continue.
Although the economic and trade relationship between China and the United States fluctuates from time to time, the chip control for China has been increasing unabated, especially the ban on high-end AI chips. This month, at the annual Reagan Defense Forum in the United States, U.S. Secretary of Commerce Raimondo made it clear that "China must not be allowed to catch up with and surpass the United States in artificial intelligence." She hopes that American technology companies can focus on the overall situation, endure a temporary loss of revenue, and defend the United States
One is how Internet companies should contribute.
China is making a big push for artificial intelligence to benefit more companies and more people, so stop thinking about the "winner-takes-all" logic of the past. If Internet companies want to develop sustainably, they must not only contribute to the development of China's artificial intelligence model industry, but also bring value to the real economy.
With an insight into the overall environment, the next road to innovation and breakthrough is meaningful.
In Brother Dao's view, for most Internet companies, the core solution to break the situation lies in two key words.
The first is "domestic substitution".
At a time when the import of AI chips is restricted, domestic substitution is destined to be imperative. Prior to this, Internet giants have already "made cores", such as Kunlun chips, Ali's high-performance inference AI chip Hanguang series, and Tencent's self-developed AI inference chip Zixiao. Other AI-related chip companies are also making efforts, such as Cambrian, Jingjiawei, Haiguang Information, Innosilicon, Suiyuan, Muxi, Biqi...
Huawei's Ascend 910 chip was called "Huawei's GPU has been benchmarked against the NVIDIA A100" by Liu Qingfeng, founder and chairman of iFLYTEK, and its performance is close to that of the NVIDIA A100.
Another key word is "cloud services".
For example, when solving the problem of computing power, compared with traditional servers, cloud services have inherent advantages such as scale, scalability, and elastic scheduling, and have more powerful computing and processing capabilities. For example, HUAWEI CLOUD's Ascend AI cloud service is backed by three AI computing power centers deployed in Gui'an, Ulanqab, and Wuhu, allowing enterprises to access AI cloud services at any time with just one click, making computing power as ready to use as water and electricity.
In this way, the technical, O&M, and engineering challenges faced by the construction of AI computing power clusters have also been solved. Large-scale model services can not only be launched faster, but also greatly improve the interconnection, latency, and heat dissipation between servers.
It can be seen that the core of the competitiveness of a high-quality cloud vendor lies in providing a practical-oriented customer experience. Reliable and easy-to-use cloud services can make the large model application of enterprises fast, well-trained, and reliable.
It can be said that in the AI era, making good use of cloud services will be the key for Internet enterprises to carry out effective scientific and technological innovation.
At present, many companies have taken the lead in seizing the opportunity.
iFLYTEK's Spark model has been iterated to 30 version, and even dare to fully benchmark ChatGPT.
Zhihu's Zhihai chart model is planned to be used to detect content, filter information, assist in creation, and improve content quality.
NetEase Youdao launched the first vertical large model in the field of education in China, "Zi Yue", and Su Peng, vice president of NetEase Youdao, said not long ago: The large model is a natural fit with education.
Dingdang Health released Dingdang HealthGPT and launched the first application-oriented pharmaceutical AI product developed based on this - Dingdang Pharmacist and Dietitian AI Assistant.
360's intelligent brain model has become the first large model product in China to disclose the realization of revenue, according to the financial report, the large model created 20 million yuan in revenue in the first half of the year.
Many of the birth and iteration of various large models are inseparable from the help of cloud services.
This also proves that AI technology and large models are by no means the patents of a few companies.
Today, the cold has not dissipated, and slogans such as "going to fat and increasing thinness" are still sounding from time to time.
At the critical moment of survival, many Internet companies have been on the road of embracing intelligence and empowering the real economy.
Other enterprises can also have their own large models, strengthen their own scientific and technological innovation in the new era, and add many chances of success in the cold winter. 100 Helping Programs