"When industry competition appears in front of most people in the form of price war, it often means that enterprises in the industry have reached the same level."
Text/Ba Jiuling (WeChat WeChat official account: Wu Xiaobo Channel)
You may often see such a scene in film and television dramas: when several factions compete for territory, their guns are raised and they confront each other. Everyone dares not pull the trigger, and everyone wants to pull the trigger. A firefight is on the verge.
This is called "Mexico deadlock", which refers to the delicate balance formed by many opposing parties because of mutual containment.
Now this deadlock has been broken in the field of AI. The person who pulled the trigger was OpenAI.
On May 14th, OpenAI released an upgraded version of GPT-4, GPT-4O, and announced that it was open to everyone for free.
A day later, gunfire finally came from the other side of the ocean, and sparks crossed the fertile soil here.
On May 15th, ByteDance announced its main model of bean curd, and the input price was 0.0008 yuan/thousand Tokens(Token is the smallest unit of input data used by the big model, such as words or characters).
Bullets began to whiz by.
Six days later, on May 21st, Alibaba announced that the price of the main model of Tongyi Thousand Questions had been greatly reduced, and the input price of the main model Qwen-Long and API had dropped from 0.02 yuan/thousand tokens to 0.0005 yuan/thousand tokens, a direct drop of 97%, with immediate effect.
Four hours later, Baidu skipped the price cut and directly announced that ERNIE Speed and ERNIE Lite, two entry-level products of its big model ERNIE Bot, were free.
A day later, in hindsight, Iflytek and Tencent also announced: iFLYTEK Spark API capability is free to open, iFLYTEK Spark Lite API is permanently free to open, and the price of iFLYTEK Spark pro/Max API is reduced to 0.21/ 10,000 Tokens；;
Tencent’s mixed model has been reduced in price, and the price of mixed -lite model, one of its main models, has been adjusted from 0.008 yuan/thousand tokens to full free.
……
Shots fired and people were killed, which didn’t seem to happen in this scene. But no one will doubt that this kind of thing will not happen in the next time.
After all, when the competition in an industry appears in front of most people in the form of price war, it often means that the enterprises in the industry have reached the same level.
However, enterprises also have their own interpretations of the price war. For example, Tan Dai, president of Volcano Engine, said: "Losing money for income is not sustainable, and we will not do this."
For example, Liu Weiguang, vice president of Alibaba Cloud Intelligent Group, said: "Only when the cost of AI reasoning is reduced by ten times or even a hundred times every year can we really promote the outbreak of AI applications in all walks of life in the industry."
If we want to go back further, the price of DeepSeek, a large model of magic square quantification of private equity funds, and the entry-level model of Zhipu AI will be reduced earlier:
On May 7th, DeepSeek announced that it would reduce the price of its large model DeepSeek-V2 to 1 yuan and 2 yuan per million token (32K context).
On May 11th, Zhipu AI Tune reduced the price of its entry-level large model GLM-3-Turbo (with a context length of 128k) from 0.005 yuan/thousand tokens to 0.001 yuan/thousand tokens, while the batch API of GLM-3 Turbo Batch was 50% cheaper, reaching 2 million tokens in 1 yuan.
However, does this mean that ordinary people have a better chance to use large models at low cost or even zero cost?
I am afraid the answer is no.
Today, there are two main commercial charging modes in the big model:
The first is to let C-end users pay., that is, ChatGPT Plus’s membership subscription mode of $20/month.
The second is to let B-end users pay.That is, the developer API calls the service to let the developer connect the "faucet" of the large model and use the "water" inside.
It is the second kind of price reduction this time.
Comparatively speaking, the B-end market is much smaller than the C-end market, so the seemingly massive price war is not too much cost pressure for enterprises.
For ordinary people, the good time for everyone to use AI at low cost has not yet arrived.
At the same time, further controversy followed:While foreign companies are trying to run hard to keep ahead of technology and rush to the sea of stars, domestic companies are commercializing and gaining market share?
For example, Huang Renxun said that our company never talks about market share, which means that everyone is doing the same thing; Musk said, I don’t pay attention to technical barriers, I only pay attention to the speed of innovation; Altman said that the development of AI is like a tornado, and OpenAI should break through the limit.
The above words are more encouraging, but in the real environment, when the industry leaders are in a leading position in technology, they can only keep ahead and continue to obtain the most resources in the industry.
At this time, followers have many ways to catch up, and full commercialization can provide financial endurance for continuous follow-up.
It is not uncommon for followers to surpass the front runners before the marathon finishes.
Of course, there are still various problems, and we have invited professional experts to answer them.
These days, the price reduction of big models is free. If this development continues, will there be any post-posting in a couple of days? What will you choose then?
My choice is, I will choose whoever delivers eggs. Before the eggs were delivered, I started from rational logic and made three points:
◎ First of all, the price war has brought about the popularity of AI. Some people say that AI is like air, and we cannot leave it in the future. And the big model price war, properly played the role of "air purifier".
Giants such as Baidu and Ali have cut prices, even for free, so that ordinary people can easily get in touch with better fresh air and better AI tools.
In fact, the fundamental reason for the price reduction is that the cost of large model reasoning is gradually decreasing. OpenAI has reduced the price by an astonishing 90% in the past year or so.
However, OpenAI has dropped so much in a year, and we have dropped so much overnight, so we feel the shock wave is even bigger.
The inner OS of the domestic big model: I can’t surpass you technically for the time being, and I can’t surpass you in price reduction?
◎ Secondly, the price war has also brought about internal friction and homogenization in the industry.
The big factories are fighting fiercely, while the small factories may face "the disaster of extinction".
Imagine the situation of a small AI company in the price war, just like a mouse tap dancing at the foot of an elephant. It is necessary to maintain elegance and avoid being trampled on the city.
The price war may also lead to the mismatch and waste of resources in the market. Excessive price competition will force manufacturers to cut costs and thus invest less in R&D and innovation.
This is not only not conducive to the long-term development of the whole industry, but also may lead to excessive market concentration, forming an oligopoly, further inhibiting the vitality and innovation power of the market.
Finally, to win the price war, we must persist in innovation and provide differentiated services.
The price war is only the first round of the Hundred Models War, and the real "ultimate confrontation" lies in the innovation of technology and service.
Just like the competition in the food industry, discounted noodles may temporarily capture consumers’ stomachs, but what really makes people forget is the bowl of "secret beef noodles" with unique flavor.
The same is true in the field of large models. After the price war, whoever can make that bowl of "secret beef noodles" will stand out in the future competition.
The price war of the big model is not terrible. There are two main reasons for the price war.
The first reason is that the computing power of big technology companies is surplus.
When the friction between China and the United States escalated, China’s technology giants were afraid of cutting off supply, so they frantically hoarded NVIDIA’s chips and piled up computing power resources quickly.
Later, Huawei’s rising series of chips were made, and the big manufacturers made a lot of purchases. As can be seen from the financial reports of Baidu and Ali, the expenditure in this area is very huge, reaching several billion yuan.
However, the promotion speed of AI application is not as fast as they thought, which leads to the idling of computing power and the waste of resources in the early stage.
Now big manufacturers are trying to dilute the cost and let the computing power roll up.
It is worth noting that,NVIDIA chips in the second-hand market are hard to sell now.
If you want to sell it to the government, the government won’t want it. They will only use domestic chips. I want to sell it to the cloud platform, but the cloud platform has plenty of computing power, so there is no need to buy new chips.
I received several calls from the United States asking if I could contact them. They have NVIDIA A100 or H100.
The second reason is the progress of technology and services.
In fact, ChatGPT is totally different from Alibaba Cloud and Baidu Cloud.
ChatGPT doesn’t do public cloud, doesn’t consider the dynamic distribution of public cloud, and doesn’t consider enterprise application scenarios, but piles up resources on the big model crazily.
On the other hand, cloud service providers in China should consider for enterprise users, and let them use artificial intelligence and big models on their own platforms, so the service of China Dachang will be better.
Both Alibaba Cloud and Baidu Cloud have been able to dynamically allocate the original computing resources with greater value and more efficiency.
For example, the original 100 cards can serve 150 users. Now, after technical upgrading, 100 cards can serve 250-300 users, which invisibly improves the utilization efficiency of counting cards.
The surplus of computing power resources and the improvement of applied technology have the basis for price war.
Throughout history, the current big model war, like the Thousand Regiments War and the Taxi APP War in the Internet era, all gave subsidies and gave low prices. In the end, only two or three giants were left.
It is expected that the domestic big model will decide the outcome in the first half of next year, and the public cloud market will also change.
Whoever has an advantage in this will probably take down the public cloud business. This is also an important reason why Alibaba Cloud took the lead in price reduction, and it wants to maintain its dominant position.
There are three main reasons why the domestic big model suddenly launched a price war.
First, domestic manufacturers use price wars to circle users;
Second, the middle and low-level competing products with stronger foreign strength have been free.The previous tariff standards of domestic manufacturers are invalid;
Third, long text applications and multimodal applications need more Token.If the tariff remains unchanged, the user’s use cost will be ridiculously high, and the manufacturer will become an island of the platform.
Price war is definitely not a good thing, although it is inevitable that prices will go down. Free is not the key, but the principle iteration, level improvement and multi-modal evolution of the model are the key.
The massive users accumulated for free may be lost overnight because the model intelligence level is left behind.
Now, the price war has become the communication highlight of domestic manufacturers. This is not right. What everyone should be concerned about is the innovative breakthroughs in model principle iteration, level improvement and multi-modal.
Companies with different volumes have different breakthrough directions. Startups with shallow funds should focus on applications, fine-tuning models or RAG or scene-based professional models.
But powerful manufacturers must dare to make breakthroughs at the bottom. If you don’t do your homework well, you won’t understand if you want to copy your homework in the future, and copying mistakes is even more likely.
You can’t copy your homework, nor can you copy your experience. If you want to bring Internet thinking to the wave of super intelligence, I am afraid that you have misunderstood the value source and development paradigm of AI, and the result is tantamount to self-satisfaction.
China’s Internet failed to move into the next era in time, which is behind the application, traffic and quick money thinking at work.
In the era of artificial intelligence, science and technology have become the force of value, facing the competitions of AGI, EI and II, and the traffic is only attached to science and technology.
Traffic in the sense of online celebrity will not bring user loyalty to AI products, especially for the general generative model, users with slightly poor standards will be lost instantly.
Without the continuous evolution of the core competence driven by the underlying technology, it is tantamount to building a so-called strong intelligence with application scenarios.
Finally, I want to say that AI must benefit everyone and AI must be people-oriented.
Science and technology should have values. Today and in the future, science and technology and its operators should always think about helping people, serving people and benefiting mankind, not dominating people, not controlling people, expelling people or squeezing people.
With the price reduction and the increasing number of open source of foreign big models, China big model manufacturers announced price reduction or free of charge, mainly because they wanted to expand the number of users.
Due to the limited scale of paying users of large models, it is difficult to train useful AI. Price reduction can attract more users to participate, increase the data scale, promote the improvement of models, and also help to expand market share.
Price war can be regarded as a strategy of market competition.
On the one hand, it helps consumers to obtain lower-cost products or services and promotes technology popularization and innovation.
On the other hand, excessive price war may lead to profit compression, affect the healthy development of the industry, and even lead to vicious competition.
There are not many paying users of the big model, and the price reduction will not bring adverse effects to themselves.
Facing the price war of big factories, startups should also be prepared to burn money and form their own competitive advantages as soon as possible.
Start-ups should focus on market segmentation, provide differentiated services, strengthen technological innovation or improve operational efficiency, and also seek the support of partners or investors to enhance their competitiveness.
In addition to the free strategy, upgrading the technical level is the core strategy.
In addition, the following measures can be taken to expand the user base: providing high-quality customer service and technical support; Cooperate with industry leaders to develop industry solutions; Improve public acceptance of AI technology through education and training.
"Let more people use AI" is a one-dimensional vision goal at the application level. Theoretically, the popularization of AI technology will bring huge social and economic benefits.
But in fact, the input-output ratio of the big model does not meet this expectation. We are suitable to advocate the long-term goal of defeating the big model from the perspective of bottom-level research and development.
Author | Mei Haoyu | Rao Zufen | responsibilityEdit | He mengfei
Editor | He mengfei | Source | VCG