There is a fault line in computing power demand
At present, there is a serious shortage of computing power. The main reason for this problem is: GPT3Upgrades from 5 to GPT4 require at least 3 to 5 times more computing power. The growth of this demand is not carried out in accordance with the normal linear pattern, but directly jumps and explodes, resulting in the inability of the domestic computing power required normally. Overall, the shortage of computing power is very serious. Secondly, in the process of upgrading, China also encountered the restrictions of cloud computing blockade, coupled with the complete cessation of computing power leasing, resulting in the interruption of domestic computing power. At the same time, the price of existing computing power is also rising. Therefore, the current situation of large factories is very difficult. Many people wonder if they can continue to get equipment such as A100 and H800, but it is very difficult! Don't think about normal channels, the cost of small batch procurement is very high, and now the only hope is the domestic self-developed computing power chip.
The level of the Internet large model is in GPT35 levels
The large models of the Internet giants have generally reached GPT35 horizontals, most of which are vertical class models, and only a few are generic large models. It is expected that by mid-2024, it is expected to reach the level of GPT4, but the premise is that the domestic computing power can catch up. At present, the domestic computing power capacity is basically transferred to China, and the problem will basically be completely solved in the second quarter, that is to say, in the first quarter, large enterprises will still face the constraints of shortage. The outbreak of the demand side exceeded expectations, taking the east and west as an example, this year's demand should be about 60 billion, and the mobile operators alone occupy the 100 billion level, plus telecom and Unicom, the scale may exceed 50 billion. Since large internet companies have snapped up and upgraded to GPT4 levels, the average demand for more than 60,000 cards per company is only a conservative estimate. In other words, this year, China needs to make up for the card gap of 250 billion, which is really terrifying. Therefore, experts have been emphasizing that computing power will be a limiting factor in the development of artificial intelligence last year.
In the face of the impact of GPT5, how should we respond?
There are more important problems that have not been solved, then it is GPT5, now the United States has fully broken through GPT5, the demand for computing power is basically going to the galaxy, Xiaozha does not say a word to hoard 600,000 H100, such a large demand, domestic chip manufacturers are afraid that they will be unable to breathe for a while, our process factory is non-stop to no avail, how to solve the shortage of computing power, it is a very big problem now, I believe that with the wisdom of our people, it will be solved, and we must not let Lao Mei succeed.
If you like this article, you can click "Watching, Following, Liking, Favorites". 
I will also update the article from time to time, if there are errors and good suggestions to the author, please correct! The data** comes from the Internet, and the infringement must be deleted.