Zero, one, everything falls down?

Overnight, the news that all things were acquired by Ali Cloud spread throughout the AI circle.

An AI blogger broke the news on X, in brief: zero all things collapsed, the team broke up, and the card and training team were sold to Ali.

Kai-fu Lee refuted rumors in the comments section: nothing at all. Three sentences all wrong.

picture

Since then, Kai-fu Lee sent a circle of friends again to refute rumors!

That there is no problem with the funds of zero things, there is no need to sell Ali: in 2024, the confirmed income of zero things is more than one billion, and it will increase by multiple in 2025.

"I don't know why such rumors are being spread."

Finally, Lee Kaifu said, "2025 is the test year of China's big model, the outbreak of application, and the commercialization of the year, we still focus on these really important things!"

picture

It is interesting that Kaifu Lee used three "years" to set the tone for 2025, but it is all about application and commercialization.

Is the pre-training team really gone, and they have given up on the main attack model? What is the truth?

1. Cut the pre-training team, Ari gives the offer

Combined with the information of many parties, it is true that all things have encountered some difficulties.

Since December, there have been frequent changes in the pre-training team and Infra team:

"In mid-December 2024, we laid off the entire pre-trained algorithm team and Infra team, including the Silicon Valley team.

At the end of December 2024, the Zero One everything pre-training algorithm team won the offer of Tongyi team, and the Infra team won the offer of Alibaba Cloud; At present, Alibaba's recruitment of the zero-everything team is in the form of providing jobs, rather than acquisition."

At the end of November last year, when Yiyiall launched its latest flagship model Yi-Lightning, Kai-Fu Lee had refuted rumors of "stopping pre-training" :

"We are the third largest company on this list globally, behind OpenAI and Google. We are very pleased to know that we must be in the world's first echelon, which is very important to us, so that the team is very excited, because we know that our model can play domestic and foreign markets, no one need be afraid.

It was a positive feedback for our team and gave us the confidence to move forward and continue to model and continue to do pre-training."

However, by the beginning of January 2025, Lee Kaifu had changed his tone at an internal meeting, saying that the capital of all things was tight, the capital invested in pre-training was relatively large, and the company planned to transform to do applications.

It is worth noting that only in August last year, 01everything also received a new round of financing, reportedly amounting to hundreds of millions of dollars. People familiar with the matter said that the participants in this round of financing include an international war investment, Southeast Asian consortia and other institutions.

It seems that AI pre-training is no different from a bottomless pit of money for startups. Even if all things do not choose to "sprinkle money", it is also known that the income and expenditure of the application is basically balanced, but it still cannot maintain a benign cash flow.

2. The beginning of the rumor: a week ago, zero one just announced cooperation with Ali Cloud

On January 2, the company announced its cooperation with Aricloud.

It was announced that the two sides will establish a "joint laboratory of Industrial large models" to jointly accelerate the landing of large models from technology to application and further expand the ecological integration of industrial large models.

It is reported that the industrial large-scale model joint laboratory of the deep strategic cooperation between the two sides includes technology, business, talent and other sectors.

In the official refuting rumors, it also confirmed the deep cooperation between the two sides.

picture

In yesterday morning, there was a small red book blogger released this news, and suggested that "zero or merged into the Tongyi team", which may be the earliest source of zero rumors.

In the First financial report, it also interpreted the cooperation as "equivalent to Ali Cloud using cooperation - rather than acquisition - to merge the technology, business, talent and other sectors in the field of to B."

If everything in this way "secret warehouse", can adjust the direction of the entire company, more focused on their own To C business.

The company currently has four AI products for overseas markets: PopAi, Monaland, Shado and Bingo.AI.

Among them, PopAi is an office product, Monaland's main character accompany. According to the data provided by Kai-Fu Lee in August 2024, the users of these products were close to 10 million within nine months of their launch, and the revenue was expected to exceed 100 million.

The To C team may also be in for a shake-up. According to media reports, in addition to the revenue and growth performance of the sea application PopAi can be self-financing, some of its applications have been split team, will be in the form of subsidiaries for external financing.

3. In the year of the big model test, surviving is the hard reason

The AI world is changing.

Kai-fu Lee, the founder of all things, has refuted rumors that pre-training will not end, but in the face of the current tightening of funds, focusing on applications or has become an unavoidable choice.

The year 2025 is destined to be the battle of life and death for AI applications, and this "great escape" is completely cruel to players on the field.

However, this sense of urgency may force the development of models to break out of the cycle of "burning money". Truly pay attention to user needs, and "kill" AI products and services that can truly provide long-term value.

Before the coming of the next capital winter, we may see an accelerated reshuffle of the survival of the fittest.

Reference link:

1.https://mp.weixin.qq.com/s/NDXdK1XyFDkA_Ni9aw8HKA

2.https://mp.weixin.qq.com/s/k80pNvEMtST3gah1_V0huw

To learn more about AIGC, please visit:

51CTO AI.x community

https://www.51cto.com/aigc/