Mistral and A16z VC Anjnyy Midha said that Deepseek will not stop the hunger of GPU from AI

Andressen Horowitz General and Member of Mistral Anjnyy “Anj” Dibsic jaw performance Six months ago, tells Techcrunch.

That is when Deepsek Coder V2, which competes with GPT4-Toboard from Openai for coding tasks, according to the tasks of coding, according to paper I released last year. He said this places Deepseek on the path of launching improved models every two months through R1. R1 is his new open source thinking The model that raised the technology industry To provide the standard performance of the industry in a small part of the cost.

despite Sell ​​NVIDIA sharesMidha says that R1 does not mean that the founding models of Amnesty International will stop spending billions on GPU graphics unit chips and create more data centers as quickly as possible.

This means that they will do more with the strength of an account they can get.

“When people are like, well, I collected a billion dollars.” “Does Deepseek mean that all of this billion dollars is completely unnecessary? No, in fact, it is very important for them to be able to look at Deepseek’s efficiency improvements, and to absorb it, then throw a billion dollars.”

“Now we can get 10 times more outputs than the same account.”

This does not mean that Mistral is desperate behind Openai and HotHROPIC, as it argues. Each of them raised billions of billions more than Mistral. Openai is said to be in talks to upload another Daw-dropping 40 billion dollars.

Mistra says he is able to compete with them because he is open source. And its area has an advantage. Open Source allows the company to access the free artwork mainly from those who want to help because they use the project. The closed source competitors are guarded by their secrets and they must pay the price of all employment as well as the energy account.

“You do not need $ 20 billion. You just need to calculate more than any other open source model. So Mistral is placed (well). Midi said about his company in his wallet:“ They have the largest number of open source provider.

Llama’s Facebook, the largest AI Open Source Open Western model, will get more investment. CEO Mark Zuckerberg said on Wednesday that he It is still planning to spend “hundreds of billions of dollars” Generally on artificial intelligence. This includes $ 60 billion in 2025 on capital expenditures, most of which are databases.

Oxygen GPU sharing program from A16z “Overbooked”

Midha, who is also a member of the Ai General Black Forest Labs Board of Directors and 3D models maker (Angel in artificial intelligence raises the anthropologist, with eleven and others, another reason for not seeing the hunger of Amnesty International in favor of graphics processing units soon.

He is the leader From the A16Z oxygen program. Graphics processing units, especially on the latest NVIDIA, have become so rare that VC took matters in its hands about a year and a half ago. I bought a group of them for its governor’s companies for use.

Oxygen “now preserved. I can’t customize enough,” laughs. Startups do not only need their graphics processing units to train the artificial intelligence model, but then they need more to run continuous artificial intelligence products for customers.

“Now there is this insecurity demand for consumption,” he explains.

For this reason it is believed that the engineering breakthroughs in Deepsik are also not changing. this A large partnership of $ 500 billion from Openai I announced earlier this month with Softbank and Oracle for artificial intelligence data centers.

The main change of Deepseek in recognition by NATATIANANA, that artificial intelligence is the following constituent infrastructure, such as electricity and the Internet. Midha wants them to think about “the independence of the infrastructure”, as he calls it. Do they want to rely on Chinese models, with censorship and claws in their data? Or do they want Western models that follow Western laws, ethics and adhere to NATO agreements?

It is clear that he is defending Western countries using Western models, such as Mistral based on Paris. Hundreds of companies share this anxiety and I have already prevented Deepseek, It is the service of the consumer application and an open source model.

Not everyone buys this fear of Chinese open sources. Companies can operate locally in their data centers. Deepseek is already available as a safe cloud service from American companies such as Microsoft Azure FoundryTherefore, developers do not have to use the cloud Deepseek service.

In fact, the former CEO of Intel, Pat Gelsiner – someone Good knowledge of China – Tell Techcrunch that the starting of Gloo is a construction Artificial Intelligence Chat Services on its own version of Deepseek R1 Instead of options like Lama or Openai.

But if anyone wants to get rid of the Data Center plans in the Deepseek light, the Midra laughs and has a request: “If you have additional graphics processing units, please send them to anj.”

Techcrunch has a news message focused on artificial intelligence! Subscribe here To get it in your inbox every Wednesday.