Mistral board member and a16z VC Anjney Midha says DeepSeek won’t stop AI’s GPU hunger
Andreessen Horowitz General Partner and Mistral Board member Anjney “Anj” Midha spied the deep performance of Deepseek for the first time six months ago, he says WAN.
That is when Deepseek introduced V2, which, according to the GPT4 turbo from OpenAi, was for coding-specific tasks, according to a paper It was released last year. This put Deepseek on a path to release improved models every few months by R1, he said. R1 is the new Open Source Reasoning Model that has increased the technical industry for offering standard performance in the industry against a fraction of the costs.
Despite the sale of the shares of NVIDIA, Midha says that R1 does not mean that AI-Fundamental models will stop publishing billions to babble GPU chips and build more data centers as fast as they can.
It means that they will do more with the computing power they can obtain.
“If people are like that, okay, Mistral has collected a billion dollars,” he says. “Does Deepseek mean that all that billion dollars is completely superfluous? No, it is actually extremely valuable for them to be able to look at Deepseek’s efficiency improvements, they internalize and then throw in a billion dollars. “
He adds: “Now we can get 10 times more output from the same calculation.”
That does not mean that Mistral is hopeless behind Rivals OpenAi and anthropic, he argues. Each of them has raised many more billions than Mistral. OpenAi is said to be in conversation to pick up a stunning $ 40 billion.
Mistral remains competitive with them because it is open source, he says. And his logic has merit. Open Source gives a company access to essentially free technical labor of those who want to help because they use the project. Rivals with closed source guard their secrets and pay for all work and the power of the calculation.
‘You don’t need $ 20 billion. You only need more calculation than any other Open Source Model app. So Mistral is positioned [well]. They have the most calculations of every open source provider, “said Midha of his portfolio company.
Facebook’s Llama, the largest Western Open Source Ai Model Rival from Mistral, also gets many more investments. CEO Mark Zuckerberg said on Wednesday that he is still planning to spend “hundreds of billions of dollars” on AI. This includes $ 60 billion in 2025 on capital expenditures, mainly data centers.
A16Z’s Oxygen GPU part program “OverBooked”
Midha, who is also a board member for AI image generator Black Forest Labs and 3D model maker Luma (and an angel in Ai Outfits Anthropic, Elevenlabs and others) has another reason why he does not see Ai’s hunger to GPUs who came down at any time. .
He is the leader of the A16Z oxygen program. GPUs, in particular the state-of-the-art H100s of Nvidia, have become such a scarce goods that took over the VC company of Hevel from its own hands about a year and a half ago. It bought some of them for his portfolio companies to use.
Oxygen is now ‘overbooks. I can’t assign enough, “laughs Midha. His startups not only need GPUs for AI modelt training, but then they need even more to run their current AI products for customers.
“Now there is an insatiable demand for conclusion, to consumption,” he explains.
That is also the reason why he thinks that the technical breakthrough of Deepseek will also not change Stargate. That is OpenAi’s BIG $ 500 billion partnership that was announced with Softbank and Oracle earlier this month for AI Data Centers.
The most important change deeper in nation states is recognition that AI is the following fundamental infrastructure, such as electricity and internet. Midha wants them to consider ‘infrastructure independence’ as he calls it. Do they want to rely on Chinese models, with its censorship and claws in their data? Or do they want Western models that follow Western laws and ethics and adhere to NATO agreements?
He clearly argues for Western countries with the help of Western models, such as his Mistral -based Mistral. Hundreds of companies share that concern and have already blocked Deepseek, which is both a consumer app service and an open source model.
Not everyone buys that fear of Chinese open source models. They can lead companies locally in their own data centers. And Deepseek is already available as a safe cloud service from American companies such as Microsoft Azure FoundrySo developers do not have to use the cloud service of Deepseek.
In fact, the former CEO of Intel, Pat Gelsinger – someone well familiar with China -WAN told that its startup -Gloo AI chat services builds on their own version of Deepseek R1 instead of choices such as Lama or OpenAi.
But if someone wants to throw away his data center plans in the light of Deepseek, laughs Midra and has a request: “If you have extra GPUs, send them to Anj.”
WAN has an AI-oriented newsletter! Register here to get it in your inbox every Wednesday.
