Can Scale AI and Alexandr Wang reignite Meta’s AI efforts?

Meta would invest almost 15 billion dollars In the data rebalancing company, the AI ​​scale and a 49% stake in the startup, while invoking CEO Alexandr Wang to help direct a new “superintendent” laboratory within the company.

The agreement recalls Meta’s important and risky bets, such as its Acquisition of $ 19 billion in WhatsApp and his Instagram purchase of $ 1 billion. As these mergers have closed, many people suggested that Meta had been widely paid for platforms – and today’s speech is no different. There is no shortage of investors and founders who have striked this weekend on Meta’s last link.

In the end, WhatsApp and Instagram became an integral part of the Empire of the CEO Mark Zuckerberg. The question is whether the AI ​​scale agreement will work in the same way in favor of Meta, proving again the premonitory strategy of Zuckerberg, or if the company grabs peaches in an erroneous effort to make up for its rivals like Openai, Google and Anthropic.

In this case, META does not bet on an emerging application of social media, but on the data used to train the best models of AI. In recent years, the main AI laboratories such as OpenAi have relied on the scale AI to produce and label the data used to form models. In recent months, the AI ​​scale and its data annotation competitors have started to hire highly skilled people, such as senior software doctoral students, to generate high -quality data for border AI laboratories.

He can benefit Meta to have a close relationship with a data scale. Meta leaders complained about a lack of innovation around data in the main AI teams of the company, according to a person familiar with the issue.

Earlier this year, Meta’s generative AI unit Launched Llama 4, a family of AI models This failed to correspond to the capacities of the Chinese models of the AI ​​Deepseek laboratory and which was largely considered to be a disappointment. Do not help things, Meta tries to fight against an attrition problem. According to data compiled by signalfire, Meta lost 4.3% of its best talents at AI laboratories in 2024.

Meta is not only betting on AI on a scale to rekindle its AI efforts, but also on Wang to lead the new aforementioned superintendent team. The 28 -year -old CEO turned out to be a solid founder of Startup – he is known in Silicon Valley as ambitious, a good seller and very well connected. In the past few months, Wang has been meet world leaders To discuss the impact of AI on society.

However, Wang did not direct an AI laboratory of this kind, and he does not have the same context of research on AI as many other AI laboratory chiefs, such as Ilya Sutskever de Safe Superintelligence or Arthur Mensch of Mistral. This may be why Meta is also supposed to recruit large-scale talents Like Jack Rae de Deepmind To complete its new AI research group.

The post-acquisition fate of the scale of the scale, society is a little troubled. The role of real world data in the training of IA models-some AI laboratories have led to internal data collection efforts, while others have increased their dependence on synthetic data (ie generated by AI. In April, information indicated that The AI ​​scale had missed some of its income targets.

According to the co-founder of Anyscale, Robert Nishihara, several Laboratories Frontier AI explore new ways of taking advantage and optimizing data, many of which are enough at high calculation intensity.

“Data is a motion target,” Nishihara told Techcrunch in an interview. “It’s not just a finished effort to make up for it – you have to innovate.”

It is possible that Meta and Wang’s relationship can scare other AI laboratories that have traditionally worked with the AI ​​scale. If this is the case, this agreement could be a boon for competitors of the AI ​​scale, such as Turing, Surge IA and even unconventional data suppliers such as the recently launched LM arena.

The CEO of Turing, Jonathan Siddharth, told Techcrunch by e-mail that he had received an increased interest from customers in the light of rumors concerning the meta agreement with the AI ​​scale.

“I think there will be customers who prefer to work with a more neutral partner,” he said.

Only time will tell us how Meta’s investment will take place for its IA efforts, but the company has important land to catch up. Meanwhile, the competition does not slow down. OPENAI is preparing for the release of its next flagship model, GPT -5, as well as its first openly available model for years – a model that will compete with the current and future Meta versions.