With Meta calling it a «strategic partnership,» Wang is expected to be joined by «a small number of Scale AI employees,» writes CNBC.
They will also get access to critical infrastructure and a steady pipeline of expert engineers.
Lots of AI experience
Scale AI is a major infrastructure provider for AI companies, providing anything from basic data labelling, «red teaming» and safety to optimizing for benchmarks.
It also develops its own large language models, which it sells to governments — including Qatar and the U.S. Department of Defense.
They employ many of the leading AI researchers in the world — and that’s precisely what Meta is after, as it scrambles to rebuild its in-house AI capabilities following a wave of high-profile departures after LLaMA 3.
Meta’s AI woes
Lackluster response to later models, less than 10 million downloads for the Meta AI app, getting caught cheating at LMArena, later privacy criticism of the Meta AI app, and the delaying of the Behemoth model have almost left Meta’s AI efforts in disarray, and the hope is now to revive it in big swoop.
To put the $14.3 billion in perspective, it is one of the largest acquisitions in Meta’s history, second only to the WhatsApp purchase in 2014 at $19 billion dollars. Instagram was bought for $1 billion in 2012 and Oculus came into the fold for $2 billion in 2014.
The purchase is also notable in the sense that it is only for 49 % of the company, getting access to the talent, but hoping to avoid regulatory scrutiny and the competition woes of flat out ownership.
Scare AI does not only have sensitive government contracts, but also does work for Google and Microsoft. This might be expected to continue.
UPDATE (June 14th, 2025): Reuters are now reporting that Google, OpenAI, xAI and Microsoft are looking the exit or tone down their deals, fearful of giving a competitor advance knowledge of their models and training.
Read more: CNBC, Associated Press and The Washington Post. Ars Technica on the superintelligence group at Meta.