Scale AI's Alexandr Wang in the Driver's Seat With Meta Deal
Plus, startup Handshake and others take aim at Scale as human training of LLMs becomes more critical
Wang Sells But Keeps Control of Scale AI
Once news of Meta’s investment in Scale broke, the myriad discussions rattling around the startup and AI world — and even among Scale’s own investors — could be boiled down to this: is this an Inflection, or is it an OpenAI?
That is to say, did Meta invest approximately $15 billion in Scale to supercharge an outside startup while reaping some of the benefits and casting off its image as an AI stumbler (Microsoft-OpenAI). Or did it poach the CEO and top leadership of a startup to bring that talent in house, while paying off investors and sucking the life-force from the remaining company, (Microsoft-Inflection).
Scale’s pitch to its investors was the former, one of them told us. And a key way to understand that is that Meta is essentially giving co-founder Alexandr Wang control of Scale, even as he steps away from the company he co-founded.
Meta intends to buy a 49% stake in Scale but will transfer its voting rights to Wang, according to a person familiar with the terms. Previously Wang did not have voting control over Scale, this person said.
The roughly $15 billion from Meta is structured as a dividend so investors can get some cash back to their LPs while retaining some upside.
Scale is surely set to lose business from Meta rivals such as Google and OpenAI, which are major customers. But Meta argued that its sales team could more than make up for that by selling Scale’s AI data labeling services to other enterprises. It’s similar to Microsoft’s arrangement where it sells OpenAI’s models to enterprise customers via its Azure sales team.
Not every investor bought the spin. “There’s not a future of Scale as Scale,” a person at another firm that backed the company at a later stage told me. This person said it was a great outcome for Scale but this was essentially an exit.
Scale investors were pumping the firm to us last year as one likely to go public. The company has been expected to reach $2 billion in revenue this year, and Wang had used its success to ensconce himself as a major name in AI, appearing in DC and flying out to Saudi Arabia to get face time with President Trump.
The sector is getting much more competitive though, with new rivals like Handshake crowding into the space and high-level experts increasingly needed for what had once been rote labelling tasks. Scale has talked about branching out into AI applications, but that would require further building an enterprise sales team, among other challenges.
Meta needed to do something. Its big Llama 4 model Behemoth was a disappointment, and it’s behind in other areas like reasoning, video and voice. Mark Zuckerberg wants to compete in all these areas, according to a person familiar with the matter, and has been in regular meetings with the AI team to monitor its progress closely.
He blew up the division’s structure amid the Llama 4 fallout and is assembling a so-called Superintelligence team, which Wang is now expected to lead. But there’s still a lot of confusion within the AI team about how it will all work, we were told.
Zuckerberg and Wang had been close for some time, one person close to Scale told me. The idea is for the two to combine their visions and build their respective companies together, an approach that strikes us as both promising and risky.
If the question is whether Meta-Scale is an OpenAI or an Inflection, the final answer will likely end up being: yes.
The combination could end up being a synthesis of the Inflection and OpenAI deals—taking the top executives out of a successful business and hoping the remaining company still flourishes.
— Tom Dotan
Career Startup Handshake Sees Explosive Revenue Growth from High-Skill AI Training
Last year, Handshake, a 10-year-old professional networking startup backed by Kleiner Perkins, noticed that companies like Scale AI which offered human-curated data for AI training were aggressively using their recruiting platform, seeking PhDs who could help tune foundation models for the likes of OpenAI and Anthropic.
So Handshake decided to get into the business itself. The company had built a network for recent college graduates looking for jobs, and boasted over 1 million employers and 18 million job-seekers on the platform. AI-training gigs could be another earnings opportunity for recent grads—and after launching the matching service in January, Handshake expects the new line of business to hit nearly a $100 million run rate by the end of the year.
“The inbound today was on a whole new level,” said Garrett Lord, CEO of Handshake. “We’re already working with the top labs, and it’s clear the industry knows we have the experts they need.”
Handshake came to Newcomer exclusively to announce its foray into the business of recruiting talented humans to train AI models.
The quick ramp reflects the AI industry’s increasing need for human expertise in fields ranging from software engineering and financial analysis to pharmaceutical research, education, law, and medicine. Handshake is seeing interest in more subjective topics as well like music theory and English literature.
Scale co-founder and CEO Alexandr Wang, who is now going to work for Meta as part of a multi-billion-dollar deal, was among the first to see the opportunity in providing clean and human-labeled datasets to AI training. But it’s been facing more competition as foundation model companies try to make their LLMs smarter in ever-more-specialized domains of knowledge.
Mercor, an AI-recruiting startup founded by two 21-year-olds, blasted onto the scene early this year, raised at a $2 billion valuation. The secretive company Surge AI has been running a bootstrapped business of its own.
There’s no question that Scale is the juggernaut of the sector. It’s now trying to sell a story that it could get into applications and move beyond just the grunt work of throwing humans at foundation models problem areas, though its new alliance with Meta could complicate the pitch.
Handshake is betting that a shift in demand to highly specialized domains is the big opportunity.
The company sources candidates from its network and then can help train and manage domain experts with the full alphabet soup of degrees, including BAs, BFAs, JDs, MBAs, MFAs, MAs, PhDs. Whether this sort of work can help compensate for the anticipated loss of white-collar jobs to AI is a big question for today’s graduates.
— Eric Newcomer