After selling for $14 billion to Meta, Scale AI surprisingly thrived even more

Almost a year ago, when Alexandr Wang left the company he founded, Scale AI, the general consensus was that this startup was essentially at the end of the road. But now, under the leadership of new CEO Jason Droege, Scale AI’s revenue has rebounded, and it is expected to surpass $1 billion this year.

In June last year, Alexandr Wang slowly descended the stairs of Scale AI’s open courtyard at its San Francisco headquarters. Employees were uncertain whether he was still the company’s leader at that moment. Just the day before, this data annotation giant announced a major deal: Meta would acquire a 49% stake in Scale AI for $14 billion, with founder and CEO Wang stepping down to lead Mark Zuckerberg’s newly established superintelligence lab. Once the news broke, chaos erupted inside the company, with many employees believing he had already left. So when Wang appeared, everyone was surprised, applauding him and welcoming him to speak at the all-hands meeting. Now, in an interview with Forbes, Wang admitted, “I was brought to tears on the spot. In another parallel universe, I would be more than willing to stay at Scale AI.”

He can no longer recall exactly what he said at the meeting, but an attendee revealed that Wang began by recalling his experience founding Scale AI as a freshman at MIT, then couldn’t help but tear up. The person recalled Wang muttering, “How stupid am I, what am I even doing?”

Though Wang’s lament was about his own tears, it also precisely expressed the common doubts outside. The major deal leaked days earlier had raised the same question across Silicon Valley: at that time, Scale AI was valued at $13.8 billion and growing rapidly, so why would the founder give up his own company to join Meta? It’s worth noting that in the AI race, Meta has been striving to catch up behind Google, OpenAI, and Anthropic.

Originally, Scale AI was a leading company in the data annotation sector, gathering massive crowdsourcing annotators along with PhDs, lawyers, engineers, and other professionals to provide training data for top AI models like Google’s Gemini and OpenAI’s ChatGPT.

Since delivering large language model technology to the Pentagon for the first time in 2020, Scale AI has also become an important partner for the U.S. Department of Defense.

The company also once made Wang the world’s youngest self-made billionaire. But with Scale AI’s tie to Meta, many thought it might jeopardize its solid business of providing infrastructure for leading AI companies. The industry wondered: which cutting-edge lab would be willing to entrust its core data to a company in which nearly half the shares are owned by Meta?

Wang took about ten key employees from Scale AI to join Meta. Now, he believes this partnership is a “rare opportunity” for both companies. In the second half of the all-hands meeting, Wang officially introduced his successor, Jason Droege. Droege previously served as Scale AI’s Chief Strategy Officer and had earlier worked at Uber and the taser manufacturer Axon. Despite ongoing turbulence the week before, the meeting was concise, lasting only 30 minutes, with no Q&A session. Wang and Droege spoke briefly throughout. Droege recalled, “We shook hands, hugged, and were full of hope for the future, then each went on to new journeys.” Currently, Droege still holds the interim CEO title, but internally it’s understood that he is the long-term leader.

Nearly a year has passed since the announcement of this deal, and Scale AI, founded in May 2016, has now entered its tenth year. Unsurprisingly, the company now shows a very different face. Droege’s first move after taking over was to adjust the company’s investment focus: no longer relying solely on traditional data annotation, but instead helping large enterprises like Ernst & Young, Paramount, Cisco, as well as the U.S. military and other public sector clients build internal AI applications.

This strategy appears to be paying off. Scale AI told Forbes that its revenue has grown from $870 million two years ago to nearly $1 billion last year. Since its founding, the company’s total revenue has reached $2.5 billion. Much of this growth is attributed to its new shareholder, Meta. According to Forbes, as part of the partnership agreement, Meta is committed to paying Scale AI at least $450 million annually over the next five years, or at least half of the company’s semi-annual AI expenditure (whichever is lower). Wang and Droege declined to comment, but this fee nearly accounts for half of Scale AI’s annual revenue.

Droege also did not disclose the revenue split between data services and AI application services. He said that data annotation still remains the company’s main source of income, but he expects that within 18 months, revenue from application services will surpass traditional data annotation. Wang, who is now at Meta, still cares deeply about his original company. He stated, “Some people think Scale AI is declining or at the end of the line, but that’s completely untrue.” He believes the resilience shown by Scale AI has thoroughly overturned market stereotypes.

Wang’s parents worked at Los Alamos National Laboratory, the birthplace of the atomic bomb.

Ten years ago, Wang co-founded Scale AI with Lucy Guo. Guo dropped out of Carnegie Mellon University after being selected for the Thiel Fellowship, which encourages young entrepreneurs to drop out and start businesses, funded by billionaire Peter Thiel. Both had previously worked at Quora, where they met. Before settling on AI infrastructure, they had brainstormed several startup ideas.

Guo revealed in a podcast called “Aspire” by entrepreneur Emma Grede last year that the two later had disagreements over the company’s future direction. The final spark was Guo’s claim to someone that she thought Wang “should be fired.” She speculated in the podcast that this person later told Wang about her remark. “I was very upset about that person passing on the message, after all, I thought we had a basic trust. I felt betrayed at the time.”

Scale AI co-founder Alexandr Wang leaves the company to join Meta. Photo credit: Ethan Pines for Forbes

Ultimately, Lucy Guo chose to leave the company. When asked about this past, Wang smiled awkwardly: “We rarely talk about it. When you’re in Y Combinator, no one tells you how common founder splits are. But we eventually resolved our differences, and I am genuinely proud of everyone on the team. Even after such upheaval, the company has continued to move steadily toward success.”

Seven years later, it’s Wang’s turn to leave the company he built himself. The first time he was approached by Zuckerberg was last spring, when Meta’s flagship large model Llama 4 underperformed expectations, prompting Zuckerberg to start looking for a new AI leader. The two had known each other for years; Wang had previously asked Zuckerberg for advice on running a startup. When asked how Zuckerberg made the invitation, Wang downplayed it: “I don’t really see what’s so hard to understand. It’s just that Meta has a great opportunity in AI, with a very promising outlook, so we hit it off and discussed possible collaboration.” He added that Zuckerberg’s belief in AI “is very clear and firm.”

Wang then sought his company’s board’s opinion, weighing the pros and cons. Investor Mike Volpi said, “For major deals like this, it often comes down to a close balance of advantages and disadvantages.” Volpi, a partner at Index Ventures and an early investor in Scale AI, had been discussing the deal with Wang long before the terms were finalized: “We focused less on the specific process of the deal and more on what Alexandr ultimately wants to achieve in life.”

In June last year, the $14 billion deal was officially announced, and both investors and employees reaped huge wealth. According to insiders, most employees received bonuses roughly equal to half of their equity value.

Some employees felt disappointed about Wang’s departure, which was understandable. When asked if he felt guilty about leaving, Wang said, “When things happen, there are a lot of emotions, and I completely understand that. But I’ve always believed that the team is capable of extraordinary achievements.” The company also initiated layoffs. Two months after the deal was announced, in August, Scale AI laid off 200 full-time employees, about 14% of the total, and terminated cooperation with 500 contractors (these were corporate contract workers, not the same as the data annotators paid per task). A company spokesperson said that the current number of full-time employees is about 1,300, with plans to hire another 500 by the end of this year.

Before officially signing with Meta, Scale AI had anticipated that the deal would impact its relationships with cutting-edge AI labs. Droege said, “We expected turbulence, no doubt about it. After all, the facts are clear: the founder jumped ship to another AI lab, working for a competitor’s client.” He added, “Most existing clients eventually returned to cooperation, but not all.”

To dispel concerns that Scale AI might be constrained by Meta, Droege launched a charm offensive to reassure partners. Some clients wanted to visit the company’s offices and speak directly with management and R&D staff, while others asked for detailed explanations of the deal’s terms. Droege emphasized that Meta would not gain a seat on the company’s board; Wang still retains a board position at Scale AI, but neither past nor current directors have access to details of the company’s collaborations with clients. Droege now states, “I am also a shareholder of Scale AI. My motivation is to ensure the best experience for non-Meta clients. I am not an employee of Meta.”

According to two insiders familiar with the cooperation, the most notable lost client is OpenAI. Losing this company was a huge blow, not only because of OpenAI’s industry prominence but also because it was one of the earliest top labs to bring Scale AI into generative AI, assisting in training GPT-3 before ChatGPT ignited the global AI craze in 2022. Additionally, sources say that after Meta’s investment was announced, Google temporarily suspended its data annotation partnership with Scale AI but resumed cooperation after a few months.

Of course, one client need not worry about losing business—Meta. During Wang’s tenure, Scale AI played a key role in training the first model, Muse Spark, released in April. The model received mixed reviews but was generally positive, seen as Meta’s comeback in large-model competition. Wang said of the new model, “This is the result of dedicated effort. Scale AI is a crucial partner. The success of Muse Spark wouldn’t be possible without the joint efforts of all collaborators, and Scale AI is definitely an important part of that.” Beyond that, Wang declined to discuss his work at Meta further.

Since the Meta deal closed, Scale AI has been working to reshape its business.

Droege explained that before the partnership, 70% of the company’s business was data annotation, and 30% was building AI applications for enterprises and governments. Now, the business focus has reversed. By the end of last year, annual revenue from enterprise AI applications had reached $200 million. The company has also added many large clients, including Mayo Clinic, BP, and Allianz. For example, Scale AI collaborated with Mayo Clinic to develop a system capable of reading and analyzing fragmented medical records. Tony Qui, CTO of Ernst & Young’s strategic consulting division, said EY is using Scale AI to build internal AI agents, including one that assists with due diligence in M&A transactions. Qui explained that EY initially chose Scale AI based on a recommendation from OpenAI. When news of Meta’s investment broke weeks later, the company had some concerns but ultimately decided to continue cooperation after meetings with Droege and other executives.

Scale AI claims that balancing enterprise services and data annotation is its unique competitive advantage. The company has long trained large models for cutting-edge AI labs and is deeply familiar with industry-leading technological trends—precisely the technologies that many enterprises are eager to deploy. Venture firm Coatue partner Lucas Swisher, who led the Series B investment in Scale AI, commented, “Only Scale AI can truly offer both types of services simultaneously.”

Scale AI has also further expanded its work with the U.S. government, continuing the development path Wang set. In early May, the U.S. Department of Defense awarded Scale AI a $500 million contract to lead the Thunderforge project, which aims to incorporate AI agents into military mission planning. In April, Scale AI was also selected as a contractor for the U.S. “Golden Dome” defense program alongside Palantir and Anduril. This is a missile defense system project costing $185 billion under the Trump administration, but both Scale AI and the Pentagon declined to disclose specific responsibilities.

Cameron Stanley, the DoD’s Chief Digital and AI Officer, praised Scale AI’s performance in serving the department, especially in data annotation. He noted that early involvement in the Maven project, which integrated machine learning into military intelligence workflows, demonstrated its capabilities in computer vision. He lauded Scale AI’s metadata organization as “top in the industry.” Its advantage lies in helping the Pentagon integrate scattered datasets and understanding how to align with military administrative processes to ensure smooth project execution. Stanley told Forbes, “They can gather massive heterogeneous data, clarify its logic, and organize it into structures suitable for training algorithms—such skills are very rare.” He refused to comment on whether Scale AI’s technology has been deployed in actual combat scenarios, and the company’s spokesperson did not disclose details of military cooperation.

Droege emphasized that the company’s reduced emphasis on data annotation is not due to concerns about Meta’s involvement capping growth, but because the overall growth rate of the data annotation industry has slowed. In contrast, the enterprise AI application sector offers vast market potential as major tech giants pivot toward AI.

Regardless of the reason, competitors like Surge, Mercor, Handshake, and Invisible have seized the opportunity to expand their market share, openly competing for contracts. Legal documents show that last year, Mercor hired away a former Scale AI employee suspected of stealing strategic business secrets. The lawsuit was settled in January this year, with undisclosed terms, and no media reports had covered the incident before.

A CEO of a rival company privately admitted that since the Meta deal, Scale AI has gradually faded from the radar in large-scale bids for data annotation contracts: “I haven’t heard much from them in recent months, and industry talks about partnerships rarely mention this company anymore.”

Meanwhile, Scale AI is actively advancing its IPO plans.

When the Meta deal was struck, going public was already a consideration. Early investor Mike Volpi said, “In some sense, this investment could still yield dual returns.” That is, the remaining part of the company could go public independently, allowing investors to realize another exit. Droege was cautious about timing: “Scale AI will very likely go public in the future, but it’s still in very early planning stages.”

Today, Scale AI is on an unprecedented growth trajectory. In recent years, many AI startups and tech giants have engaged in high-profile acquisition and talent-harvesting deals. Founders of hot AI companies like Inflection, Adept, Character, and Windsurf have collectively jumped to Microsoft, Amazon, and Google. But Droege firmly disagrees with categorizing Scale AI among these companies.

Why? He explained, “We already have mature business operations, whereas those companies do not yet have a proven business model. There’s simply no comparison.”