Meta’s partnership with Scale AI shows early signs of strain

Meta’s billion-dollar AI gamble

In June 2025, Meta stunned Silicon Valley with a $14.3 billion investment in Scale AI. The move wasn’t just financial; it brought Scale’s CEO, Alexandr Wang, and top executives into Meta’s new Superintelligence Labs.

The deal was pitched as a shortcut to leapfrog OpenAI and Google by pairing Meta’s compute power with Scale’s data labeling expertise.

Just months later, signs of strain surfaced, raising questions about whether Meta’s most ambitious AI partnership is already faltering.

A fast executive departure raises eyebrows

One of the first cracks appeared when Ruben Mayer, Scale AI’s former Senior Vice President of GenAI Product and Operations, abruptly left Meta after only two months.

Mayer was expected to help oversee AI data operations but never became part of TBD Labs, the core group tasked with building Meta’s most advanced AI.

His exit highlights the difficulty of integrating Scale executives into Meta’s complex structure and raises doubts about whether the promised leadership synergies are materializing.

Researchers favor Scale AI’s rivals

Perhaps most damaging for the partnership, Meta’s top AI research team, TBD Labs, reportedly prefers data from competitors Surge and Mercor over Scale AI’s.

Despite the multibillion-dollar deal, researchers describe Scale’s output as lower quality and lean toward rival vendors for critical training datasets.

That means even as Meta invests heavily in Scale, much of its cutting-edge AI development is being fueled by competitors. This awkward contradiction undercuts the entire premise of the partnership.

Scale AI’s crowdsourcing model shows limits

Scale AI initially built its reputation on crowdsourced data labeling, using large pools of low-cost workers for simple annotation tasks. This worked well for earlier AI models that needed massive quantities of basic data.

But today’s cutting-edge systems require nuanced, expert-level annotations from medicine, law, or science professionals.

Competitors like Surge and Mercor were initially designed around expert networks, leaving Scale struggling to retrofit its model for high-end requirements.

The Outlier platform isn’t enough yet

To address its data quality challenge, Scale AI launched Outlier, a platform to recruit domain experts to provide specialized training data. But the shift has proven difficult.

Transitioning from low-cost crowd labor to expensive, highly skilled talent requires entirely new business models, quality controls, and recruiting strategies.

Surge and Mercor emphasize expert networks, which some observers see as giving them an advantage. At the same time, reports suggest that Meta researchers have shown interest in alternatives, raising questions about whether Scale can meet their needs.

Meta insists Scale’s data is fine

Despite internal pushback, Meta’s spokespeople have publicly denied that Scale’s data quality is an issue. They argue the partnership remains strategically valuable and point to Scale’s established track record.

Yet the optics are tough: if your most important AI researchers openly prefer working with rival vendors, it undermines confidence in the deal.

Meta seems determined to defend its investment for now, but convincing skeptical researchers may prove more challenging than defending a press statement.

Scale AI loses key customers after Meta deal

The fallout isn’t limited to Meta’s walls. Soon after the investment, both OpenAI and Google cut ties with Scale AI, citing concerns about working with a vendor so deeply tied to a competitor.

For Scale, losing two of the most critical customers in AI was a devastating blow. The departures highlight how the Meta deal, intended to elevate Scale, has ironically narrowed its customer base by alienating the broader AI ecosystem.

Layoffs deepen concerns about Scale’s health

In July 2025, just weeks after Meta’s investment, Scale AI announced layoffs of 200 employees in its data labeling business.

New CEO Jason Droege blamed “shifts in market demand,” but industry observers tied the cuts directly to the loss of OpenAI and Google contracts.

While Scale is trying to pivot toward government and enterprise work, the layoffs underscore how vulnerable the company has become even as it secures one of the biggest checks in AI history.

Government contracts offer some relief

Scale AI has leaned into government work to offset its shrinking commercial business. It landed a $99 million contract with the U.S. Army in July.

While such deals provide stability, they are slower-moving and less glamorous than high-profile AI research partnerships.

For Meta, that’s little consolation; its AI teams need cutting-edge, domain-specific data for advanced model training, not defense contracts. The mismatch raises questions about whether Scale is the right partner for Meta’s ambitions.

Was the deal really about Alexandr Wang?

Some insiders suggest that, in addition to Scale’s data capabilities, Meta may have been motivated by the opportunity to bring in Alexandr Wang’s leadership and reputation.

By bringing him in to lead Superintelligence Labs, Meta hoped to inject entrepreneurial energy and attract top AI talent.

Wang, who founded Scale in 2016 at age 19, has credibility in the AI world. However, critics say he lacks the deep research background to guide advanced model development.

Superintelligence Labs struggles with a culture clash

Meta’s Superintelligence Labs has quickly become a battleground of organizational cultures. Startup veterans from Scale and OpenAI clash with Meta’s entrenched corporate bureaucracy.

Researchers complain about endless approvals, shifting priorities, and unclear leadership structures. Such dysfunction is a serious setback for a team tasked with building world-changing AI.

Several recruits have already left, citing frustration with Meta’s environment, while longtime Meta AI staff feel sidelined by the influx of outsiders.

Key researchers are leaving Meta’s AI team

The instability is showing in departures. Rishabh Agarwal, a respected AI researcher at MSL, announced he was leaving despite calling the team’s mission “incredibly compelling.”

Director of product management Chaya Nayak and research engineer Rohan Varma also recently exited. These losses erode Meta’s ability to compete in a high-stakes AI race where top researchers are scarce.

Talent churn so soon after the Scale partnership suggests Meta struggles to retain the people it needs most.

Meta’s Llama 4 disappointment raised the stakes

Meta’s launch of Llama 4 in April 2025 failed to match the performance of rivals’ models, according to reviews from industry analysts. The model fell short of expectations and failed to match OpenAI’s or Google’s latest releases.

Reports say Mark Zuckerberg grew increasingly frustrated with the AI team’s lack of breakthroughs. The Scale partnership was intended as a reset, a way to improve data pipelines and attract elite talent quickly. But if anything, it has added chaos rather than clarity.

Meta diversifies data vendors despite investment

Even after its $14 billion outlay, Meta isn’t putting all its eggs in the Scale basket. TBD Labs continues to source training data from Mercor and Surge, competitors that staff their projects with expert annotators.

This diversification reduces risk but also undercuts Scale’s value proposition. If Meta researchers already prefer working with rivals, then Scale becomes less of a core partner and more of an expensive symbolic investment.

What Meta hoped to achieve with Scale AI

For Zuckerberg, the partnership was supposed to solve several problems at once: better training data, new leadership under Wang, and fresh talent inflows.

In theory, combining Scale’s data pipelines with Meta’s compute could accelerate the development of AI superintelligence.

But reality has proved more complicated. Instead of unity, Meta faces talent churn, cultural clashes, and researcher skepticism. The contrast between expectation and execution is stark, and it leaves Meta still scrambling to close the gap with rivals.

See how Meta is trying to regain momentum by bringing in two more elite AI researchers from OpenAI.

Can Meta salvage this troubled partnership?

The big question is whether Meta and Scale can turn things around. Meta must stabilize its AI teams, regain researcher trust, and decide how central Scale is to its plans.

Scale must demonstrate that its shift toward expert-driven data via Outlier can meet enterprise-level quality demands before risking further erosion of trust.

If Meta’s next AI model, expected in late 2025, fails to impress, the Scale partnership may be remembered as one of Silicon Valley’s most expensive missteps.

Find out why Scale AI is cutting 14% of its staff and what the CEO revealed in a candid email.

What do you think is causing the partnership between Meta and Scale AI to weaken? Please share your thoughts and drop a comment.

Read More From This Brand:

• Meta teams up with Scale AI in $14.3B deal and a big leadership shakeup
• Meta’s CEO security budget tops those of Apple, Nvidia, and more combined
• Meta and Microsoft earnings spark a big move on Wall Street

Don’t forget to follow us for more exclusive content on MSN.

If you liked this story, you’ll LOVE our FREE emails. Join today and be the first to get stories like this one.

This slideshow was made with AI assistance and human editing.