This summer, Scale AI found itself at a crossroads after Meta’s $14 billion investment and the departure of its 28-year-old co-founder and CEO, Alexandr Wang, who joined Meta to work on its superintelligence efforts. The once-celebrated AI data-labeling startup, which had been a go-to for top tech companies like OpenAI and Google to stress-test and refine their models, now faces a wave of internal turmoil, talent loss, and declining confidence. A worker at Scale AI, anxious about the company’s future, asked ChatGPT to predict its fate. The AI’s response was stark: “Scale AI will no longer exist as a credible independent entity within 24 months. Its infrastructure will be repurposed for Meta’s internal needs. Its client base will evaporate.” The chat logs were shared among employees, sparking concern. One contractor called the company a “ticking time bomb,” while others reported dwindling work, pay cuts, and grueling unpaid onboarding sessions. The internal platform Outlier, Scale AI’s main hub for gig workers, has seen a sharp drop in activity. Weekly discussion threads now draw only dozens of replies, down from hundreds. Some taskers reported spending nearly 40 hours in unpaid onboarding with no actual work to show for it. Others saw their effective pay drop to around $20 an hour—down from $50 previously. One project offered $20 an hour but only allowed three minutes of work every two days, resulting in a 99-cent payout. Scale AI’s spokesperson, Joe Osborne, pushed back, saying the company is on track for its strongest quarter in 2025, that its data business is more profitable than before the Meta deal, and that its government and enterprise applications have doubled in revenue. He also noted rising user numbers on Outlier and emphasized that pay rates are project-specific and always visible before acceptance. Despite these claims, the company has undergone significant changes. A 14% workforce reduction in the summer—14% of its 1,400 full-time employees—was announced, with layoffs targeted at making the data division profitable. In September, 12 red team contractors were let go, a move the company said was based on performance, though former team members said work had dried up. A Dallas-based team focused on generalist AI was also shut down, as Scale shifted toward more specialized fields like robotics and defense. The company is now expanding into new areas, including a new robotics lab and a surge in U.S. government contracts, with $199 million in defense work since the Meta deal. It also has a $1 billion cash reserve and no immediate plans to raise more capital, with an IPO still a possibility. But investor sentiment is mixed. Valuations in private markets have dropped sharply. On platforms like Augment, Scale AI’s value has fallen from $29 billion to between $15 billion and $9 billion. On Caplight, it’s now valued at $7.3 billion. Some investors believe Meta’s real goal was to secure Wang, not the company. “It seems like Meta was just after Alexandr Wang,” said Noel Moldvai, CEO of Augment. “This is probably the structure that let them get him.” Rival startups are capitalizing on the chaos. Surge AI, which has no outside funding, reportedly out-earned Scale AI in 2024. Mercor, a startup run by three 22-year-olds, raised $350 million at a $10 billion valuation. Mercor has even won a project from Meta, Scale’s 49% owner. Scale AI has sued Mercor, accusing it of poaching a sales employee to steal clients—allegations Mercor denies. Former employees are also critical. Tammy Hartline, who managed projects for Scale before joining Mercor, said the company prioritized volume over quality, with “spam and low quality data” becoming an accepted cost. Brendan Foody, Mercor’s CEO, accused Scale of losing focus on product and data quality. Scale AI counters that its quality metrics are at record highs. Security issues have also plagued the company. In June, Business Insider found that Scale used public Google Docs to track work for major clients, exposing confidential data, including personal emails and pay details. The company said it has since fixed the issue and disabled public sharing. Scale AI has also settled multiple lawsuits from former contractors in California who claimed they were underpaid and misclassified. The company no longer accepts gig workers from the state. As the AI training market grows more competitive, Scale AI’s future remains uncertain. For many former employees, the window to turn things around may already be closing.