Google is severing ties with Scale AI following Meta’s shock announcement of acquiring a 49% stake in the AI data-labeling startup, Reuters reports, citing sources.
The deal, which values Scale at $29 billion, more than double its previous $14 billion valuation — has sent shockwaves through Silicon Valley, triggering an exodus by major AI companies wary of exposing proprietary research and sensitive data to a direct competitor.
Sources familiar with the development say Google had planned to pay Scale nearly $200 million in 2025 for human-labeled training data crucial to the development of its Gemini models. But that arrangement has now been scrapped. The company, reportedly Scale’s largest client, is moving swiftly to redirect contracts to rival vendors. Already, Google has begun talking to multiple Scale competitors this week in a bid to offload most of its AI annotation needs.
This rapid shift stems from rising concerns about Meta — parent of Facebook, Instagram, and WhatsApp — gaining access to privileged data from companies it competes with directly in the artificial intelligence arms race. Meta now has a powerful foothold in the AI infrastructure backbone, by owning nearly half of Scale AI and absorbing its CEO Alexandr Wang into its AI division.
The Deal Was A Silent Coup
The Meta-Scale partnership came together quietly but swiftly. Multiple sources suggest the deal was orchestrated over the past several months as Meta aggressively sought to bolster its AI development following underwhelming reviews of its Llama 4 model released in April. While the model showed promise, it failed to match the performance of OpenAI’s GPT-4 or Google’s Gemini in key benchmarks, prompting fears Meta was falling behind.
To accelerate its progress, Meta turned to Scale AI, which built a reputation since its founding in 2016 as the premium supplier of high-quality, human-labeled datasets — a crucial resource for training advanced AI systems. Scale’s services are not cheap: some annotations by PhD-level experts can cost upwards of $100 each. But its clients, which include Google, Microsoft, OpenAI, xAI, and the U.S. government, were willing to pay for precision.
With Wang set to lead Meta’s AI efforts and several Scale employees also transitioning to the company, the deal raises alarm for rival firms. Many of them rely on Scale for labeling not just raw data, but also prototype model outputs, internal prompts, and context-rich examples that are core to their development strategies. Now, those same companies fear their crown jewels could end up within Meta’s line of sight.
Industry Backlash Gathers Pace
The backlash has been swift with Google, Microsoft, Elon Musk’s xAI, and even OpenAI — which had already begun reducing its reliance on Scale months ago — all walking away. Google in particular is moving fast to dismantle all key contracts with Scale. Although the exact timeline varies by agreement, the sources say the shift could be completed quickly due to the flexible structure of many data-labeling deals.
Labelbox, Turing, Handshake, and Mercor — smaller competitors once overshadowed by Scale’s dominance — are now witnessing a surge in demand. Labelbox’s CEO expects to generate hundreds of millions in new revenue from defecting clients. Handshake said its workload tripled within 24 hours of the Meta-Scale announcement. Turing’s CEO summed up the mood across the industry: “Neutrality is no longer optional, it’s essential.”
OpenAI, despite spending far less than Google on Scale services, has reiterated it will continue working with Scale but emphasized that the startup is only one of many vendors. Elon Musk’s xAI, meanwhile, is said to be preparing to exit completely. Microsoft has not commented publicly but is also believed to be shifting its data-labeling contracts.
A Strategic Gamble for Meta — and a Risk for Scale
The deal is undoubtedly a win for Meta. Alexandr Wang’s appointment is expected to inject new technical vigor into Meta’s AI roadmap. But for Scale AI, the Meta alliance could come at a high cost. Much of the company’s revenue — $870 million in 2024 — comes from providing services to companies that now view it as compromised. Unlike its government and automotive contracts, which may be insulated from competitive threats, the lucrative generative AI sector that powered its growth now stands on shaky ground.
The company’s statement following the deal tried to project stability, insisting that its business “remains strong” and that it is committed to customer data protection. But it did not comment directly on the specifics of Google’s departure or the ongoing client exodus.
Beyond immediate business impacts, the Meta-Scale deal is expected to reshape the AI industry’s supply chain. Companies have come to recognize that control over data infrastructure — including labeling, annotation, and fine-tuning processes — is just as critical as access to GPUs or large model architectures. This realization is pushing more labs to build in-house data-labeling teams and secure their own AI training pipelines, even at greater cost.
Meta’s strategic bet on acquiring a direct line into that ecosystem is high-risk, high-reward. While it stands to gain a wealth of internal capability through Scale and Wang, the fallout may permanently alienate Meta from industry collaborations at a time when AI research increasingly hinges on trust, interoperability, and data security.
In the end, what Meta gains in scale, it may lose in credibility, at least in the eyes of its fiercest AI rivals.
