The recent acquisition of 49 per cent of the shares of Scale AI, the leader in the field of artificial intelligence data at $15 billion by Meta, triggered the “earthquake” of the AI data supply chain. According to Reuters, Scale AI was suddenly “discarded” by its largest client, while many head AI companies, including Microsoft, XAI and others, were revisiting or terminating their cooperation with Scale AI.
According to reports, Google had planned to pay approximately $200 million to Scale AI in 2024 to obtain high-quality manual labelling data, which were the key “breeds” for training its Gemini megamodel. This is not the case.Following Meta ‘ s announcement of Scale AI, Google has moved fast and has begun to approach other data providers this week, planning to transfer most of its core business from Scale AI.
Meta ‘ s $15 billion investment raised the Scale AI valuation from approximately $14 billion to $29 billion, with an infinite surface. Ironically, however, the transaction has also directly shaken the confidence of its largest client and served as a catalyst for Google to accelerate “de-scaleization”.
According to sources, Google has sought to diversify its data suppliers for more than a year, while Meta’s entry became the last straw to crush camels, prompting Google to resolve to move all critical operations. Thanks to the “change of control” clause commonly found in the contract, Google’s evacuation may have been completed quickly.
Google’s departure was undoubtedly a heavy blow to Scale AI, with a total revenue of $870 million in 2024, with a Google company alone contributing $150 million, the largest single source of income.
The most horrific is the chain reaction with Google’s “evacuation”. According to sources, Microsoft is “re-assessing” its cooperation with Scale AI, and Mask has a clear plan to terminate it. OpenAI, for its part, significantly reduced the use of Scale AI several months ago, even though its CFO recently indicated that Scale was one of its “multiple data suppliers”.
At the heart of the collective “runaway” of Google, Microsoft and others lies the great risk of data security and disclosure of commercial secrets. Similarly, the leading company in the AI industry, which has handed over sensitive information, including prototype tools and private data sets, to Scale AI, which is held by nearly half of Meta (49 per cent), for labelling purposes, is tantamount to exposing its core strategic road map and technical details to Meta, the main competitor. One of the industry’s executives said, “It’s like keeping the competition’s eyes on your research blueprint.”
In the face of dramatic changes, Scale AI plans to continue to operate, but CEO Alexandre Wang and some core staff will join Meta. The spokesperson for Scale AI stressed that the company was “relative in its relations with large enterprises and government agencies” and “always committed to protecting client information”, but did not comment on Google’s decision.
Scale AI relies heavily on high-end data labelling services for generating AI. Its core competitiveness lies in building a platform to connect AI with experts with a deep professional background who will fine-tune the data and complete the later fine-tuning of the AI model. Companies also provide data indications for auto-driving car manufacturers and government agencies in the United States, and these customers are expected to have relatively stable partnerships due to their different fields or regulatory requirements. But it is undeniable that the generating AI giant is the engine of growth.
“Meta’s share in Scale AI’s deal is a watershed moment”, Jonathan Siddharth, Chief Executive Officer of Turing, commented, “the top AI laboratory is now well aware that the neutrality of data suppliers is no longer a plus, but a necessity for survival.” Google’s action was the flash of this cognitive shift.
Because of the need for absolute data control, more and more AI laboratories tend to form professional data labelling teams internally. The CEO Brendan Foody of Mercor, the talent recruitment technology starter, observed a significant trend: “Many AI laboratories now have a strong desire to employ in-house data labelling experts, which is a fundamental way of ensuring the security of their core data assets.” This marks an unprecedented level of industry attention to data sovereignty and security.
