Scale AI and Google Are Looking To Part Ways

This year, Google had intended to pay Scale AI roughly $200 million for the human-labeled training data that is essential to technological advancement.

Some of the startup’s clients could be hesitant to do business with Meta because of its significant investment in Scale AI.

After learning that rival Meta is acquiring a 49% share in the AI data-labeling business, Alphabet’s Google, Scale AI’s biggest client, intends to sever its relationship with the company, five people familiar with the situation told Reuters.

According to Reuters, Google had intended to give Scale $200 million this year for the human-labeled training data that is essential for creating technology, including the advanced AI models that underpin Gemini, its rival ChatGPT, but it is now discussing severing relations with its rivals. According to reports, OpenAI took a similar choice months ago, but its CFO stated that the business will continue to cooperate with Scale as one of many vendors. Microsoft is also apparently considering withdrawing.

As it looks to delegate a large portion of that task, the search engine giant has already spoken with a number of Scale AI’s competitors this week, according to sources.

The U.S. government and self-driving vehicle manufacturers are among Scale’s clients, but according to Reuters, its largest customers are generative AI firms looking to hire experts who can annotate data for model training.

Regarding the report, Google chose not to comment. Although he declined to comment on the company’s partnership with Google, a Scale representative told TechCrunch that Scale’s business is still doing well and that it will continue to function as a separate entity that protects the data of its clients.

According to earlier reports, Scale CEO Alexandr Wang joined Meta to spearhead the company’s attempts to develop “superintelligence” after Meta invested $14.3 billion in the business, acquiring a 49% interest.

Significant business loss for Scale coincides with Meta’s acquisition of a sizable share in the company, which is valued at $29 billion. Prior to the transaction, Scale was valued at $14 billion.

Scale AI plans to continue operating while Alexandr Wang, its CEO, and a few staff members transfer to Meta. It might suffer a significant deal if it loses important clients like Google because its core business is centred around a small number of clients.

A Scale AI representative stated in a statement that the company’s business, which includes working with governments and other corporations, is still robust because it is dedicated to preserving consumer data. The business refused to discuss information with Google.

Google reportedly paid about $150 million on Scale AI’s services last year, and the company generated $870 million in sales in 2024, according to sources.

Microsoft and other significant tech firms that Scale’s serves are also retreating. According to one of the sources, Elon Musk’s xAI is also trying to sell. Despite spending significantly less than Google, OpenAI chose to withdraw from Scale a few months ago, according to people familiar with the situation. Scale AI is one of OpenAI’s numerous data vendors, and the firm will continue to collaborate with them, the CFO announced on Friday.

Five sources stated that companies who compete with Meta in the development of advanced AI models are worried that working with Scale may reveal their research priorities and roadmap to a competitor. Customers who hire Scale AI frequently provide prototype products and confidential data, for which Scale’s employees provide data-labeling services. AI businesses are worried that one of their main competitors may learn about their technological blueprints and business strategy now that Meta has a 49% interest.

Microsoft, OpenAI, and Google all declined to comment. A request for comment from xAI was not answered.

Competitors See Opportunities and the majority of Scale AI’s income is generated by charging generative AI model developers for access to a network of highly qualified human trainers, including scientists, historians, and some with doctorates. The demand for sophisticated human-provided examples has increased as AI models have become more intelligent, and the cost of an annotation can reach $100. Humans annotate complicated datasets that are used to “post-train” AI models.

According to the sources, Scale also performs data-labeling for businesses that are probably here to stay, such as the US government and self-driving car manufacturers. According to the sources, however, collaborating with generative AI model developers is its major revenue generator.

Three of the people claimed that Google had been working for more than a year to diversify its data service providers. According to the sources, Google is attempting to remove Scale AI from all of its important contracts as a result of Meta’s actions this week. That procedure might occur rapidly due to the structure of data-labeling contracts, according to two people.

This will give competitors of Scale AI a chance to enter the market.

Turing, a rival of Scale AI, CEO Jonathan Siddharth stated, “The Meta-Scale deal marks a turning point.” “Leading AI labs are realizing neutrality is no longer optional, it’s essential.”

Manu Sharma, the CEO of Labelbox, a rival company, told Reuters that the company will “probably generate hundreds of millions of new revenue” by the end of the year from customers who are leaving Scale.

Top AI laboratories that compete with Meta increased their workloads at Handshake, a rival that focuses on creating a network of PhDs and experts.

“After the news, our demand tripled overnight,” Handshake CEO Garrett Lord said.

Brendan Foody, CEO of Mercor, a startup that competes directly with Scale AI and develops technology around being able to recruit and vet candidates in an automated way, said that many AI labs now want to hire in-house data-labelers because it allows their data to remain secure. This allows AI labs to quickly scale up their data labeling operations.

In order to create complex products like OpenAI’s ChatGPT, Scale AI, which was founded in 2016, offers enormous volumes of labeled data or curated training data.

Scale AI’s investors, including Accel and Index Ventures, as well as its present and past staff, will benefit greatly from the Meta purchase. 

Wang, the CEO of Scale AI, will have a key role in overseeing Meta’s AI initiatives as part of the agreement.

After its first batch of Llama 4 big language models, which were published in April, failed to meet performance standards, Meta is battling the idea that it may have lagged behind in the AI race.