Scale AI's biggest customers Google plans to split after meta deals

SAN FRANCISCO (Reuters) – Scale AI's biggest customer, Alphabet (Googl.O) Google plans to cut its relationship with Scale after news broke that rival Meta (Meta.O) is taking a 49% stake in AI data label startup, five sources used to the issue said.

Google was planning to pay ChatGpt's competitors about $200 million in AI this year for human sign training data that is critical to developing technology, including sophisticated AI models including Power Gemini.

The search giant has already held conversations with several scale AI rivals this week as it looks to shift much of its workload this week, sources added.

Scale's significant business loss is because Meta has developed a major stake in the company, valued at $29 billion. The size was worth $14 billion before the transaction.

Scale AI intends to continue running the business while CEO Alexandr Wang moves to Meta with several employees.

Because core businesses are focused around a small number of customers, if you lose a major customer like Google, you can suffer a lot.

In a statement, an AI spokesperson said businesses that work with major companies and governments are still strong as they work to protect customer data. The company declined to comment on Google's details.

Scale AI was laked up in 2024 with revenue of $870 million, and Google spent about $150 million on AI services last year, sources said.

Other major tech companies, which are customers of scale, including Microsoft (MSFT.O), are also falling back. Elon Musk's Xai is also asking for a departure, one source said. Openai spends much less money than Google, but sources familiar with the issue decided to pull back from its size a few months ago. Openai's CFO said on Friday that as one of many data vendors, the company will continue to work with Scale AI.

Companies competing with META in developing cutting-edge AI models are concerned that doing business at scale could expose research priorities and roadmap to rivals, five sources said. By contracting with Scale AI, customers often share their own data and prototype products in which scale workers offer data signing services. With Meta taking a 49% stake, AI companies are concerned that one of their major rivals will gain knowledge about business strategy and technical blueprints.

Google, Microsoft, and Openai declined to comment. Xai did not respond to requests for comment.

Rivals look at the opening

The large amount of AI revenue comes from billing generative AI model manufacturers to provide access to a network of expert knowledgeable human trainers, from historians to scientists to doctoral students. Humans annotate complex datasets that use AI models “after training,” and as AI models become smarter, the demand for sophisticated human-provided examples could skyrocket, with one annotation likely to cost as much as $100.

Scale also labels data for businesses such as self-driving car companies and the US government, and companies such as the US government, according to sources. But its biggest money makers are partnering with generative AI model makers, sources say.

Google has already been trying to diversify its data service providers for over a year, three sources say. However, this week's Meta move has led Google to try to turn off Scale AI on all major contracts, sources added. Two sources said that the process could occur quickly due to the structure of the data label contract.

This will provide an opening where Scale AI rivals will jump in.

“Metascale trading marks a turning point,” said Jonathan Siddharth, CEO of AI competitor Turing. “Major AI labs recognize that neutrality is no longer an option. That's essential.”

Another competitor, Labelbox, will “produce probably hundreds of millions of new revenue” from customers who will flee the scale by the end of the year, CEO Manu Sharma told Reuters.

A competitor focused on building a PhD and professional network, Handshake saw a surge in workloads from top AI labs competing with the meta.

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

Many AI Labs now want to adopt internal data labels. This states that the data remains secure. In addition to competing directly with scale AI, Mercor CEO Brendan Foody also builds technology that allows candidates to be recruited and screened in automated ways, allowing AI Labs to quickly scale data labeling.

Founded in 2016, Scale AI provides a vast amount of labeled or curated training data that is important for developing sophisticated tools such as Openai's ChatGPT.

Meta trading is a boon for Scale AI investors, including Accel and Index Ventures and its current and former employees.

As part of the contract, Scale AI CEO Wang will take the top position to lead Meta's AI efforts.

Meta is battling the perception that the first set of the large language models of the Llama 4 released in April may have been behind in the AI ​​race after failing to meet performance expectations.