There is a major shakeup afoot in the world of artificial intelligence: Alphabet is said to be in the process of selling off one of its businesses that provides data-labeling services to the AI industry, and specifically companies that develop machine learning models.
This week’s strategic realignment comes hot on the heels of rival firm Meta Platforms Inc.’s news that it has bought a significant 49% chunk of Scale AI, news that has set off shock pulses across the competitive landscape for sophisticated AI model creation.
Google is said to have budgeted about $200 million for the year on Scale AI’s vital human-labeled training data for models like its Gemini AI — and now, they are actively considering other data service providers.
The biggest concern here, according to those who know about the discussions, is that proprietary research, roadmaps and technical blueprints could be exposed to a direct rival.
When companies hire Scale AI to label their data, they tend to give it access to sensitive, early-stage product data, provoking fears that the significant ownership that Meta could end up with might give the company an unfair view into rivals’ AI work.
The Meta-Scale AI deal — which saw the value of Scale AI soaring from $14 billion to $29 billion — also involves the owner of Scale AI, Alexandr Wang, coming aboard to assist Meta’s AI efforts.
Though Scale AI has promised to protect customer data and keep the business independent, with operations in a range of different industries, like self-driving cars and government contracts, its key source of revenue is in working with the companies that make generative AI models.
That Google did this, and the other reports that Uber and Microsoft and Elon Musk’s xAI are doing it too (with OpenAI cutting back their toes months ago), represent a nifty little watershed. Rivals to Scale AI, including Turing and Labelbox, are also already bracing themselves for an influx in new business as AI labs make data neutrality and security a top priority.
This shifting landscape may also nudge more AI companies toward developing in-house data-labeling capabilities in order to protect their most prized intellectual property. It’s data mining — and it’s great data, and lots of data.” The arms race in AI, which has previously been won and lost based on compute power, is increasingly pivoting around the security and control of high-quality human-annotated data.