The Equation Series
Menu

Part II: The Markets

Chapter 9

Tools and Data

On 10 June 2025 reports described Meta investing about $14–15 billion for 49% of Scale AI, the San Francisco company that supplied frontier labs with the human-written examples, preference rankings and evaluations they used to post-train their models. On 12–13 June Scale confirmed a "significant investment" valuing it at more than $29 billion. Its founder and chief executive, Alexandr Wang, would leave to join Meta while keeping his seat on Scale's board, and Jason Droege, its chief strategy officer, would become interim chief executive (company). The stake was non-voting and below half, so no merger filing was required (reported). Most of the roughly $14.3 billion went to existing shareholders and vested employees as a distribution rather than staying in the company (reported).

Scale had earned about $870 million of revenue in 2024 and was targeting about $2 billion in 2025 (reported). Its customers included Google, OpenAI, xAI, Microsoft and Meta itself. Within weeks Google, which had planned to spend about $200 million with Scale in 2025 (reported), moved to wind down its work; OpenAI said it had already been reducing its use; xAI also pulled back (reported). Rivals such as Surge AI, Mercor, Turing and Handshake picked up the business (reported). On 16 July Scale cut about 14% of its staff, roughly 200 employees, and about 500 contractors, with Droege citing over-expansion of its generative-AI data business (reported). The 2025 target was not met (reported, verify), and the company later projected more than $1 billion of revenue for 2026 (company claim, reported).

The deal made Scale's shareholders rich and weakened Scale. It is the starkest example in this book of a rule that governs the whole tools-and-data layer: a supplier to model builders is worth what its neutrality and its position in the customer's workflow are worth, and both can disappear in a week. This chapter looks at the businesses that sell picks and shovels to the people building and using models, and at why some of them compound while others are absorbed, bundled or abandoned.

The preview ends here. The full chapter is in the book.

In this chapter

  1. 9.1The layer between the model and the work
  2. 9.2Scale AI: a services business priced like a platform
  3. 9.3A worked example: concentration and operating leverage
  4. 9.4Databricks: owning the place where the data lives
  5. 9.5Picks and shovels that were bundled or bought
  6. 9.6How an expert reads a tools-and-data business