Google's $1.5B AI Deal Is Not Legally A Merger

A $1.5 Billion Deal That Buys No Company

Google is in talks to pay close to $1.5 billion for a non-exclusive license to Mechanize's AI coding technology and to hire the researchers who built it, according to SiliconANGLE. Mechanize builds simulated environments and grading systems that let AI coding agents practice software engineering tasks and get scored on the result, the kind of reinforcement-learning infrastructure that Google needs to close a gap: its own coding agents are widely seen as behind Anthropic's Claude Code and OpenAI's Codex in enterprise adoption. The company Mechanize itself is not being bought. Its investors, reportedly including former GitHub chief executive Nat Friedman, Stripe chief executive Patrick Collison and podcaster Dwarkesh Patel, keep their equity in a business that keeps operating on its own.

Mechanize is barely a year old. It raised $9.1 million earlier in 2026 at a $500 million valuation, according to SiliconANGLE, meaning the licensing-and-hiring terms now under discussion value the underlying technology and team at roughly three times that figure within months.

The Third Time Is a Pattern

Google has now run this same play three times in under two years, and the repetition is the story as much as any single deal. In each case, Google licensed a startup's technology on a non-exclusive basis and hired its key people, while the original company itself survived as an independent entity under different ownership or direction. TechFundingNews and SiliconANGLE both place the Mechanize talks in that same lineage.

The pattern holds regardless of what happens to any single founder afterward, as Shazeer's later move to OpenAI shows. What stays constant is the structure: Google gets the technology and the talent, and the seller keeps a company that can be sold again, wound down, or redirected without Google's say.

Why the Merger-Review Line Never Gets Crossed

A conventional acquisition of Mechanize would put the deal in front of antitrust reviewers in the US and, if the parties have sufficient EU turnover, the European Commission, both of which examine full changes of control for their effect on competition. A non-exclusive license paired with hiring the team is not a change of control in the legal sense reviewers use, and SiliconANGLE's reporting frames the structure explicitly as a way to sidestep that scrutiny, without describing any formal review actually underway. Google has not disclosed the deal as a merger filing in either jurisdiction, and neither report identifies one.

Regulators on both sides of the Atlantic have flagged reverse acquihire structures like this before, in deals involving Microsoft and Inflection and Amazon and Adept, without blocking any of them. The scrutiny has so far produced attention, not enforcement action that stopped a deal of this shape from closing.

What This Means If You Bet on the Losing Bidder

A business that adopts an AI coding tool from a startup Google might later license this way takes on a risk a standard acquisition would not create. When a company is bought outright, its product has one clear owner going forward, for better or worse. When a company is licensed and its team hired away, the original business keeps existing but loses the people who built what a customer is relying on, while remaining free to sell the same technology to someone else, change direction, or wind down.

Procurement teams evaluating AI coding vendors now have a concrete reason to ask not just who owns a tool today, but who actually built it and whether that team is still there. A licensing deal like the one under discussion with Mechanize can hollow out a vendor's engineering bench while leaving its logo, contracts and support obligations unchanged on paper.

Read next: Discovery Loop's Price Tag Jumped 5x With No Product  |  Five Billion In Defense Money Is Now Chasing GPUs