The great agentic acqui-hire: what six months of AI deals reveal

The agentic-AI story of the last year gets told in models and demos. The more revealing story is in the deal flow. Over the past six months, a pattern that took shape across 2025 has hardened: the largest technology companies are not out-building independent agent startups so much as absorbing them — their people, their code, and sometimes the company itself. What follows is drawn entirely from reported, on-the-record deals; sources are linked throughout.

The structure of the deals is the story

The striking thing isn't the dollar amounts — it's the shape. Increasingly these are acqui-hires and reverse acqui-hires: the acquirer licenses the startup's technology and hires its founders and key researchers rather than buying the company outright.

The archetype was Google's roughly $2.4 billion deal for Windsurf in July 2025 — Google took a nonexclusive license to the coding-agent startup's technology and hired its CEO Varun Mohan and co-founder Douglas Chen, after OpenAI's ~$3 billion agreement to buy Windsurf collapsed. Cognition, the maker of the Devin coding agent, then acquired what remained of the company.

Meta's move on Scale AI followed the same logic at ten times the size: **$14.3 billion for roughly a 49% non-voting stake** in June 2025, with founder Alexandr Wang leaving to run Meta's new Superintelligence Lab.

This structure is not an accident. Buying the talent and a license — rather than the company — acquires capability while sidestepping the antitrust scrutiny a full merger would attract. It also leaves something behind: reporters have taken to calling the hollowed-out remainders "zombie startups." The acqui-hire has, by several accounts, become the defining deal structure of this AI cycle.

Incumbents are assembling stacks, not placing bets

The enterprise platforms buy differently — outright, and in bulk, to assemble an agentic stack. ServiceNow has been the clearest: it paid **$2.85 billion for Moveworks** in March 2025 to fold a front-end AI assistant into its agentic-orchestration platform — then kept going, acquiring Data.World two months later and, in December, agreeing to buy Veza to govern AI-agent permissions.

OpenAI, meanwhile, has been acquiring at a pace that reads like a roll-up. It bought product-experimentation platform Statsig for **$1.1 billion** in September 2025 and installed its CEO to lead engineering for OpenAI's applications business — one of what TechCrunch counted as roughly nine acquisitions in a single year.

The last six months

Into 2026 the pace hasn't slowed; it has spread to smaller, more targeted deals. In January, OpenAI acqui-hired the team behind executive-coaching tool Convogo and bought a small health-records startup, Torch, for a reported ~$100 million; Google hired the team behind voice-AI startup Hume AI. In May, Asana bought no-code agent-builder StackAI for $75 million to position itself as an "AI-native" work platform.

The through-line: the buyers are the same handful of names — OpenAI, Google, Meta, ServiceNow, Asana — appearing again and again. That repetition is the tell. These aren't opportunistic grabs; they're roadmaps being filled in.

What it means for everyone else

For founders and investors, the last six months clarified the exit math. A strong agent team is now more likely to be hired than to IPO, and the premium is on people and distribution rather than standalone products. For the market, capability is consolidating fast into a few platforms that own both the models and the channels.

Which is exactly why the interesting independent position is not another agent a hyperscaler can absorb in a weekend, but the connective layer between them — the identity, trust, and record-keeping that let agents from different companies transact safely. Models and teams get acqui-hired. Neutral infrastructure that both sides have to integrate against is much harder to buy, and much more valuable to build. The deal flow of the last six months is, in that sense, a map of where value will not stay independent — and, by contrast, where it will.

Every deal referenced above is linked to a contemporaneous report from CNBC, TechCrunch, or Forbes. Figures and dates reflect those reports; deal terms for privately held companies are as disclosed.