OpenAI has made six acquisitions in 2026 alone (nearly matching its total for all of 2025) with its most recent purchase, Astral, a maker of open-source developer tools, following its earlier acquisition of Promptfoo, an open-source AI application testing tool. Anthropic has moved more slowly by comparison, with one known 2026 purchase (Vercept, a two-year-old software development startup) after two in 2025 (Humanloop, an LLM evaluation platform, and Bun, a JavaScript runtime). The contrast between these two labs' acquisition pace is itself a useful signal about how differently they're building out their platforms.
Why AI coding tools became the hottest acquisition category
AI coding tools have specifically become the most active area for both acquisitions and the founder-poaching deals covered below. Anysphere, the company behind Cursor, has passed $1 billion in annual recurring revenue, a scale that's reshaping what labs are willing to pay to either compete with or absorb adjacent coding-tool capability. OpenAI's Astral and Promptfoo purchases both fit this pattern directly: developer-tooling companies being folded into a much larger platform rather than left to compete independently.
The stranger, larger trend: acquisitions that aren't acquisitions
The more structurally significant pattern isn't the acquisitions themselves. It's a different kind of deal entirely. Between March 2024 and January 2026, Google, Microsoft, Amazon, and Meta spent more than $20 billion hiring away the founding teams of AI startups without technically acquiring a single company. The mechanism: a large company buys the assets (typically an IP license) of an early-stage startup and brings on one to three members of the founding team, without a formal change-of-control acquisition of the company itself.
Why this structure exists
This isn't a technicality. It's a deliberate structure that avoids the regulatory scrutiny a full acquisition would trigger, particularly given how much antitrust attention Big Tech's AI dealmaking has already drawn. It also lets the acquiring company get the thing it actually wants (the founding team's expertise and, often, an IP license) without taking on the acquired company's full cap table, existing customer contracts, or other liabilities a traditional acquisition would carry. For the startup's other investors and employees left behind, it's a genuinely different (and often worse) outcome than a full acquisition, since they don't share in the same payout the founders effectively receive through the new employment arrangement.
What this means for diligence, and for everyone else at the startup
This structure changes what "AI acquisition" diligence actually needs to examine, connecting to the broader capital-allocation questions we cover in OpenAI's own funding and structure evolution: the real asset changing hands in many of these deals is a small team's expertise and an IP license, not a company with an ongoing, transferable customer base. That's a genuinely different risk and value profile than a traditional acquisition, and it's part of why these deals have moved faster and more quietly than a full M&A process would.
Why the pace itself is the story
OpenAI closing in on its full 2025 acquisition count within a matter of months of 2026, alongside the ongoing $20 billion-plus acquihire wave, both point to the same underlying pressure: frontier labs are consolidating coding-tool and infrastructure capability faster than at any point in the AI industry's history so far. Whether that consolidation continues at this pace, or whether regulatory attention to the acquihire structure specifically catches up to it, is one of the more consequential open questions in AI business coverage right now.