Stepmark Partners

How Big Tech Is Rewriting M&A: The License and Acqui-hire Era

—

by

Da
in

Authors: Justin Wang (Analyst), Evelyn Xie (Analyst)

Table of Contents

Introduction

A new chapter of Silicon Valley dealmaking has emerged: the license and acquihire (L&A). Rather than pursuing traditional M&A, Big Tech now licenses models, data, and infrastructure while hiring only the core team members (e.g., researchers, technical staff, etc.) and founders behind the breakthroughs. These transactions deliver speed and certainty, sidestepping regulatory barriers that slow traditional M&A. Yet, while acquirers secure elite engineering talent and intellectual property, they leave behind the wider workforce and investors, who often see little upside. This shift is redefining corporate consolidation around people rather than companies, concentrating AI expertise within a handful of incumbents.

Inflection → Microsoft (Mar 2024)

The transaction was structured as a $620 million non-exclusive licensing fee for Inflection’s models including Inflection 2.5, later deployed on Microsoft Azure, plus a $30 million payment to waive legal claims tied to mass hiring. In total, the $650 million all-cash structure contained no stock or earn out provisions. Reid Hoffman, a co-founder of Inflection, is both a Microsoft Board member and Partner at Greylock, an investor in the startup.

This calculated approach allowed Microsoft to absorb critical AI expertise and technology while technically preserving Inflection’s independence as it pivoted toward an enterprise AI studio model. The transaction highlighted Microsoft’s regulatory caution: integrating top researchers and IP without crossing acquisition thresholds or triggering merger review.

Adept AI & Amazon (Jun 2024)

For $330 million (plus $100 million retention bonus) Amazon secured a non-exclusive license to Adept’s agent technology, multimodal foundational models, and select datasets, with the transaction structured entirely in cash rather than equity. The deal circumvented traditional acquisition frameworks. Amazon did not purchase Adept’s equity or assume ownership of the entity, instead, combining technology access with mass talent recruitment.

The arrangement drew immediate scrutiny from regulators questioning whether it constituted a de facto acquisition designed to bypass antitrust notification rules. For Amazon, the deal was strategic: onboarding Adept’s entire founding team to accelerate AGI ambitions while ensuring rapid access to agentic model infrastructure. For Adept’s investors and remaining staff, however, it represented a limited financial outcome and organizational downsizing.

Covariant & Amazon (Aug 2024)

Through the licensing component, Amazon gained rights to Covariant’s “Brain” foundation model for robotic perception and motion planning, integrating it into its fulfillment center robotics systems while permitting Covariant to continue licensing the same technology externally. The arrangement included selective hiring of Covariant’s three co-founders and roughly 25% of its team.

This hybrid deal strengthened Amazon’s robotics automation capabilities while preserving Covariant’s limited independence as a boutique AI robotics studio. It followed Amazon’s growing use of L&A frameworks, rapidly absorbing R&D leadership while leaving behind sales and operations functions that fall outside Big Tech’s automation focus.

Character.AI & Google (Aug 2024)

By pursuing a reverse acquihire rather than a traditional merger, Google avoided antitrust filing thresholds under Hart Scott Rodino rules while effectively absorbing Character.AI’s intellectual property and core team. Founders Noam Shazeer and Daniel De Freitas rejoined Google under DeepMind alongside roughly 30 researchers, with the $2.7 billion transaction structured around a non-exclusive technology license.

The move reinforced Google’s focus on expanding Gemini’s multimodal conversational layer and rebuilding expertise in socialized AI systems, a domain Character.AI had pioneered through customizable “AI personas.” Though Character.AI remains nominally independent, the exit of its founders and loss of proprietary access effectively transformed it into a licensing shell, emblematic of a broader 2025 trend consolidating model assets while skirting merger scrutiny.

Scale AI & Meta (Jun 2025)

Negotiations between Mark Zuckerberg and Alexandr Wang evolved from an initial $5 billion non-voting investment to a final $14.3 billion deal, reflecting Meta’s urgency to resolve delays in its Llama 4 “Behemoth” model. The transaction gave Meta a 49% non-voting stake while extracting Wang and select employees to lead Meta’s new superintelligence group.

This structure mirrored the Microsoft Inflection model but with a twist: Meta transferred its voting rights back to Wang to maintain Scale’s operational neutrality. Yet the arrangement triggered industry turmoil, with key customers including Google terminating contracts. The deal raised broader concerns about conflicts of interest when founders lead both independent startups and Big Tech AI divisions.

Windsurf & Google + Cognition AI (Jul 2025)

Google’s parent company Alphabet made a decisive move in the AI talent wars by stepping into the stalled OpenAI negotiations to acquire Windsurf, a fast-growing AI-native coding platform. OpenAI had pursued a $3 billion acquisition, but concerns from Microsoft, its largest investor, derailed the talks. Within days, Alphabet offered roughly $2.4 billion for a non-exclusive technology license and hired Windsurf’s founder Varun Mohan, co-founder Douglas Chen, and its core R&D team. The hires joined DeepMind’s Gemini division to advance agentic AI coding systems for autonomous software development.

Shortly later, Cognition AI acquired Windsurf’s remaining assets, brand, and workforce, creating what analysts described as a dual or “distributed M&A” structure. Google secured the key leadership and IP access, while Cognition inherited product and operational teams, ensuring continuity for customers and investors. This two-part deal allowed Alphabet to sidestep antitrust scrutiny while reinforcing its edge in agentic AI development. It also marked a turning point in competitive AI dealmaking, proving that licensing and acquihires could rival traditional acquisitions in speed and strategic impact.

Enfabrica & Nvidia (Sep 2025)

The deal included no disclosed earn out provisions and represented a non-traditional acquisition, as Nvidia licensed Enfabrica’s Accelerated Compute Fabric (ACF) SuperNIC technology rather than acquiring the company’s assets outright. Nvidia had previously invested in Enfabrica’s $125 million Series B round in 2023, giving it early insight into the startup’s networking breakthroughs.

The rationale centered on Enfabrica’s ability to scale GPU interconnectivity beyond the 100,000 chip ceiling that constrained existing architectures, a critical enabler as Nvidia transitions from selling processors to integrated rack scale systems. The transaction reinforced Nvidia’s strategy of using licensing plus hiring structures to expand core engineering capacity while avoiding formal antitrust filings.

Takeaways

License and acquihire transactions have become the defining mechanism of AI consolidation. They allow acquirers to onboard top researchers, absorb intellectual property, and accelerate product timelines, all without triggering merger reviews. Founders who depart through acqui-hire deals while leaving their teams behind break a fundamental promise: that captain and crew would see the journey through together. While efficient for incumbents, these deals disadvantage investors and support staff (e.g., accountants, marketers, etc.) through limited financial upside and double taxation. By emphasizing selective absorption over organizational integration, Big Tech has created a system optimized for speed but short on inclusivity, where innovation accrues to incumbents, while the startup ecosystem faces diminishing room for growth.