Introduction to AI Transaction Complexity
In July 2025, Google made headlines by shelling out USD 2.4 billion for a rather unconventional acquisition. Instead of purchasing Windsurf—a startup formerly known as Codeium—Google focused on securing the talent behind the company. This included hiring the CEO and core research leaders while obtaining a non-exclusive license for Windsurf’s technology. Shortly after, Cognition AI acquired the remaining assets, encompassing the intellectual property, product line, brand, operations, and crucial teams such as engineering and go-to-market.
Understanding Capability Acquisition
This dual acquisition raises an important question: which capabilities did each buyer actually secure? Traditionally, transactions have been straightforward, involving shares, intellectual property, or contractual rights. However, AI transactions are more intricate, involving elements like source code, training data, customer feedback, third-party models, and the expertise of individuals who can integrate these components.
The Role of Regulators
Regulators have acknowledged this shift. In March 2024, Microsoft acquired the core team of Inflection AI and obtained a non-exclusive license for its intellectual property without acquiring the company itself. The UK Competition and Markets Authority (CMA) deemed this a “merger situation”, showing that transferring a team with essential know-how can itself constitute a merger.
Mapping Capabilities
For a successful AI transaction, it’s crucial to map out the capabilities being acquired and the conditions for their continuation. This mapping exercise is essential not just for technical reasons but also for assessing risks and valuation. A failure in this area could result in the buyer acquiring software without the necessary data to improve it, leading to a significant reduction in anticipated value.
Inputs, Outputs, and People
Understanding the sources of value in AI systems is crucial. Inputs often involve various types of data, each governed by different rights and restrictions. Contracts must specify the data’s origin, purpose, and usage rights, especially under laws like the Digital Personal Data Protection Act, 2023.
Outputs, such as evaluation datasets and customer-specific improvements, also need clear ownership and usage rights. A rights matrix can help delineate customer-specific information from generalized improvements.
Finally, a significant part of a company’s capabilities lies with its people. Indian law complicates this by voiding agreements that restrain trade, making employee retention crucial through incentives and knowledge-transfer obligations.
Conclusion
The future of AI contracting involves the application of conventional principles to dynamic scenarios. Lawyers must identify the capabilities being acquired, map all essential components, and determine their legal status before finalizing the contractual framework. While traditional elements like the company or software remain central, they no longer constitute the entire transaction.
About the Authors
Ashima Obhan is a Senior Partner and Arzu Chimni is an Associate Partner at Obhan Mason.
Disclaimer: The views expressed in this article are those of the authors and do not necessarily reflect those of Bar & Bench.
