M&A in AI: Redefining dealmaking

For boardrooms confronting AI as this decade’s defining disruptive force, a strategic imperative has emerged: if you cannot build it fast enough, acquire it. From Google’s USD32 billion acquisition of Wiz to Krutrim’s purchase of BharatSahAIyak in India, acquirers are paying for capability, and not just revenue. As this trend gathers pace, the traditional M&A playbook must also evolve with the market.

AI dealmaking reshapes India compliance

AI’s reliance on constant collection and processing of data to train models, raises multiple questions on the regulatory front. Whether training models on copyrighted content qualifies as fair use remains a critical question with divergent global precedents. Though the Delhi High Court’s prima facie opinion in the ANI Media-OpenAI case favours OpenAI, final outcome of the suit remains awaited.

The Digital Personal Data Protection Act (DPDP Act), 2023, has elevated training data legitimacy to a primary concern as acquirers must now examine data lineage, consent mechanisms, cross-border transfers and breach history. These concerns are compounded by the challenge of determining fiduciary versus processor status and identifying whose personal data is embedded in model outputs.

Representations and warranties can hence no longer be boilerplate: sellers must warrant DPDP Act compliance and data inventory accuracy. Indemnity carve-outs for pre-closing non-compliance, with caps calibrated to the DPDP Act’s penalty exposure are likely to become sharply contested as sellers resist open-ended liability for historical data practices.

Talent retention has also driven the rise of acqui-hire structures as seen in Meta’s investment in Cred and Microsoft and Inflection AI’s transaction to secure talent without triggering a full merger review. In India, where non-competes have limited enforceability, acquirers rely on milestone-based earnouts, extended employee stock option plan (ESOP) vesting, and retention bonuses.

Valuing AI deals in India

Valuation sits at the heart of any investment, but AI assets derive value and lose value in ways that conventional intangibles do not. Traditional valuation relying on revenue multiples, earnings before interest tax and depreciation (EBITDA), and discounted cash flow models strain under this reality. A startup with a curated dataset and trained model may generate negligible revenue while representing extraordinary strategic value.

This is significant given the recently introduced threshold of the INR20 billion deal value under the Competition Act, 2002, as investments in AI companies with modest revenues, but substantial strategic value may now trigger mandatory notification from the Competition Commission of India, requiring extended long-stop dates and careful structuring from the outset.

Open-source licensing can also undermine deal value as acquirers may inherit licences, which later require public disclosure of proprietary source code as it happened in Cisco’s acquisition of Linksys. Similarly, third-party dependencies on cloud providers or distribution partners, is itself a risk to be priced.

The ongoing OpenAI-Apple dispute case illustrates how platform integration concerns and unmet revenue expectations can trigger breach of contractual provisions.

Such risks underscore the need for enhanced contractual protections. Representations regarding data ownership, training data licensing, model architecture accuracy, and absence of undisclosed third-party dependencies may warrant classification as “fundamental” with longer survival periods and higher indemnity caps.

Earnouts tied to AI-specific metrics such as model performance, deployment milestones, or compute-efficiency targets can bridge gaps. Interim covenants should restrict material changes to models or datasets, require the retention of key engineers, continuation of data centre contracts and maintain sufficient GPU or cloud capacity. Representations and warranties insurance may help manage such risks, though insurers are increasingly developing exclusions for risks arising from AI-generated outputs, which necessitates further policy negotiation.

The defining challenge is that the assets driving value, namely data, models and talent, are often the hardest to value and protect. Parties that build these considerations into deal architecture from the outset will be better positioned to navigate India’s evolving M&A landscape.

Aneesh Gupte is a partner and Mahak Saxena is an associate at Shardul Amarchand Mangaldas & Co