In a standard asset sale, the buyer lists the specific assets being acquired and the specific liabilities being assumed. For AI companies, the primary value is often in trained model weights, training datasets, and fine-tuning infrastructure — assets that are difficult to separate cleanly from the company's other operations. The line between what is being acquired and what is being left behind is harder to draw for AI assets than for conventional software code or customer contracts.
In a traditional software company acquisition, an asset sale transfers clearly identified IP assets: registered copyrights, patents, trademarks, trade secrets, and source code. The ownership of each asset can be traced, the transfer can be documented, and the buyer receives clean title to the acquired IP. AI company IP doesn't transfer this cleanly. Trained model weights are the product of a training process that may have used data from third-party sources, open source foundations, and licensed datasets — each with its own terms governing transferability. The fine-tuning pipeline may include proprietary methodologies layered on top of a foundation model licensed from a third party whose agreement restricts transfer. An asset sale of AI IP requires mapping the entire provenance chain of the model before the asset schedule can be written. This due diligence work often reveals transfer restrictions that make a stock sale cleaner despite its tax disadvantages.
In an asset purchase, the buyer acquires assets at fair market value and can depreciate or amortize them over their tax lives, creating a tax step-up. For AI companies, the most valuable assets — trained model weights, training data, and customer relationships — have varying amortization periods under Section 197 and other tax provisions. The tax step-up benefit is one of the primary reasons buyers prefer asset purchases, and sellers demand a premium to accept asset sale treatment that triggers corporate-level tax on the gain.
Buyers in asset sales receive a step-up in the tax basis of acquired assets to their fair market value at closing, creating future tax deductions through depreciation and amortization. For AI companies, the most valuable assets — trained models, proprietary datasets, fine-tuning pipelines — are intangible assets amortizable over 15 years under Section 197. The step-up benefit can be substantial when AI IP represents the majority of purchase price value. But Gurpreet S. Bal notes that in 2026, the analysis has become more complex: the step-up benefit from an asset sale may be offset by the cost and risk of IP transfer, the loss of favorable license terms that don't survive asset transfer, and the time required to clean up the IP provenance chain before the deal can close. The structure decision requires quantifying both sides of that equation.
In a merger or stock purchase, the target entity continues to exist as the surviving corporation or subsidiary, so contracts and licenses generally do not require assignment — they remain with the same legal entity. In an asset purchase, each contract must be individually assigned to the buyer, requiring consent from the counterparty if the contract has anti-assignment provisions. AI foundation model licenses and key customer contracts with anti-assignment provisions are the most common obstacles to clean asset purchase structures.
Forward triangular mergers — the most common structure for public company acquisitions and many private deals — transfer all assets and liabilities of the target by operation of law, without requiring individual assignment of contracts. This is a significant advantage for AI companies whose customer agreements, compute contracts, and data licenses may include change-of-control or anti-assignment provisions that would require third-party consent in an asset sale. Gurpreet S. Bal describes the merger structure as increasingly preferred in AI acquisitions precisely because it sidesteps the consent problem: "The structure decision used to be primarily a tax question. For AI companies, it's also an IP question." The 338(h)(10) election — which allows a stock sale to be treated as an asset sale for tax purposes — can sometimes capture the best of both structures, but requires specific conditions and careful planning.
Parties should evaluate AI deal structure through four lenses: IP transferability (can model weights and training data be assigned cleanly in an asset sale), regulatory continuity (do any AI regulatory approvals or certifications transfer automatically or require reapplication), tax efficiency (does the step-up benefit to the buyer outweigh the additional tax cost to the seller), and contract consent burden (how many key agreements have anti-assignment provisions requiring third-party consent). The structure that minimizes friction across all four dimensions is the right starting point for negotiation.
Gurpreet S. Bal recommends that parties evaluate AI company deal structure decisions against a four-factor framework. Tax efficiency: what is the quantified value of the step-up benefit, net of the tax cost to the seller? IP transferability: does a clean asset transfer require third-party consents that will be difficult or expensive to obtain? Contract assignability: do the target's key contracts — compute agreements, data licenses, customer agreements — have change-of-control or anti-assignment provisions that are easier to navigate through a merger structure? Speed: what is the timeline impact of each structure, and does deal certainty require a faster closing? In recent 2026 AI transactions, Gurpreet S. Bal has seen the IP transferability and contract assignability factors override the pure tax analysis more frequently than in any prior period — a structural shift that reflects the unique characteristics of AI as an asset class.
Gurpreet S. Bal is a corporate partner with 16 years advising on private equity, merger transactions, and public offerings for companies and investors at three of the world's top law firms. He has represented clients in hundreds of transactions with aggregate deal value exceeding $60 billion across AI, semiconductors, fintech, and emerging technology. For more information and to get in touch, visit gurpreetbal.com.