Fine-tuned AI models are no longer experimental. For many mid-market firms, a model fine-tuned on proprietary data has become a competitive asset, embedded in customer-facing tools, internal workflows, and pricing decisions. But the legal and commercial frameworks governing these models often lag behind the technology.
A growing number of mid-market buyers are now negotiating model portability clauses — contractual terms that grant them rights to export, replicate, or continue using a fine-tuned model if the vendor relationship ends. This shift reflects a broader recognition that AI models are not just software features but strategic property.
This explainer examines what model portability clauses are, why they are gaining traction, how mid-market firms are approaching them, and what the implications are for vendors, investors, and the wider AI economy.
What Changed: From Black Box to Negotiable Asset
Historically, enterprise AI contracts treated models as part of a broader SaaS subscription. Customers paid for access, not ownership. Fine-tuning — the process of adapting a base model on proprietary data — was often bundled into the vendor's service, with the resulting model remaining the vendor's intellectual property.
That model is now being challenged. As mid-market firms invest significant time and money in fine-tuning, they are asking a simple question: if we leave the vendor, do we lose the model we helped create?
The answer, in many contracts, is yes. But that is changing. Legal advisers and procurement teams report that model portability clauses are becoming a standard ask in mid-market AI negotiations, particularly for use cases where the fine-tuned model is central to operations — such as customer support automation, document processing, or industry-specific prediction tools.
The shift is not universal. Many vendors resist portability because it undermines lock-in and complicates their own IP strategy. But the commercial pressure is mounting, especially as open-source alternatives and multi-vendor AI strategies become more common.
Why It Matters: Control Over Strategic Assets
For a mid-market firm, a fine-tuned model can be the difference between a generic service and a differentiated one. If the model is not portable, the firm faces a stark choice: stay with the vendor indefinitely, or lose the accumulated value of their fine-tuning investment.
This is not a hypothetical concern. Vendor consolidation, pricing changes, or shifts in a vendor's product roadmap can force a migration. Without a portability clause, the buyer may have to start from scratch — retraining a new model, re-annotating data, and re-integrating workflows. The cost is not just financial; it is time-to-market and competitive position.
Model portability clauses address this by defining what happens to the fine-tuned model on termination. They typically cover:
- Export rights: The buyer can obtain a copy of the fine-tuned model weights, or a functional equivalent, in a usable format.
- Continued use rights: The buyer can continue to use the model internally, even after the vendor relationship ends.
- Third-party transfer: The buyer can move the model to another vendor or an in-house infrastructure.
- Data rights: Clarity on who owns the training data and any derived datasets.
These terms are not standard. They are negotiated case by case, and the outcome depends on the vendor's bargaining power, the buyer's leverage, and the technical feasibility of portability.
Who Is Affected: Buyers, Vendors, and Investors
Mid-market buyers are the primary drivers. They are large enough to have custom AI needs but often lack the legal resources of enterprise giants. They are learning to ask for portability early, rather than discovering the lack of it at renewal time.
Vendors — from AI-native startups to established SaaS platforms — face a strategic tension. Granting portability may reduce lock-in and increase churn risk, but refusing it can make their offering less attractive in a competitive market. Some vendors are responding by offering portability as a premium feature, or by structuring it around additional fees.
Investors are beginning to scrutinise portability clauses as part of due diligence. A vendor's willingness to grant portability can signal confidence in its technology and service quality, while a rigid stance may indicate a reliance on lock-in for revenue retention.
Legal and procurement teams are the operational actors. They are developing playbooks for portability negotiations, often borrowing from data portability precedents in GDPR and cloud computing contracts.
Commercial Impact: Pricing, Lock-In, and Valuation
The commercial implications are significant. For buyers, a portability clause can reduce switching costs, which in turn strengthens their negotiating position at renewal. It also protects the value of their fine-tuning investment, making AI projects more justifiable to finance teams.
For vendors, portability can affect revenue predictability. If a customer can leave with the model, the vendor loses a key retention lever. This may lead to higher upfront pricing or separate fees for portability rights. Some vendors may also restrict portability to certain model sizes or deployment types, such as on-premise versus API-only.
In the broader AI economy, portability clauses could accelerate the commoditisation of fine-tuned models. If models become portable, the competitive moat shifts from the model itself to the data pipeline, ongoing optimisation, and integration services. This is a fundamental change for AI vendors that have relied on model exclusivity as a differentiator.
Risks and Unknowns
Portability is not without complications. Technical feasibility is a major constraint. Some fine-tuned models are tightly coupled to the vendor's infrastructure, making export impractical. Even when export is possible, the model may depend on proprietary preprocessing or postprocessing steps that are not covered by the clause.
Legal uncertainty also remains. The IP status of fine-tuned models is still evolving, and courts have yet to establish clear precedents. A portability clause may be enforceable, but its scope can be contested — for example, whether it covers derivative works or only the exact weights.
There is also the risk of unintended consequences. A buyer who secures portability may find that the model degrades without the vendor's ongoing updates or support. The clause protects the asset, but not the ecosystem around it.
Finally, there is the question of data privacy. If the fine-tuned model contains sensitive customer data, portability may trigger compliance obligations under GDPR or other regulations. Buyers must ensure that their portability rights do not conflict with data protection duties.
FY Outlook
Expect model portability clauses to become more common in mid-market AI contracts over the next 12 to 18 months. As more firms treat fine-tuned models as strategic assets, the absence of a portability clause will be seen as a red flag, similar to the way data portability became a standard expectation in cloud contracts.
Vendors will likely respond with tiered offerings: basic access without portability, and premium tiers that include export rights. Some may also offer portability as a paid add-on, creating a new revenue stream.
Legal frameworks will evolve, but slowly. In the meantime, mid-market buyers should prioritise portability in negotiations, but also consider the practical realities of model transfer. A clause is only as good as its implementation.
Conclusion
The model portability clause is a pragmatic response to a structural shift in how AI value is created and captured. For mid-market firms, it is a way to protect their investment and maintain optionality. For vendors, it is a strategic concession that may reshape their business models. For investors, it is a signal of how the AI market is maturing.
As with any contractual innovation, the details matter. Buyers should seek clear definitions of what is portable, how it will be delivered, and what ongoing support — if any — is included. Vendors should be transparent about technical limitations and pricing. And both sides should recognise that portability is not a zero-sum game; it can create trust and long-term value.
The AI economy is still young, but the contracts that govern it are being written now. Model portability is one of the clauses that will define who holds power in the next phase of enterprise AI.
