Perpetual vs term data license: which should you grant?

What is the difference between a perpetual and a term data license?#

A perpetual data license grants use rights with no end date, while a term license grants them for a defined period, after which the rights stop unless renewed. Both are licenses: the company keeps ownership of its records in either case, and the buyer receives only the rights the contract describes.

Two other words often get confused with perpetual. Irrevocable means the licensor cannot take the rights back at will, though many irrevocable licenses can still end for uncured breach. Fully paid-up means no further payments are owed. A license can be perpetual but terminable for breach, or term-limited but fully paid-up at signing, so read each word on its own.

What survives the term in an AI data license?#

What survives the term is the question that decides most AI license negotiations, because a model trained on licensed records keeps what it learned after the files are deleted. Removing specific training examples from a trained model is difficult in practice, so buyers commonly ask that models trained during the term stay in use after expiry.

That leaves three separate things to negotiate: the raw records, derived artifacts such as cleaned subsets, labels and embeddings, and the trained models. A common term structure requires the buyer to delete raw records and derived files at expiry and certify the deletion, while allowing models already trained to remain in use. Without that split, a term license can quietly behave like a perpetual one.

Perpetual vs term license compared#

The table compares the two structures on the points CFOs and counsel raise most often. Revenue timing is shown as a general pattern only: under ASC 606, a license of functional intellectual property is generally a right to use recognized at a point in time, and renewal revenue is recognized no earlier than the start of the renewal period. How a specific data license is classified is a judgment for your auditors.

When a perpetual license makes sense#

A perpetual license makes sense when the records are a closed set that will not change and the price reflects the full value of permanent use. Archives from a retired help desk, a migrated ERP or a wound-down business fit this pattern: no refreshes are coming, and the company may prefer one clean payment over a long contract it must administer.

Perpetual grants also work better when the permitted use is narrow. Perpetual rights to train and evaluate models in a defined field, with no right to resell the records and no exclusivity, carry less long-term risk than broad perpetual rights across every use.

The cost is leverage. Once a perpetual grant is signed, any later rise in what the records are worth belongs to the buyer. Price that in before agreeing, and confirm the grant covers only records delivered at signing, not records the company creates afterward.

When a term license makes sense#

A term license makes sense when the records keep growing and recency is part of their value. Support conversations, dispatch histories and order exceptions from a live business become more useful with each refresh, and a term structure lets each renewal reflect what the newer records are worth.

Term licenses also limit exposure to change. If customer contracts, privacy rules or the company's own plans shift, an expiring license is easier to adjust than a perpetual one. For a company that may be sold, a license that expires or renews on clear terms is usually simpler for an acquirer to review. Hybrid structures can combine the strengths of both.

• 1Perpetual training rights on a fixed historical snapshot, plus a term subscription for refreshes.
• 2Term access to raw records, with continuing rights only in models trained during the term.
• 3Perpetual non-exclusive rights, with exclusivity limited to a short period after delivery.
• 4Perpetual rights in one field of use, with other fields licensed separately on a term.

Clauses to read together before you sign#

Clauses that decide duration rarely sit in one place. Read them together, because a grant section that says term can be undone by a survival clause that keeps everything alive.

• 1Grant: whether the license is perpetual, term-limited, irrevocable or fully paid-up, and for which records.
• 2Term and renewal: start date, renewal mechanics and what notice ends the license.
• 3Effect of termination: deletion or return of raw records and derived files, with a written certificate.
• 4Models and derivatives: which trained models and artifacts survive, and whether new versions may be trained after expiry.
• 5Survival: which duties, such as confidentiality and no re-identification, continue after the term.
• 6Assignment and change of control: whether the rights move if either party is acquired.

Illustrative: a 3PL with a closed archive and a live feed#

Illustrative: a fictional third-party logistics company retired its old warehouse management system after a migration, leaving years of order exceptions, inventory adjustments and carrier claims in a read-only archive. Its current TMS keeps producing new exception records every day.

The company licenses the closed WMS archive on a perpetual, non-exclusive basis, limited to training and evaluating models for logistics operations, with no resale. For the live TMS exceptions it grants a term license with periodic refreshes, deletion of raw records at expiry and continuing rights only for models trained during the term. The archive is priced once; the live feed is repriced at each renewal.

How SourceX approaches license duration#

SourceX settles duration during the Rights and Approval steps of the SourceX five-step transaction, after the records are scoped and before the supplier signs. The supplier approves the term, the permitted use and what survives, and those choices are recorded under permitted use and release authorization in the SourceX Evidence Packet.

SourceX does not default to perpetual or term. The right structure depends on whether the records are a closed archive or a growing feed, on advice from the supplier's counsel and auditors, and on what the buyer needs to train and evaluate its models.