The scale of multi-year AI compute purchase commitments-agreements between model developers and infrastructure providers to reserve computing capacity years into the future-has grown large enough that a handful of these deals can materially shift aggregate deal volume statistics for the entire AI sector, in a way that smaller, more numerous transactions in previous technology cycles never did.
Why compute deals are structurally different
A large compute deal is often a multi-year commitment rather than a single transaction, reported at its full contracted value even though the actual spending will be realized incrementally over the life of the agreement, contingent in some cases on usage that hasn't yet materialized. This makes a single compute deal announcement capable of distorting a reporting period's aggregate deal volume far more than a comparable-sized deal in a less capital-intensive sector would.
What this means for interpreting deal volume trends
Aggregate AI deal volume statistics that don't separate large, multi-year compute commitments from other categories of transaction risk giving a misleading sense of the breadth of dealmaking activity across the sector. A quarter with one very large compute deal and otherwise modest activity can look, in aggregate volume terms, similar to a quarter with genuinely broad-based dealmaking across many companies-even though these represent very different states of the market.
A more useful way to read the data
Separating compute infrastructure commitments from other deal categories, and tracking the number of distinct transactions alongside total reported value, gives a more accurate picture of whether AI dealmaking activity is genuinely broadening or concentrating in a small number of very large infrastructure commitments.
Why this distinction is worth making explicit
Readers using deal volume as a proxy for overall sector momentum deserve to know whether that volume reflects broad-based activity or a small number of unusually large infrastructure bets. Conflating the two produces a picture of the market that's technically accurate in aggregate but substantively misleading about what's actually happening.