Enterprises consolidate from four LLM vendors to two

The multi-vendor LLM strategy that made sense two years ago is being unwound. Enterprises that approved four providers to avoid lock-in are cutting to two, and the reason is not price.

Why the list is shrinking

Each approved vendor carries a fixed overhead: a security review, a data processing agreement, an evaluation suite, and someone who understands its quirks. Multiply that by four and the cost is real even if the licence spend is small. When the capability difference between the leading models was large, that overhead bought something. For most production tasks it now buys less.

What is kept, and why two rather than one

The pattern we are seeing is one primary provider for the bulk of workloads and one secondary kept warm for negotiation leverage and for the specific tasks where the primary underperforms. Nobody we spoke to is going to a single vendor, which is the sensible read: the migration cost of being wrong is high enough to justify keeping one alternative live.

What it means for services firms

If your delivery partner’s practice is deep on a provider that just fell off your list, that is now a procurement problem rather than a technical one. Ask any firm you are shortlisting which providers they are genuinely fluent in, and treat model-agnostic tool layers as a real differentiator rather than a marketing line.

What we do not know yet

We have not seen a published policy from any of the enterprises involved, so the scope may be narrower than the practice suggests. It is also unclear whether the consolidation survives the next capability jump.