Cloud AI commitments, short, capped, portable.

Cloud AI commitments price a market that is repricing itself downward every quarter, which makes short terms, caps, and portability worth more than any launch discount.

What do cloud AI commitments actually look like?

Cloud AI commitments take three shapes: provisioned capacity reservations, committed spend folded into the wider cloud agreement, and enterprise platform agreements with AI vendors directly. The provisioned capacity options are documented on Amazon Bedrock pricing, Azure OpenAI pricing, and Vertex AI pricing.

• Provisioned throughput: reserved model capacity at committed rates on platforms like Amazon Bedrock, the AI equivalent of reserved instances.
• Cloud commit integration: AI consumption counting toward EDP, MACC, or Google Cloud commitments.
• Direct platform deals: enterprise agreements with model providers, separate from the cloud bill.

Each shape locks a different thing: capacity, spend, or relationship. Knowing which one the seller is actually selling is the first negotiation skill in this category.

Why do falling model prices change commitment math?

Model prices per token have fallen repeatedly as providers compete and hardware improves, so a multi year commitment at today's rates is a bet against the most reliable trend in the market. Every long commitment we reviewed in 2024 to 2025 was underwater against list within its term.

What this does to standard procurement logic

• Discounts age badly: 20 percent off a rate that drops 40 percent is a loss, not a saving.
• Capacity outlives enthusiasm: provisioned throughput sized at launch ran 30 to 50 percent over steady state in our benchmarks.
• Model churn is real: the model you commit to today may be two generations old before the term ends.

The clauses that compensate

Negotiate price reopeners that reprice committed capacity when list rates fall, model substitution rights that let newer models inherit committed terms, and downscale windows on provisioned throughput. All three are achievable and almost never requested.

How does workload portability create AI negotiation leverage?

A model abstraction layer that lets workloads switch providers is the strongest single lever in cloud AI negotiation, because it converts the renewal from a captive conversation into a competitive one. Estates with demonstrated portability negotiated 10 to 25 percent better outcomes in our engagements.

Portability has a cost: abstraction layers add engineering work and mute provider specific features. Price that cost honestly against the negotiation value it creates, workload by workload.

What negotiation levers work on cloud AI agreements?

Five levers move cloud AI terms: short terms with renewal options, price reopeners, model substitution rights, consumption based ramps, and counting AI spend inside the existing cloud commitment rather than stacking a second one. The stacking error is the expensive one.

1. Commit short: one year with options beats three years flat in a falling market.
2. Write price reopeners tied to published list rate reductions.
3. Secure model substitution rights so committed terms follow capability forward.
4. Ramp provisioned capacity on measured consumption, not launch forecasts.
5. Fold AI usage into the existing cloud commit instead of signing a parallel one.

Treat every AI commitment as an experiment with an exit, because the market will reprice beneath you either way. The contract should let you follow it down.

Where the common advice on AI commitments is wrong

The standard advice from cloud providers and many advisors is to lock multi year AI commitments now because usage will only grow and early commitment earns the best discounts. We disagree. In roughly 10 of the 15 plus AI commitment reviews Morten Andersen ran in 2024 to 2025, the discount captured at signing was smaller than the list price reduction that arrived during the term, meaning the committed buyer paid more than a patient one. The buyer side move is to commit short, cap exposure, write reopeners, and let provisioned capacity follow measured consumption. In a market repricing itself downward every quarter, patience is a discount no seller can match.

What the engagement data shows

Three cuts of our advisory engagement file frame the size of the opportunity.

How to use these numbers

Treat the ranges as negotiation benchmarks, not promises. Your estate sets the baseline; the engagement file tells you what disciplined buyers achieved against the same vendor playbook.

In a market repricing itself downward every quarter, patience is a discount no seller can match.

What to do next

The moves below turn this analysis into a lower invoice at the next renewal.

A sequence you can run this quarter

1. Inventory current AI consumption by provider, model, and workload.
2. Measure steady state throughput needs against any provisioned capacity quoted.
3. Draft price reopener and model substitution language before the term sheet arrives.
4. Evaluate a model abstraction layer for the highest spend workloads.
5. Confirm AI usage counts toward the existing cloud commitment before signing anything parallel.
6. Commit one year with renewal options and revisit against list prices each cycle.

White Paper · GenAI

Cloud AI Commitment Negotiation. The buyer side framework for the contracted cloud AI commit cycle. Three principal cloud AI publishers. One buyer side framework

How to cut a cloud AI commitment across AWS Bedrock, Azure OpenAI, and Google Vertex: the commit bands, the overage traps, and the levers that hold. Read it free.

Read the white paper

Frequently asked questions

Should you sign a multi year AI commitment?

Generally no in current conditions: model prices have fallen repeatedly, and in our 2024 to 2025 reviews the signing discount was usually smaller than the list reduction that arrived mid term. Commit short with reopeners and options instead.

What is provisioned throughput in cloud AI?

Reserved model capacity at committed rates on platforms like Bedrock, Azure OpenAI, and Vertex AI, the AI equivalent of reserved instances. Launch sized reservations ran 30 to 50 percent above steady state consumption in our benchmarks.

Does AI spend count toward cloud commitments?

Yes, consumption on Bedrock, Azure OpenAI, and Vertex AI generally counts toward EDP, MACC, and Google Cloud commitments respectively. Folding AI into the existing commit avoids the double commitment stacking error.

What is a price reopener in an AI agreement?

A clause that reprices committed capacity when published list rates fall, protecting the buyer from paying above market mid term. Reopeners are achievable in negotiation and almost never requested.

How does model abstraction help negotiation?

An abstraction layer keeps workloads movable between providers, which converts renewals into competitive events. Estates with demonstrated portability negotiated 10 to 25 percent better outcomes in our engagements.

Which cloud AI platform is cheapest?

The ranking changes quarter to quarter as providers reprice, which is itself the point: no static answer survives a contract term. Benchmark your specific workloads at each renewal and keep the comparison alive in the contract.