The end of elastic: cloud procurement in the post-credit AI era

For the past two years, AI founders have been racing. Building fast. Hiring dead early. Scaling infra like growth was guaranteed.

And in many ways, I cannot blame them as the system was very much encouraging it:

• Investors poured in pre-product
• Cloud providers handed out (unreal) generous credits
• Infra commitments could be pooled, resold, or arbitraged behind the scenes

You had every right to burn like an enterprise, even if you were still acting like the seed-stage startup you were.

But that window just closed.

The loophole era is over

This summer, AWS quietly updated its terms of service. No more pooling reserved instances. No more reselling unused savings plans. No more soft exit routes for overcommitment.

So what’s actually changed?

• No more cross‑customer discount sharing/pooling via resellers
• Each company must hold its own RIs/Savings Plans to realize discounts
• Existing pooled discounts won’t cover you; idle vendor commits don’t transfer

What used to be a flexible, if somewhat unofficial, ecosystem of third-party optimisers and cloud brokers has now been shut down…at its very root.

If you do nothing: usage prices revert to on‑demand in your account.
If you overreact: you lock into 1–3‑year commitments you might not use.

The message is now clear:

Every commitment is now a liability. Your liability. And yours alone.

No transfers. No splitting across orgs. No discount arbitrage through your VC’s infra partner.

For AI startups, the timing could not be worse

The current iteration of AI has been built on a mix of ambition and abstraction.

Ambition to define a category. Abstraction from real costs.
Founders were chasing adoptions.
Investors were underwriting a generational platform shift.

And cloud providers were throwing out $100k-$1m credit packages as if it were standard PLG practice.

But those credits do not last forever. Most of these expire in 12-24 months. If it hasn’t already, the expiration date is fast approaching. And when they do, many companies will find themselves holding the bag of aggressive, inflexible commitments: based on usage forecasts that were never real in the first place.

This silence is the LOUDEST signal

I’ve spent this month speaking with Founders and CFOs. Some were vaguely aware AWS had “changed something.” None of them had a concrete plan for what to do next.

For me, that is what makes this moment in history so, so risky.
There’s no alert flashing. No email from your rep. No pricing model overhaul.

Just a new set of rules. Quietly pushed live as of June 1, 2025. Fully enforceable.
And likely already in motion against your cost base.

What to do now

If you’re still within your first two years, and running on credits:

• Pull forward your post-credit cost model
• Recalculate what your commitments look like once real spend begins
• Ask your infra vendors what happens now that resale is off the table

If you’re already spending (hard):

• Double-check the savings plans you’ve locked into
• Make sure commit forecasts match actual usage
• Treat cloud like debt: know the terms, know the exit conditions, know the risk

Best thing to do here is to ask your vendor a single question:

“What is our savings and commitment utilisation if we remove your pooled capacity tomorrow?”

If they can’t show it per account, assume it does not exist.

This isn’t the crash. But it might be the cliff

This policy shift will not kill this current generative AI-infused boom. But for sure, it will accelerate the separation between narrative and unit economics (ie., shiny shiny vs actual business).

Because when those credits vanish and the commitment terms tighten, Founders and CFOs won’t be negotiating with hope. They’ll be negotiating with AWS.

And this time, AWS (re)wrote the rules.

Cloud moved. Quietly. Permanently.

Ask yourself: are you still building like it didn’t?