← PastBehavior Research
Download PDF
PastBehavior Fieldwork 051 Evidence to 8 September 2026

When the Commitment Becomes Currency

Companies like AWS and Anthropic are letting customers use money they already committed to buy other companies' products. History says that can move a deal, but it does not automatically change how companies buy.

What is happening

Anthropic is letting customers spend their Claude budget on other companies' software

Since March 2026, enterprise customers can use part of the money they have already committed to Anthropic to buy products from companies such as GitLab, Harvey, Snowflake and Replit. Anthropic handles the purchase and sends the invoice. It takes no commission.

At first glance that looks like Anthropic becoming a new enterprise software store, and much of the coverage has read it that way. The historical record points to something narrower, and more interesting.

This arrangement is not new. AWS, Microsoft, Google Cloud and Snowflake have all run a version of it, and one large company ran the same structure for thirty years and deliberately refused to open it. Watching what those companies actually did, rather than what they said, gives a fairly clear picture of what Anthropic is buying, what it is giving up, and how far this is likely to go.

The historical lesson

AWS already showed us the rule

On 1 May 2025, AWS moved two levers in opposite directions on the same day. Almost nobody outside the trade noticed, and it is the clearest thing in this study.

It opened the shelf

Any software company could now list in AWS Marketplace, including products that do not run on AWS at all. More sellers, easier discovery, one invoice.

It closed the wallet

Only software hosted entirely on AWS could be paid for with a customer's committed AWS dollars. Everything else had to be bought with new money.

The catalogue tells you what a platform is willing to sell. The eligibility rule tells you what it actually wants customers to spend committed money on.

The implication is worth sitting with. AWS was happy to distribute almost anything. It was only willing to let an outside product absorb dollars a customer had already promised AWS when that product still drove more AWS usage. A customer's committed budget is money AWS has effectively already won. Letting an outside company take a slice of it costs AWS real revenue, unless the outside product brings the consumption back through the front door.

That is why the two rules moved apart. Distribution is cheap to give away. Committed dollars are not.

The same choice, made four more times

This pattern is not just AWS

Microsoft

Customers can put Azure commitments toward selected partner products, with no published limit on how much. The catch is which products qualify: only ones Microsoft has already designated as strategically tied to Azure.

Google Cloud

Eligible products have to run on Google Cloud, which is a condition of appearing in the marketplace at all. Partner purchases are capped at a quarter of the customer's commitment.

Snowflake

Customers can set aside part of their Snowflake commitment for products sold inside Snowflake Marketplace. They have to opt in when the contract is signed or renewed, and only products sold through Snowflake itself qualify.

AT&TThe counterexample

AT&T sold outside companies' hardware and software to its enterprise customers on the same contracts and the same invoices, and had done so for decades. It still chose not to let any of those purchases count toward the customer's minimum revenue commitment. Equipment and outsourcing were written out of eligibility in the contract language itself. Nothing technical stood in the way. AT&T simply decided that money promised to AT&T should be spent on AT&T.

AT&T is the entry that turns an observation into a rule. When a company has every reason and every capability to open committed money to outside products, and still writes the contract to prevent it, the choice is deliberate.

The rule underneath

Platforms only open committed money when the outside purchase still helps them

This was tested deliberately, across several industries and several decades, looking for a single commercial company that voluntarily let customers spend guaranteed commitments on unrelated products that gave the company nothing back. None was found.

If the company is a distributor

Selling other companies' products is already the business. A wholesaler or an outsourcer letting you buy third-party goods against your commitment is not opening anything, because those goods are what you contracted to buy from them in the first place.

If the company makes its own product

Letting unrelated purchases count against guaranteed revenue means giving up revenue for no return. There is no version of that trade that improves the company's position, which is why nobody has been found doing it.

Energy contracts make the point from the other direction. Under take-or-pay gas supply agreements, buying from an alternative supplier did the opposite of retiring the commitment. It created the shortfall. Through the 1980s, pipeline companies watched their obligations mount precisely because customers had gone elsewhere. The remedy on offer was always more gas from the same seller, later.

The only arrangements found anywhere that let committed money reach genuinely unrelated products were created by regulators or by governments as a condition of a deal. No company chose it.

The second lesson

Moving the first deal is not the same as owning the buying channel

Committed money is powerful because it removes one of the hardest objections in an enterprise sale: finding new budget.

A customer may already want the software. What stops the purchase is having to create a new budget line, onboard a new vendor, run a new security review and negotiate a new contract. If the purchase can instead come out of money already promised to AWS, Microsoft or Anthropic, most of that disappears. Finance stops being asked to approve a cost and starts being offered a way to recover money that was going to be forfeited. Sellers report faster deal cycles.

What is much harder to find is any sign that the customer then keeps buying through that marketplace once the pressure to consume the commitment is gone. Three things stand out.

AWS

Enterprise deals on AWS Marketplace are almost all privately negotiated offers, and until 1 September 2026 those could not auto-renew at all. Every renewal required an affirmative new action. AWS ran the largest marketplace of its kind for more than a decade without one of the basic features that helps a buying channel become habitual, which sits awkwardly with the idea that it had already become permanent buying infrastructure.

AWS announcement, 1 September 2026
Microsoft

Its own documentation states that Marketplace does not connect to enterprise procurement and workflow systems, and that companies with procurement policies have to manage those separately. A buying route that a company's procurement system cannot see is not yet that company's buying route.

Microsoft documentation
Buyers

No named enterprise policy was found that made marketplace purchasing the default independent of committed-budget pressure. Where such a policy is claimed at all, it is unnamed, it is limited to cloud-related software, and the reason given is always the committed budget first.

Winning the transaction

Well established. Every party to these deals describes it the same way, and none of them disputes it. Committed budget gets the purchase approved.

Owning the route

Not established. Cancelling a marketplace subscription is easier than cancelling a direct contract, not harder. The machinery that would make the route permanent is only now being built.

Applying the history

Claude Marketplace will probably win transactions before it wins procurement

Anthropic has recreated the same basic structure the cloud providers built, and it has done so from day one rather than a decade in.

Enterprise customers already hold committed Anthropic dollars. Selected Claude-powered products can absorb some of those dollars. Anthropic takes no cut, which removes the objection that has slowed sellers down on every other marketplace. All of that should make it meaningfully easier for those partner products to get bought, and that is a real commercial advantage worth having.

What the historical pattern does not support is treating Anthropic as becoming the default place enterprises buy software. Four reasons, all of them visible today rather than speculative.

  • Products have to be Claude-powered, which is the same class of rule Google Cloud used and which by design keeps the catalogue inside Anthropic's own consumption.
  • The partner set is still small, roughly six to eight products, against thousands on each cloud marketplace.
  • How much of a commitment can actually be spent this way is not public. Anthropic describes it only as a portion.
  • No renewal or procurement workflow has been demonstrated, and that is what turned out to be missing at AWS for fourteen years.

If Anthropic ever lets products with no Claude connection absorb Anthropic commitments, everything above stops applying.

No company in the historical record has done this, which is why it would be the single most informative thing Anthropic could do. It would mean the arrangement had become something genuinely new rather than a faster version of what AWS built.
What happens next

What to watch now

Anthropic keeps eligibility narrow

Products will continue to need a clear Claude connection.

Every company that has run this arrangement held that line and none loosened it. Expect tightening instead, most likely a requirement that partners run inference through Anthropic directly rather than through a cloud provider.

A limit appears on how much committed spend can go to partners

Anthropic will disclose or enforce a cap, and it will sit well below half the commitment.

AWS and Google both cap this at a quarter. Snowflake makes customers reserve a portion in advance. Anthropic already describes the eligible amount only as a portion, which suggests the limit exists and has not yet been published.

Renewal becomes the next battleground

Anthropic will build auto-renewal and easier repeat purchasing before it builds much else.

Staying power needs renewal pricing that carries forward and purchasing that does not restart each year. AWS shipped exactly this in September 2026. When Anthropic follows, read it as finding the ceiling rather than clearing it.

Procurement-system integration matters more than adding hundreds of apps

Partner count will be the number everyone reports and the least informative one.

The signal worth tracking is whether enterprise procurement systems can see the route, approve against it and enforce it. None of the marketplaces reviewed here has demonstrated that yet, Microsoft included, by its own account.

The most important transaction will be one made after the budget advantage is gone

Watch for a customer that has used all its eligible Anthropic commitment and still buys the next product through Anthropic.

At that point the choice is no longer financial and the behaviour has become structural. One such purchase would tell you more than any volume figure Anthropic publishes.

Competing commitments create a new problem

Large buyers will start deciding which commitment a given purchase should consume.

A company may hold committed spend with AWS, Microsoft, Snowflake and Anthropic at once, and one purchase cannot retire all of them. Snowflake has already written the conflict into its rules by excluding contracts bought through Google Cloud.

What to do about it

Three decisions, depending on which side of the table you sit

If you sell software

  • Ask about committed cloud or AI spend early in the sales process, not after a budget objection.
  • Use it to remove budget friction and win the first transaction. This is the best-supported move in the whole study.
  • Do not assume a marketplace sale makes the customer sticky.
  • Forecast the renewal as a sale you still have to win.
  • Make sure salespeople are not financially penalised for marketplace deals. That is where this fails first.

If you run a platform

  • Treat the eligibility rule as a strategic decision, not a marketplace feature.
  • Decide exactly what kinds of outside spending you are willing to let replace your own revenue.
  • Expect to cap it. Almost everyone else has.
  • If you want this to become permanent infrastructure, build renewal and procurement workflow rather than more listings.

If you are the buyer

  • Use committed spend to reduce waste, and do not let an expiring budget make the product decision for you.
  • Understand which purchases count before signing a large commitment, not after.
  • Watch the limits, and check whether spending this way costs you discount elsewhere.
  • When several platforms could claim the same purchase, decide deliberately which commitment you want it to consume.
The one unresolved question

The answer may already be sitting inside AWS

Marketplace spend counts No longer counts 0% 25% limit 100%

AWS caps eligible Marketplace spending at a quarter of the annual commitment. Once a customer reaches that limit, buying more through Marketplace no longer helps retire the commitment, and under some contracts it costs the customer discount value it would have earned by spending with AWS directly.

If customers keep buying through Marketplace anyway, that would be strong evidence the Marketplace has become genuinely valuable as a buying channel in its own right. If they stop at the line, then committed budget was the whole story.

AWS has this data. It is not public, and answering the question requires access rather than more history.
Sources and how this was tested

What was examined

The study compared five companies that ran committed-spend arrangements with enterprise customers: AWS, Microsoft, Google Cloud, Snowflake and AT&T. Salesforce AppExchange and the HubSpot App Marketplace were used as comparisons, since both run large partner ecosystems without any committed-budget component.

A separate search then looked across telecommunications, energy, television and digital advertising, aviation, defence trade, pharmaceutical and industrial distribution, IT outsourcing, cooperative purchasing and general commercial contract law, covering roughly fifty years, for any company that voluntarily let guaranteed commitments be spent on unrelated outside products. Primary sources were preferred throughout: filed contracts, published tariffs, regulatory records and operator documentation.

Important limitations

  • No public transaction-level dataset exists to measure the size of the effect. No company publishes how much of its marketplace volume is driven by committed budget rather than by ordinary demand, and no rule change anywhere has a published before-and-after measurement.
  • In every case where committed money could be spent on outside products, those products also had a technical connection to the platform. That makes it impossible to separate the budget effect from the technical one using public information.
  • Nothing was found showing when during a contract year these purchases actually happen, which is the most direct test of whether this is a year-end clean-up or a standing habit.
  • Most published figures on why buyers prefer these marketplaces were commissioned by the platforms themselves and are treated accordingly.
  • Four of the five companies examined are in enterprise technology and all are recent. The one case from another industry and era is AT&T, and it is a negative.
  • Whether offering broader eligibility causes customers to sign larger commitments in the first place could not be settled, and is not asserted in either direction.

Sources