Brokers turn unused AI credits into a resale market
This piece is a follow-up to an earlier article the author wrote about the token relay market. That piece left out something the author kept running into: "token brokers," people who buy unused AI inference credits from startups and resell them. The author first heard about this from a friend who was getting offers for Anthropic tokens at steep discounts. Talking to more founders turned up the same pattern: they were getting inbound email from people looking to buy or sell off-market inference. Startups swapping credits was not new, but the author realized the practice was being commercialized, so they sourced brokers' email addresses and reached out directly.
The author's own outreach used email addresses sourced from friends. The first two emails bounced; the third got a response. The seller was offering $100k in spend per day. Rather than handing over provider keys directly, brokers of this kind act as a proxy, probably drawing from a pool of keys and forwarding requests through it.
The author also found dedicated marketplace sites. AI Credits bills itself as a credit marketplace, with a straightforward onboarding process that lets a seller choose a delivery method; the author listed their own credits there and it was still pending approval. AICreditMart is described as a pure-play credit reseller marketplace. Both offer credits across most major cloud and inference providers.
A separate category of sites position themselves as bulk-discount routers. CheapCredits says it achieves its discounts through "bulk pricing"; Tokvana and Neokens work along similar lines. Based on industry experience, the author considers a 40% discount very unlikely unless the buyer is one of a provider's top customers, and their hunch is that CheapCredits is acquiring its supply some other way. CheapCredits offers a Data Processing Agreement for buyers who need to stay GDPR compliant.
Beyond the marketplaces, the author checked the places underground trading usually shows up: a handful of Telegram channels, one of them fairly active, sporadic Reddit posts, and similar posts circulating in closed startup groups.
Across the sites, forums, and resellers examined, the author's rough estimate is that tens of millions of these credits are being offered. Tokens have become a kind of pseudo-currency, and there is enough liquidity in the market for real abuse. As the market matures and companies grow more cost-conscious, the author expects crackdowns on that abuse are not far behind.
Key facts
- A broker who answered the author's direct outreach, after two of the first three emails bounced, offered $100k in spend per day, and works as a proxy rather than handing over provider keys directly.
- Dedicated marketplaces AI Credits and AICreditMart, and bulk-discount router sites CheapCredits, Tokvana, and Neokens, sell AI credits across major cloud and inference providers; CheapCredits offers a GDPR Data Processing Agreement.
- The author doubts CheapCredits' claimed bulk-pricing discounts, calling a 40% cut "very unlikely" unless the buyer is a top customer, and suspects the supply is sourced some other way.
- Brokers also operate through a handful of Telegram channels, sporadic Reddit posts, and closed startup groups.
- The author's rough estimate is tens of millions of credits being offered across the sites, forums, and resellers examined, with crackdowns expected as companies grow more cost-aware.
Why it matters
AI inference credits have quietly grown a secondary resale market that runs outside providers' own billing systems: dedicated marketplace sites, bulk-discount routers, and brokers who reach startups directly by email. The author's earlier piece tracked the token relay market that moves this supply around; this one shows the commercial layer built on top of it, with tokens functioning as a kind of pseudo-currency that trades on its own terms.
Who it affects
Startups sitting on unused API budgets are the target of broker outreach emails looking to buy or sell off-market inference. On the buying side are people chasing cheap access to providers like Anthropic, including the author's friend who received discounted Anthropic token offers directly. Privacy-conscious buyers are also a target audience: CheapCredits markets a Data Processing Agreement specifically for GDPR compliance.
How to use it
AI Credits and AICreditMart operate as marketplaces; AI Credits has a straightforward onboarding flow that lets a seller pick a delivery method, and the author listed credits there, still pending approval at the time of writing. CheapCredits, Tokvana, and Neokens position themselves as bulk-pricing routers instead of marketplaces. Brokers can also be reached by direct email, and the same activity shows up in a handful of Telegram channels, on Reddit, and in closed startup groups.
How solid is it
This is a first-person investigation. The author sourced broker email addresses from friends and tested them directly, two of three bounced, the third produced a live offer, and browsed the listed sites both as a prospective buyer and seller, backing the account with screenshots. The "tens of millions" credit-volume figure is explicitly a rough estimate across what the author looked at, not a measured total, and the claim that CheapCredits sources its supply through other means is labeled a hunch, not a confirmed finding.
Risks and caveats
The piece does not say which provider's credits were behind the $100k-a-day broker offer; the Anthropic mention is a separate case, the author's friend receiving discounted Anthropic token offers, not the same transaction. It also does not identify who runs AI Credits, AICreditMart, CheapCredits, Tokvana, or Neokens, gives no date for when the outreach or site research took place, and does not explain how CheapCredits or the other router sites actually obtain the credits they resell beyond the author's own speculation. The author frames the market itself, now that tokens carry real liquidity, as ripe for abuse.
“a 40% discount is very unlikely unless you are one of the provider's top customers”
— the author of the Vectoral piece on AI credit brokers