An informal resale market for artificial-intelligence computing credits was attracting attention on August 16, 2026, as startups looked for ways to sell capacity they did not expect to use. A report from Vectoral described brokers buying credits from young companies and reselling access through marketplaces, direct outreach, private groups and services presented as discounted inference routers.
The account is based on the author’s conversations with founders, emails to brokers and attempts to list credits as a prospective seller. Several founders reportedly received unsolicited approaches from people seeking to buy or sell inference outside providers’ official channels. Credit swapping among startups was already familiar in private forums, the author wrote, but a growing number of dedicated sites suggested that the activity was becoming commercialized.
One respondent offered access representing as much as $100,000 in daily spending. Rather than transfer a provider key, the seller proposed acting as a proxy, receiving requests and apparently routing them through a pool of credentials. The report did not independently establish where those credits came from, who owned the underlying accounts or whether providers permitted the arrangement.
The investigation identified different business models. AI Credits and AICreditMart presented themselves as marketplaces covering major cloud and inference providers. The author submitted a listing to AI Credits, which remained pending at the time of writing. CheapCredits, Tokvana and Neokens were described as routers advertising savings through bulk purchasing. The author questioned whether discounts of around 40 percent could be explained by ordinary volume pricing, but offered that as an industry-informed suspicion rather than a verified finding.
Activity was also visible in Telegram channels and closed startup groups. Across websites, forums and resellers reviewed, the author estimated that tens of millions of dollars in credits might be on offer. That figure was explicitly rough and was not supported by audited transaction data, so it should not be read as a measured market size.
The resale model can reduce the gap between prepaid capacity and actual demand. A startup with expiring or surplus credits may recover some value, while a buyer gains cheaper inference. It also introduces uncertainty about data handling, account control and compliance with provider terms. One broker promoted a data-processing agreement for customers concerned about European privacy rules, but documentation alone does not verify the origin or authorized use of credits.
Vectoral’s report shows that off-market inference is becoming easier to find, not that every listed service is illegitimate. Its evidence consists of public marketing, private outreach and the author’s own marketplace interactions. Provider responses, completed sales and independently verified volumes were absent, leaving the market’s scale and durability unresolved.
