The Rise of AI Credit Resellers and the New Token Broker Economy

Lead

AI inference is expensive, and the credits handed to startups by cloud and model providers are no longer just a nice subsidy that quietly expires. In a recent security research post, Vectoral’s Matt Lenhard describes the rise of “token brokers”: intermediaries who buy unused AI credits from startups and resell discounted access to other customers.

The story is not simply about founders informally swapping perks. The more important signal is that a loosely connected resale market appears to be becoming commercialized, giving token spend a kind of liquidity that makes it behave like a quasi-currency.

Key points

• Brokers are becoming more organized. According to the article, several founders received inbound emails from people offering to buy or sell off-market inference capacity. In some cases, the broker does not provide the original provider key. Instead, the service likely acts as a proxy, choosing from a pool of keys and forwarding requests.
• The apparent supply is meaningful. In one exchange described by the author, a seller claimed to offer $100,000 in spend per day. Based on websites, forums, and reseller channels he reviewed, the author estimates that tens of millions of dollars in credits may be available, though this remains a rough observation rather than a verified market size.
• Distribution channels are broadening. The article points to credit marketplaces, discount routers, Telegram channels, Reddit posts, and private startup communities. Some sites present the discount as the result of bulk purchasing, while others look more directly like marketplaces for buying and selling credits.
• A more formal commercial layer is emerging. Some services are adding business-friendly elements such as data processing agreements. That suggests the market is trying to appear legitimate enough for companies that might otherwise hesitate to route inference through an unknown intermediary.

Why it matters

For buyers, the appeal is obvious: lower inference bills. A startup trying to stretch runway may be tempted by access to popular models at a significant discount. But discounted access can hide serious unknowns. Buyers may not know where the credits came from, who controls the keys, whether prompts and outputs are being logged, or whether user data is passing through an additional proxy.

For providers, resale creates a different problem. Credits are usually designed for a specific purpose: customer acquisition, startup support, cloud migration, or ecosystem growth. When they become transferable inventory, they can be arbitraged. API access that was supposed to be tied to identity, limits, and monitoring can be blended and redistributed through a routing layer, making abuse detection harder.

The larger implication is that inference capacity is becoming a tradable cost asset. If this market continues to grow, cloud and model providers are likely to tighten terms, audit credit usage more aggressively, and restrict transfer or proxy-based resale. Discounted APIs may reduce short-term spending, but opaque supply chains can create security and compliance risks that outweigh the savings.

Source: Hacker News