Background & Context§
The AI industry runs on compute, and compute is often prepaid through cloud provider credits or API access tokens. Startups, flush with venture funding, often purchase large blocks of inference credits from providers like Anthropic, OpenAI, and others at face value. However, when a startup pivots, shrinks, or simply over-provisions, those credits become idle assets. Historically, such credits were swapped informally among founders in private forums. But a new class of intermediaries—dubbed "token brokers"—has begun commercializing this activity, creating a secondary market for AI credits. This emerging economy matters because it introduces liquidity, pricing arbitrage, and regulatory gray areas into the AI supply chain, potentially affecting provider revenue models and enterprise compliance.
The News: What Happened Exactly§
In an investigative piece published on August 10, 2026, Matt Lenhard of Vectoral shed light on the rise of "token brokers"—individuals and companies that purchase unused AI inference credits from startups and resell them at substantial discounts. Lenhard first encountered this phenomenon through a friend who received unsolicited offers for Anthropic tokens at steep discounts. As he spoke with other founders, he realized that inbound emails offering to buy or sell off-market inference were widespread. What was once a casual practice in startup forums has now been commercialized by dedicated brokers and marketplaces.
Lenhard's own outreach revealed striking details. After several cold emails, one broker responded, offering a staggering $100,000 in spend per day in supply. The broker did not provide direct provider keys but instead acted as a proxy, likely routing requests through a pool of keys. This architecture allows them to hide the original credit holders and aggregate demand. Such scale underscores the maturity of the gray market, which Lenhard estimates to be worth tens of millions of dollars in listed credits across various platforms.
Several dedicated websites have emerged to facilitate credit trading. AI Credits bills itself as a credit marketplace, allowing sellers to list credits for major cloud and inference providers. Lenhard even created his own listing, which was pending approval at the time of writing. AICreditMart offers a similar pure-play reseller marketplace. Another site, CheapCredits, positions itself as a router, claiming its discounts come from "bulk pricing." Lenhard, however, expresses skepticism: a 40% discount is unlikely unless the broker is a top-tier customer of the provider, so he suspects CheapCredits acquires supply through other, undisclosed means. Other brokers like Tokvana and Neokens follow similar patterns. Some even offer GDPR-compliant Data Processing Agreements, suggesting they cater to enterprise clients concerned with data governance.
Beyond websites, the market thrives in less formal channels. Lenhard found relatively active Telegram channels dedicated to trading credits, as well as sporadic Reddit posts. Closed startup groups are also rife with such offers. The liquidity and pseudo-currency nature of tokens have inevitably attracted abuse, and Lenhard warns that as companies become more cost-conscious, crackdowns on these practices are likely imminent. The article includes screenshots of real conversations and site interfaces, grounding the investigation in verifiable evidence.
Historical Parallels & Similar Incidents§
This gray market for AI credits mirrors the cloud computing reserved instance resale market that emerged in the mid-2010s. AWS, Microsoft Azure, and Google Cloud sell reserved instances (RIs) at discounted rates in exchange for upfront commitments. As startups scaled down or folded, they found themselves with unused RIs, leading to a secondary market on platforms like Cloudability and Reserved Instance Marketplace (now part of AWS's own offering). In 2016, AWS officially launched a Reserved Instance Marketplace, allowing third parties to sell their unused RIs at prices they set. This legitimized what had been a shadow market, bringing transparency and control to AWS's ecosystem. The current token broker phenomenon is analogous, but with a key difference: while AWS embraced the secondary market, major AI providers like Anthropic have not officially sanctioned any such resale, leaving it in a legally ambiguous gray zone.
Another parallel can be drawn from the GPU cloud credit scalping that occurred during the Ethereum mining boom of 2017-2018. Miners purchased massive amounts of cloud GPU time, and when Ethereum's price crashed, they offloaded their unused compute contracts on forums and marketplaces. This created a volatile secondary market where credits traded at deep discounts, often with questionable terms. The AI credit market is similar in that it thrives on startup over-provisioning and subsequent distress. However, unlike crypto mining, AI inference credits are tied to specific API access, raising questions about terms of service violations. Providers like OpenAI explicitly prohibit the transfer of API keys, making these trades a violation of contract. The eventual consequences could include account terminations and legal action, similar to how cloud providers cracked down on RI reselling before AWS legitimized it.
A third historical precedent is the mobile data plan sharing phenomenon, where individuals on family plans sold unused data to strangers via apps like Roam Mobility and Flexiroam in the mid-2010s. These services faced regulatory scrutiny and carrier resistance, as they contradicted the carriers' ability to monetize data usage. In the AI context, token brokers similarly undermine providers' pricing tiers and volume discounts. Providers like Anthropic offer tiered pricing based on volume; brokers aggregating demand could negotiate better rates and then undercut the provider's own retail pricing. This could erode profit margins and force providers to implement stricter usage monitoring. The lesson from these parallels is that secondary markets often force primary providers to adapt—either by legalizing and regulating the market, as AWS did, or by cracking down with technical and legal measures, as mobile carriers did.
For AI providers, the risk is not just financial but also reputational, as a gray market invites abuse, including money laundering or service abuse. The fact that brokers operate as proxies, obscuring end users, could complicate enforcement of safety policies. Lenhard's investigation suggests that a crackdown is likely, but he did not indicate any immediate action from providers. As the market matures, providers may follow AWS's example and create official resale marketplaces, or they may aggressively enforce their terms of service. Startups holding idle credits face a dilemma: sell at a discount in a legally gray area, or hold and risk losing the full value as the credits expire. This uncertainty is reminiscent of the early days of cloud RI trading, where early sellers benefited before the market was regulated. For now, the token broker economy thrives in the shadows, with participants betting that the arbitrage window will remain open.
All facts and figures are based on the original investigation by Matt Lenhard of Vectoral, published on August 10, 2026.