OpenAI’s Trillion-Token Club Exposes Who’s Leading at AI

OpenAI recently released a list of its top customers — 30 companies and individuals that have each processed more than a trillion tokens using its models.

To give that some context, that’s hundreds of billions of words across thousands of applications, all running continuously. Every query, every document summary, every chatbot reply counts. Trillions of times.

The first thing the list shows is that AI at scale isn’t happening in startups. The heaviest users are established, product-heavy companies quietly embedding AI across operations. Uber, Duolingo, Salesforce, Notion, Shopify, T-Mobile, Zendesk, Canva, Datadog, Mercado Libre — these names dominate.

A few startups like CodeRabbit, Cognition, OpenRouter, and Sider AI appear, but they are the exception.

Token usage reflects real integration, not experimentation. While some on social media argue about the privacy policy of OpenAI as they revealed the name of the top customers, others are mostly concerned with what this tells the creators and users of AI about where AI is actually being used the most.

Scale and data determine usage

Uber, to start with, is running AI across logistics, pricing, driver and rider safety, and deliveries. Every trip uses predictive models, increasingly backed by language models and multimodal AI. The token count reflects continuous, real-time operations.

Duolingo uses AI to personalise lessons at scale. Every adaptive exercise and correction burns tokens. Salesforce automates sales and service workflows. Notion powers writing, summarisation, and task assistance. Indeed refines job matching.

Each of these requests adds to the trillion-token total, which isn’t sandbox testing, but live continuous AI powering core business functions. This is different from other startups in India using OpenAI APIs to become AI-native while using it to create a completely new product.

Most companies on the list aren’t AI companies. Shopify, Zendesk, T-Mobile, WHOOP, Datadog, Mercado Libre — AI is invisible, running behind the scenes, powering logistics, support, recommendations, and monitoring. It’s not a feature. It’s a core layer.

That’s why token-heavy usage is almost always tied to companies with mature products and established operations.

Startups innovate, but few can hit this level because they lack user volume and integrated data streams. Even AI-first startups that appear on the list burn tokens at a fraction of what scaled companies do. Token-heavy usage correlates directly with scale. Millions of users, continuous workflows, and integrated data streams drive the numbers.

These numbers, along with the fact that OpenAI released AgentKit for developers at DevDay to build agents inside ChatGPT itself is also a threat to a lot of startups building solutions on top of OpenAI APIs, and considering it as their moat.

AI has become infrastructure

OpenRouter, CodeRabbit, Sider AI, and Cognition show that small teams can scale fast. But for most startups, AI usage is focused on experimentation, early feature development, or internal tooling. Reaching the trillion-token mark in production requires millions of live interactions, which most startups haven’t achieved.

Processing a trillion tokens isn’t cheap. It signals a strategic bet. These companies are investing heavily in AI, whether through API usage, fine-tuning, or internal integrations.

At the same time, token consumption signals both influence and dependence. Being on this list shows who is deeply reliant on OpenAI infrastructure.

Competitors can see who is committing at scale. If OpenAI raised prices or throttled access, these companies would feel it first. Social media discussions highlighted the scale — OpenAI alone is processing roughly 3 quadrillion tokens annually, approaching the total human word output.

That gives these companies operational leverage, but also dependence. Alex Issakova, CEO of Huckr AI, had earlier posted this question on LinkedIn: “80% of AI startups depend on APIs from companies burning billions. How long can that last?”

That dependency is the core weakness. When OpenAI raises prices, when Anthropic cuts back credits, or when Google shifts its model tiers, entire businesses get shaken.

Read: AI Startups Depend on Costly APIs of Companies Burning Billions

This list is a wake-up call. AI is no longer experimental. It is infrastructure. Duolingo, Uber, Notion, Salesforce, and others aren’t just customers; they are the engines of the AI economy, running models continuously across millions of interactions.

Startups can innovate, but at scale, the real power lies in companies quietly wiring AI into the systems we use every day.

The post OpenAI’s Trillion-Token Club Exposes Who’s Leading at AI appeared first on Analytics India Magazine.