A list materialized online recently, reportedly showing OpenAI’s top 30 customers, each having processed more than a trillion tokens through its models. OpenAI hasn’t confirmed the list, which first surfaced on its developer forum and was amplified on Reddit, but its existence offers one of the most candid looks yet at the inner workings of the AI reasoning economy. This isn’t just about who’s spending the most on AI, it’s a map of who’s building the future, and how.

Forget the polished keynotes and press releases for a moment. This is the raw data. It shows a frantic race to compound reasoning, where a trillion tokens is just the entry ticket. The list reveals a clear stratification in the market, pointing to four distinct archetypes of companies that are not just using AI, but actively defining its role in the world.

The Four Archetypes of the AI Reasoning Economy

The names on this list aren’t random. They fall into clear, strategic categories that show where the real value (and the real compute) is being concentrated.

1. The AI-Native Builders

These are the challengers, the companies born from the primordial soup of large language models. Their entire product is a reasoning engine. You see names like Cognition (the creators of the Devin AI coding agent) and Perplexity, the AI search engine. They aren’t layering AI onto an old business model, the AI is the business model. They’re living on the edge, pushing the boundaries of what’s possible, and their massive token burn reflects the immense cost of innovation at this scale. Also in this camp are startups like CodeRabbit and Sider AI, which are laser-focused on reimagining the developer workflow from the ground up.

2. The AI Integrators

This is where the incumbents are flexing their muscles. Companies like Salesforce, Shopify, Notion, and Canva are not building foundational models, they’re weaponizing them. Their strategy is to embed AI deeply into existing workflows to increase stickiness and defend their market moats. The “revolutionary” policy mostly revolutionized paperwork for their users. For these giants, every AI-powered feature is a defense against a smaller, more agile AI-native startup trying to eat their lunch. Their trillion-token usage signifies a massive, company-wide commitment to making their platforms indispensable in the age of AI.

3. The Enablers

Every gold rush has its sellers of picks and shovels. In the AI token war, the enablers are the developer tools that make building with AI possible. The presence of JetBrains, Warp.dev, and Datadog on this list is perhaps the most telling statistic of all. It proves that the people building the future are also the heaviest users of the tools to build it. Warp, for instance, has built an entire “Agentic Development Environment” designed to let developers steer AI agents from prompt to production, a task that is extraordinarily token-intensive. They are the critical infrastructure, the plumbing of the AI reasoning economy.

4. The Vertical Specialists

This is where AI moves from a general-purpose tool to a domain-specific expert. Abridge is using AI for medical transcription, tackling the high-stakes, high-complexity world of clinical documentation. Harvey is doing the same for the legal profession. WHOOP is integrating it into fitness and health tracking. These companies aren’t just applying AI, they are training it on the nuanced, proprietary data of their respective fields. Their token usage reflects a deep, intensive process of fine-tuning and inference to solve problems where generic models simply won’t cut it.

The Plot Twists: Duolingo and the Developer Takeover

The list is full of surprises. Why is Duolingo, a language-learning app, reportedly at the very top? The answer is interaction. Every personalized lesson, every grammar explanation, every conversational practice is a reasoning task. It’s a masterclass in applying AI at massive scale to a consumer-facing product, proving that the AI reasoning economy isn’t just for B2B enterprise software.

The other stunning revelation is the sheer dominance of the developer tooling sector. Beyond the obvious enablers, you have Cognition, CodeRabbit, Sider AI, and OpenRouter (an AI model routing service). The prevailing sentiment on developer forums is that AI is fundamentally reshaping software creation, and this list is the proof. It’s a self-reinforcing loop: developers use AI tools to build the next generation of AI-powered applications, which in turn drives even more demand for advanced AI tooling.

The Nagging Question: Is the Token War Just a Bubble?

But before we anoint these companies as the undisputed winners of the new economy, a dose of reality is in order. For every story of incredible adoption, there’s a counter-narrative of a potential bubble. A recent analysis from Yale Insights points out the dizzying web of circular investments between tech giants like OpenAI, Nvidia, Microsoft, and AMD, drawing parallels to the dot-com bubble.

The warnings are coming from the top. Goldman Sachs CEO David Solomon expects “a lot of capital that was deployed that [doesn’t] deliver returns”, while Amazon founder Jeff Bezos called the current environment “kind of an industrial bubble.” Even Sam Altman has warned that “people will overinvest and lose money.”

These concerns aren’t just hot air. A recent study from MIT revealed that a staggering 95% of generative AI pilots at companies have failed to produce any return on investment, despite spending billions. This creates a fundamental tension: the token counts on the leaked list show incredible engagement, but do they translate into sustainable, profitable businesses? Or is this just a massive, well-funded experiment where the bill is eventually coming due?

Beyond the Bubble: The Enduring Shift

Whether we’re in a bubble or not is almost a secondary question. The more important takeaway is that a fundamental shift in software has already occurred. The companies on this list, for all their potential flaws and sky-high burn rates, are the ones actively defining the next paradigm of human-computer interaction.

The “token war” is the new platform war. The real winners won’t just be the companies that burn through the most tokens, but those who can compound that reasoning into durable, valuable products that solve real problems, whether that’s learning a new language, debugging production code, or transcribing a doctor’s notes. The scoreboard is fascinating, but the game is far from over.

Related Articles

llm

We Accidentally Trained AI to Lie to Us

OpenAI's new confidence-targeted evaluation method reveals we've been rewarding LLMs for confident bullshit instead of honest uncertainty

#llm#hallucination#ai-evaluation...

Read More

llm

We Accidentally Trained AI to Lie to Us

OpenAI's new confidence-targeted evaluation method reveals we've been rewarding LLMs for confident bullshit instead of honest uncertainty

#llm#hallucination#ai-evaluation...

Read More