Micro1, a startup that supplies training data for AI models, has reached a $500 million gross run rate. That’s the kind of growth that happens when you’re selling picks and shovels during a gold rush.

The company trains and manages human annotators who label data, review AI outputs, and perform the tedious work that makes machine learning possible. It’s not glamorous, but it’s essential. Every frontier model needs millions of examples to learn from, and someone has to create those examples.

Micro1’s timing is perfect. As AI labs race to train bigger models, they’re burning through training data at an unprecedented rate. OpenAI, Anthropic, Google, and others are all competing for the same limited resource: high-quality human feedback. That competition is driving prices up and creating opportunities for companies that can deliver data at scale.

The $500 million run rate puts Micro1 in rare territory for a data labeling company. Traditional players in this space like Scale AI have raised billions and achieved valuations to match, but they took years to get there. Micro1’s rapid ascent suggests the market has intensified significantly.

What’s different now is that AI companies aren’t just buying labeled images or transcribed audio anymore. They need human evaluators who can assess whether a model’s response is helpful, accurate, and safe. They need domain experts who can verify technical claims. They need creative writers who can generate training examples for tasks that don’t have naturally occurring data. This more sophisticated work commands higher prices, which is good news for Micro1 and its competitors.

The business model is straightforward: hire and train people to do data work, then sell their output to AI companies at a markup. But execution is harder than it sounds. You need robust quality control, fast turnaround times, and the ability to scale up or down based on client demand. You also need to manage a distributed workforce across multiple time zones and languages.

Micro1’s growth also reflects a broader trend in AI spending. Companies are pouring money into model development right now because they believe better models will translate into better products and, eventually, revenue. Whether that bet pays off remains to be seen, but for now, the money is flowing.

The stickiness question

New data from enterprise usage suggests businesses are less loyal to specific AI providers than investors might hope. A recent analysis indicates OpenAI is gaining ground on Anthropic with business users, but the real story is how willing companies are to switch providers as new models launch.

When Anthropic releases a better model, businesses shift usage toward Claude. When OpenAI releases GPT-5 or whatever comes next, they shift back. This volatility should concern both companies because it suggests enterprise AI spending isn’t sticky yet. Businesses are treating AI providers like commodity services, not strategic platforms.

That’s a problem if you’re trying to build a sustainable business with predictable revenue. It’s much easier to justify a high valuation when customers are locked in through integrations, workflow dependencies, or switching costs. Right now, those switching costs are minimal. Most businesses access these models through APIs, and swapping one API for another is trivial.

The race to build stickiness is already underway. OpenAI is pushing ChatGPT Enterprise and custom GPTs. Anthropic is investing in Claude for Work and vertical-specific solutions. Both companies are trying to move up the stack from raw API access to integrated products that embed themselves into daily workflows.

Whoever figures out how to make their AI indispensable wins. Right now, nobody has.

Brockman’s moment

While usage patterns shift and startups chase data contracts, OpenAI is dealing with its own internal dynamics. Greg Brockman, the company’s president and cofounder, has quietly accumulated significant power over the past year.

OpenAI has been through a lot recently. A messy legal battle with Elon Musk. A trade secrets lawsuit from Apple. Scrutiny over an unreleased model that reportedly hacked another AI company. And as the company prepares for an IPO, key executives have left.

Through all of it, Brockman has consolidated influence. He’s always been important at OpenAI, but his role has expanded as others have departed. The company hasn’t announced major structural changes, but people close to OpenAI say Brockman is increasingly central to major decisions.

This matters because OpenAI’s leadership structure has always been unusual. Sam Altman is the face of the company and the CEO, but OpenAI’s power dynamics are more distributed than a typical startup. Brockman’s growing influence could signal a shift toward more traditional hierarchy, or it could just reflect the natural evolution of a company preparing to go public.

Either way, it’s worth watching. OpenAI’s internal politics have always influenced the broader AI industry, whether through personnel moves, research priorities, or strategic decisions. If Brockman is taking on a larger role, that will shape where OpenAI goes next.