The $4 Billion Hint About Why AI Adoption Is Hard

OpenAI launched a new company this month. The OpenAI Deployment Company raised $4 billion at a $10 billion valuation. Its entire job is to embed OpenAI engineers directly inside organizations to help them actually use AI.

That's not a product announcement. That's an admission.

For the first several years of its commercial run, OpenAI's implicit bet was that a powerful enough model sells itself. Make the AI extraordinary, price it reasonably, keep the interface simple, and businesses will figure out the rest. That theory worked up to a point. ChatGPT reached more organizations faster than almost any software product before it.

But wide adoption and real adoption are not the same thing. What OpenAI found, like every enterprise software company before it, is that most organizations have access to AI and have changed almost nothing about how they work. The license is there. The transformation is not.

What they couldn't solve by making the model better

The announcement for the Deployment Company is unusually candid. In OpenAI's own framing, the bottleneck for most enterprise customers is not whether the model can solve their problem in principle. It is whether the customer's data, processes, governance, and security infrastructure can be reorganized to let it.

Read that carefully. The model works. Everything around the model is the problem.

This shows up in a specific way for most businesses. You try AI on a task. In isolation, it does something genuinely impressive. You try to bring it into an actual workflow, and it stalls, or it requires so much manual correction that you've saved nothing. The issue usually isn't the AI. It's that the workflow it's supposed to improve was built for humans doing the handoffs and judgment calls that never got formally documented anywhere. The AI has no way to know what those are. Neither does your tool evaluation process.

OpenAI's answer at enterprise scale is what they're calling Forward Deployed Engineers: people who understand both what the model can do and how the organization actually works, and who sit at that intersection to redesign workflows from the inside. They're backing that approach with a $4 billion joint venture and partners from TPG, Bain Capital, Goldman Sachs, and Brookfield.

That's the price of the deployment problem, when you're trying to solve it across thousands of large organizations simultaneously.

What this means for a business that's not Goldman Sachs

You don't need a team of embedded OpenAI engineers. But you probably do need to stop treating AI adoption as a software evaluation and start treating it as an integration project.

A software evaluation asks: does this tool work? An integration project asks: what does our process need to look like for this tool to work? Most businesses are still asking the first question when the hard part is the second.

A practical place to start: pick one workflow where AI seems like it should help but hasn't delivered. Don't ask whether the AI is capable enough. Ask what information it would need, in what form, to handle this task reliably. Ask who currently catches the exceptions and what they're looking at when they do. Ask what "done" actually looks like, specifically enough that you'd know immediately if the output was wrong.

What you usually find is that the process has to shift before the AI can. Information needs to be accessible. Exceptions need to be defined. Outputs need to be precise enough to evaluate. That is not a technology problem. It's an organizational one. And it's the same problem whether you're a 12-person firm or a 12,000-person bank.

For a smaller business, the same thinking that costs OpenAI billions to deliver at scale costs you an afternoon and a whiteboard.

What the existence of this company actually tells you

OpenAI didn't have to do this. They could have kept selling API access and ChatGPT Enterprise licenses and let deployment be a third-party problem. They didn't, because the data told them that deployment is where value either gets created or stays theoretical.

The same is true at any scale. The AI is capable. The gap between what it can do in principle and what your operation is actually set up to let it do — that's the work. It doesn't require a billion-dollar consulting venture to close. But it does require someone in your business who is deliberately thinking about it.

If that person doesn't exist yet, that's probably the more honest answer to why AI hasn't moved the needle for you.