My take: OpenAI's Broadcom Chip Deal Lacks $18B Financing Plan

The future is bright, but announcements are not architecture

There is something I have always believed about building things that genuinely work: the gap between a public declaration and a functioning system is precisely where the hard work lives. Anyone can draw a diagram on a whiteboard. The web itself, in those early days in Geneva, was not a press release. It was a machine, switched on, doing something real. That distinction matters enormously, and right now, the OpenAI and Broadcom chip story is asking us to forget it.

The partnership was announced last autumn with considerable confidence. Custom silicon, a plan to deploy capacity consuming 10 gigawatts of power by 2030, and a stated ambition to reduce dependence on Nvidia. Fine goals, every one of them. But here is the part that deserves honest scrutiny: the estimated $18 billion required to finance this project has no confirmed funding plan. None. The deal was presented to the world as settled, and it is not settled in the one dimension that actually determines whether it happens.

OpenAI and Broadcom announced a custom AI chip partnership last fall positioned as finalized, with plans to deploy enough capacity to consume 10 gigawatts of power by 2030 and reduce reliance on Nvidia. However, the companies have not resolved how OpenAI will finance the estimated $18 billion project.

I am not raising this to be discouraging. The ambition here points in a genuinely interesting direction. Owning your own silicon stack, building infrastructure that is not beholden to a single supplier, thinking seriously about the compute requirements of the next decade—these are the right instincts. When we built open standards for the web, the principle was simple: no single company should control the critical layer. OpenAI moving toward custom hardware follows that spirit. The concern is not the destination. It is the navigational honesty getting there.

When the announcement becomes the product

We have developed, in the technology industry, a rather curious habit of treating announcements as achievements. A partnership deck becomes a competitive moat. A roadmap slide becomes delivered infrastructure. And the people who fund, regulate, and depend on these systems are left making decisions based on intentions rather than realities.

Ten gigawatts of power by 2030 is not a small claim. For context, that is roughly the continuous output of ten large nuclear power plants, directed entirely at AI compute. Whether or not that scale is desirable, the question of whether it is financed is not a secondary detail. It is the primary question. Infrastructure at that scale requires not just ambition but land, contracts, energy agreements, manufacturing partnerships, and yes, eighteen billion dollars that currently exists only as an estimate.

What this actually means for the open web

Here is what I think is genuinely being missed in the coverage. The conversation is framed as a chip story, a competition story, a Nvidia-versus-everyone story. But underneath it is a question about who controls the infrastructure of intelligence on the web, and whether that infrastructure will be built on solid foundations or on a sequence of well-timed press moments.

Open data, open standards, open government: the things I have spent years working toward all share one property. They are accountable. You can check them. You can see whether the data is actually there, whether the standard actually works, whether the code actually runs. What we need from AI infrastructure is the same quality of accountability. Not polished announcements, but verifiable progress.

The Broadcom deal may well find its financing. The project may well deliver. I genuinely hope it does, because more diverse chip manufacturing capability is good for the ecosystem and good for the web. But the technology industry does itself no favours when it presents unresolved questions as closed chapters. The people building on top of these systems—the developers, the researchers, the companies integrating AI into real services—deserve to know the difference between a plan and a promise.

I hope you enjoyed the article. If you have any questions about the Toronto market reach out directly:

555.555.0322 or start@oneupai.com