Roads generate the record of everything that happens on them — speed, failure, near-failure, signal timing, incident history. That record belongs to the public that generated it. For decades, it has lived inside proprietary systems, one vendor contract at a time, while the industry's attention stayed fixed on the layers above it: the software, and now the models.

Last week, Nvidia's CEO published a letter tracing that argument to its origin. In the 1980s, open-source software pioneers challenged the belief that software could only advance behind closed doors. Developers chose to study, modify, and improve code together instead. That choice built the shared foundation of knowledge that runs most of the internet today, along with the U.S. military's own systems. The letter's conclusion: the United States now faces the same choice with artificial intelligence. Open weights are the next chapter of the same instinct.

That lineage has one more step. Open-source software freed the code. Open weights free the model. Neither frees the data underneath either one. In transportation, the data is the entire argument. Every signal, every sensor reading, every near-failure a road records about itself sits inside proprietary systems that no open license and no open model will ever reach, because the wall sits one layer below where the openness argument has been fighting.

Open, interoperable transportation data should be the default architecture for every publicly funded intelligent transportation and connected-vehicle program in this country. Not the exception. Not something earned after proving good behavior. The default. The same experiment already ran once, with software, in the 1980s, and it became the floor the internet now stands on without anyone noticing.

The stakes are structural, not rhetorical. Whoever controls the base data layer controls everything built on top of it, permanently — the way whoever controlled the operating system once controlled everything running on top of it, before open source changed that math. The maps that will govern autonomous and connected vehicles for the next several decades are being drawn now, in private, while public attention stays fixed on the model layer above them. Once that map exists in one set of hands, it does not reopen; there is no future open-sourcing that unbuilds a monopoly once the terrain is owned. And once the base layer is owned, using it becomes a toll — a real, recurring charge for access to data that a public road generated in the first place, paid by the public that generated it.

This is not a pitch, and it does not require a majority to be true.

The open-source pioneers of the 1980s did not need agreement to be right; their argument now runs most of the internet. Being early and being correct are not the same as being popular.

There is no list of names attached to this. No coalition has signed on — not because none exists in agreement, but because this argument does not wait for a coalition to feel safe before it becomes public. Nvidia had twenty-five companies standing next to its letter on day one because that conversation had already happened in private. This argument starts that conversation now, in public.

An AI model trained on data it does not control is not open, regardless of the license on its weights. Openness has to reach the layer that generated the record in the first place, or it is not openness — it is a better-decorated version of the same wall.

This is the layer beneath the model layer, beneath the software layer — the one nobody has priced into the openness conversation yet. It is already being decided. The only open question is whether it gets decided in public or in private.

Love,
Vikasa

‍

Download a PDF

‍

Works Cited

1. Huang, Jensen. "Open Weights and American AI Leadership." NVIDIA, July 24, 2026.

2. Muir, Don. "Jensen Huang's Warning to Every CEO: Don't Let One Vendor Own Your AI." Forbes, July 29, 2026.

‍