By moving away from Nvidia, Big Tech isn't just chasing margins—it's turning frontier AI into a regulated infrastructure where the government decides who has the right to compute.
A silent consensus is taking hold in Cupertino, Mountain View, and San Francisco: depending on Nvidia for everything has ceased to be a product strategy and become a strategic vulnerability. According to TechCrunch, companies like OpenAI and SpaceX are accelerating the development of their own silicon—such as the inferential "Jalapeño" chip, designed in partnership with Broadcom—to escape vendor lock-in. But reading this shift merely as a cost-cutting move or a way to ease production bottlenecks misses the bigger picture. The true consequence of AI labs baking their own chips is that frontier computing ceases to be a free-market commodity and becomes a regulated infrastructure under the direct tutelage of the State.
The mathematics of risk have changed. While processing power was concentrated in the hands of a single outsourced manufacturer, the market absorbed geopolitical tension as a mere supply chain problem. However, when the very architecture of frontier models begins to be designed and operated internally by a handful of companies holding the densest capital on the planet, silicon gains the status of a critical utility. And critical utilities, in any developed economy, attract regulation. It is no coincidence that the U.S. executive branch has positioned itself not merely as a dispenser of factory subsidies, but as the gatekeeper of access to technology.
In this new landscape, the unrestricted availability of artificial intelligence dies by asphyxiation. In my view, the premise that any well-capitalized startup can buy its way to the AI frontier is numbered. The United States government has already adopted a case-by-case approval stance—whether by blocking chip exports abroad or dictating the national security terms under which massive data centers operate. By internalizing hardware design, Big Tech takes on the role of an accidental bureaucrat: it becomes the operational arm of state policy. If you control the custom chip factory of a frontier model, you control what the model can—and, crucially, cannot—do on a global scale.
The subtle irony of this engineering is that the pursuit of private autonomy results in a tightening of the public regulatory noose. Custom-designed silicon promises to free companies from Jensen Huang's abusive margins, but it ties them to an implicit social contract with Washington. AI hardware becomes the new equivalent of 19th-century railroads or 20th-century telecommunications spectrum: you may own the tracks, but the government dictates the tariffs, the routes, and who gets to board. AI has ceased to be just a software product to become a public utility grid, and custom chips are merely the visible brick of this new regulatory wall.
Companies are moving away from Nvidia to escape vendor lock-in and high margins. However, the deeper consequence is that by internalizing hardware design, frontier computing transforms from a free-market commodity into a critical utility that attracts government regulation.
Custom silicon turns Big Tech into an operational arm of state policy. Because controlling frontier chip factories means controlling what AI models can do globally, the government dictates national security terms, transforming AI hardware into a regulated public utility grid.
The idea that any well-capitalized startup can buy its way to the AI frontier ends. The government now acts as a gatekeeper, approving access on a case-by-case basis and dictating who has the right to compute based on national security interests.