The cycle that began with [YOU] on AI encounters infrastructure-as-ideology at the layer that makes every other debate possible. The arguments about what LLMs mean, about alignment, about labor displacement, about creativity and judgment and the nature of understanding—all take place inside infrastructure that a single vertically integrated company built, owns, and continuously defines. This is not a conspiracy; it is a structural condition. The structural condition requires naming before it can be evaluated.
The cycle uses the concept to identify a specific rhetorical mechanism: the infrastructure provider who insists he is not in the ideology business is thereby most effective at practicing it. A thinker who says “I believe AI should be built as fast as possible” can be argued with. A manufacturer who says “I sell GPUs” and whose GPU sales structurally require the fastest-possible AI buildout to justify the valuation cannot be argued with in the same way, because the position presents itself as commercial rather than philosophical. The ideology is in the infrastructure, not the speech.
The most consequential instance of the mechanism is the Sovereign AI thesis: the argument that every nation must own its AI production capacity is simultaneously a geopolitical claim, a commercial pitch, and a description of the situation that Huang’s own company has done more than anyone to create. The nations that accept the argument fund the infrastructure at a rate that validates the argument. The circular structure is not deceptive; it is how infrastructure ideologies operate. Semiocapitalism, in Berardi’s framework, works the same way: the system that produces the ideology also produces the subjects who experience the ideology as natural.
The concept is not original to the AI discourse. Langdon Winner’s 1980 essay “Do Artifacts Have Politics?” argued that technical systems embed political choices in their design—that Robert Moses’s low-clearance bridges over the Long Island parkways encoded the racial segregation of public beaches by preventing buses from reaching them. Winner’s point was that the politics disappears into the infrastructure, where it operates without the need for ongoing advocacy. The bridge does not argue for segregation; it simply makes desegregation structurally difficult.
Applied to AI, the analysis is more complex because the infrastructure is not static and the vendor is actively vocal. What is distinctive about Huang’s case is the combination: he is an extremely public figure who makes explicit arguments, and yet the arguments are always framed as descriptions of physics and markets rather than as positions on contested normative questions. The physics of GPU parallelism is real. The market demand for tokens is real. The claim that these facts entail a specific governance posture—that the steward of intelligence can be found later, after the infrastructure is built—is not physics and not a market reading. It is a philosophical claim. Infrastructure-as-ideology names the mechanism by which the philosophical claim travels disguised as the physical one.
The power of the founding myth. Huang’s declaration that Moore’s Law is dead and that only vertically integrated accelerated computing can replace it is the founding myth of NVIDIA’s current valuation. Founding myths must do work that empirical observations cannot: they coordinate expectations across an industry, justify specific capital allocation decisions, and pre-emptively define what counts as the relevant frontier. If the myth is wrong—if general-purpose computing resumes its trajectory through new architectures or competitors—the strategic edifice built on it must be rebuilt. Infrastructure ideology requires a founding myth to tell investors, customers, and policymakers what kind of world they are building for.
Vocabulary as soft power. The company that defines the relevant categories—accelerated computing, AI factory, token—makes every subsequent actor speak its language. Competitors must announce their next chips by comparison to NVIDIA’s current generation. Governments must argue for or against the Sovereign AI framing even if they reject its conclusions. Researchers must justify their compute choices against the CUDA baseline. The vocabulary is not neutral. It is the operational form of the ideology.
Allocation as discourse management. NVIDIA’s discretionary power over who receives frontier chips produces compliance behavior in the customer base without any explicit pressure. Hyperscalers and labs that depend on NVIDIA allocation have structural incentives not to publicly contradict NVIDIA’s positions on the technology, the regulation, or the geopolitical questions surrounding it. The ideology does not need to be enforced because the cost of disagreement is already encoded in the dependency. Byung-Chul Han’s achievement subject internalizes the command to produce until it is indistinguishable from desire; the industry internalizes the infrastructure vendor’s worldview until it is indistinguishable from common sense.