In the span of 24 hours in early May 2026, two of the most powerful figures in AI said things that, read together, reveal something important about the industry's cognitive dissonance.

Nvidia CEO Jensen Huang, speaking at the Milken Institute, called AI "the United States' best opportunity to re-industrialize" and said his "greatest concern is that we scare people... to the point where AI is so unpopular in the United States, that they don't actually engage it." He argued that AI creates jobs, that people misunderstand the relationship between task automation and job replacement, and that fear-mongering about AI threatening humanity is counterproductive.

Elon Musk, in federal court testimony related to his lawsuit against OpenAI, acknowledged that xAI trained Grok by "distilling" OpenAI models. When asked directly if xAI trained Grok on OpenAI models, he replied: "Partly." He added that "it was a general practice among AI companies."

Two statements. One bullish on AI's economic potential. One revealing that the industry's competitive dynamics include training on rivals' models without their consent. Both from people with enormous financial stakes in AI's success.

The distillation problem nobody wants to talk about

Distillation — training a new model by prompting existing models and using the outputs as training data — has been discussed in AI research for years. It's how students learn from teachers. It's also, as Musk acknowledged, how companies train competitors' capabilities into their own products.

The practice is controversial for obvious reasons: if Company A trains on Company B's API outputs without permission, is that competitive innovation or something else? The major AI labs have been fighting this, particularly against Chinese companies. But Musk's testimony suggests it's widespread among U.S. companies too.

What makes this notable isn't the ethics of distillation — it's the implication for AI capability claims. If frontier models are built partly through distillation of other frontier models, then the capability gap between top labs is narrower than public claims suggest, and progress may be more iterative than revolutionary. Labs competing on "our model is much better than theirs" may be competing on marketing more than substance.

Jensen Huang's re-industrialization pitch

Huang's framing is worth taking seriously because Nvidia's incentives aren't simple. He sells the most expensive chips in AI. He wants the market to grow. But his re-industrialization argument has a kernel of truth that's independent of his commercial interests: AI does enable new categories of economic activity, not just job displacement.

The history of automation technology suggests that task displacement and job creation aren't zero-sum. ATMs didn't eliminate bank tellers; they made branches cheaper to run, which led to more branches and more tellers per capita for a decade. The internet displaced enormous amounts of existing commerce while creating far more new economic activity.

Huang's point that "the purpose of a job and the task of a job are related but not the same" is genuinely important. AI automating the routine components of a job doesn't eliminate the job — it changes what humans in that role focus on. Radiologists using AI don't stop being radiologists; they focus on cases that require judgment while AI handles the screening work. This is a different model than "AI replaces radiologists."

What the tension reveals

The contradiction between Huang's optimism and Musk's admission of aggressive competitive tactics points to a deeper issue: the AI industry doesn't have a unified theory of what it is.

Is AI a transformative general-purpose technology that will broadly benefit society? Or is it a competitive landscape where the only thing that matters is who builds the best model, regardless of how? Both narratives exist simultaneously, served depending on the audience.

For regulators and the public: AI is a technology with risks that require careful governance. For investors and enterprise buyers: AI is a competitive advantage that's worth billions. For courts: AI companies train on each other's outputs and compete aggressively.

These aren't contradictory; they're the same industry at different moments. But the gap between the public narrative and the actual competitive dynamics is wider than the industry's public figures usually acknowledge.

Huang wants us not to be scared. Musk showed us exactly why people are scared. The truth is somewhere between the pitch deck and the court transcript — and it's more complicated than either.

Sources: TechCrunch, The Verge