Nvidia's chief executive Jensen Huang rejected calls for AI regulation at a recent conference, arguing that artificial intelligence safety falls squarely on the shoulders of individual companies rather than government bodies. Huang framed AI as fundamentally no different from other computing technologies, claiming that hardware and software engineering practices already address the risks inherent in machine learning systems.
The statement positions Nvidia against a growing chorus of AI researchers, policymakers, and international bodies pushing for structured regulatory frameworks. The EU's AI Act, which took effect in 2024, established tiered risk categories and compliance requirements for AI systems. Meanwhile, the U.S. government has issued executive orders and guidance on AI risk management. Huang's position dismisses this regulatory momentum as unnecessary overreach.
His core argument rests on a specific view of what AI actually is. By treating large language models and other AI systems as conventional software with known engineering constraints, Huang suggests existing industry practices suffice. Companies can implement safety measures through model testing, red-teaming, alignment techniques, and deployment controls without external mandates.
This framing conveniently aligns with Nvidia's commercial interests. The company manufactures the GPUs that power most large-scale AI training and deployment globally. Strict regulations could slow adoption rates, impose costly compliance burdens, and give advantage to larger incumbents with resources to navigate bureaucratic requirements. Huang's anti-regulation stance protects Nvidia's growth trajectory in the expanding AI infrastructure market.
Yet the "just engineering" framing glosses over genuine differences between AI systems and traditional software. Machine learning models exhibit emergent behaviors not fully predictable from their training data or parameters. Their outputs can perpetuate bias, spread misinformation, or cause societal harm at scale. The inability to fully interpret how neural networks arrive at specific decisions creates opacity that standard software testing cannot entirely resolve. Foundation models trained on internet-scale data introduce risks around cultural toxicity and hallucinated outputs that differ in kind from conventional software failures.
Huang's characterization of AI as merely "hardware and software" also ignores the concentration of capability among a handful of companies. Nvidia itself holds approximately 80 percent of the AI chip market. A handful of labs control access to state-of-the-art foundation models. This concentrated power structure creates asymmetries that market-driven safety measures alone may not address. A small competitor cannot easily implement safety measures comparable to those available to well-resourced incumbents.
The "leave it to industry" approach has historical precedent in tech. Self-regulation failed to prevent privacy breaches, data misuse, and platform harms in social media and other sectors. Government intervention eventually arrived after years of documented problems. Huang's position assumes Nvidia and peer companies will consistently prioritize safety over competitive advantage and profit margins, an assumption not supported by decades of tech industry behavior.
What matters most is whether safety measures actually work. Internal company testing and voluntary standards can catch certain classes of risks, but they operate without external auditing, public transparency, or enforcement mechanisms. Government regulation introduces accountability mechanisms, standardized testing protocols, and consequences for noncompliance. Neither approach alone suffices. A hybrid model combining industry expertise with external oversight has proven more effective in domains ranging from pharmaceuticals to aviation safety.