When Peter Navarro recently claimed that “Americans are paying for India and China to use ChatGPT,” he revealed a fundamental misunderstanding of how artificial intelligence evolves, how value is created in large language models, and what global adoption actually means for U.S. technological leadership.
Navarro’s framing reflects a 20th century industrial mindset, one in which physical inputs and outputs define trade but AI is not steel, automobiles, or energy commodities. Modern AI systems derive their power and long-term value from scale, diversity of use, and ubiquity to further evolve models, not simply from where the electricity bill is paid
Navarro’s argument assumes that every query from India or China represents a cost borne by Americans. What he fails to grasp is how that usage itself is an asset. Even when user interactions may not directly retrain a model in real time, global usage generates indispensable intelligence about performance, bias, safety, and relevance. Large language models are not static products; they improve through exposure to varied language structures, cultural contexts, and problem-solving approaches and behaviors. This is known as the Data Network Effect. Treating that exposure as a liability rather than a strength is not consistent with the economics of AI platforms.
The scale of AI adoption in India makes this hard to miss. According to recent data, India has become the largest market for ChatGPT globally, with more than 65 million daily active users and roughly 145 million monthly users, accounting for about 16 percent of the platform’s global user base. Year-on-year growth in active users has exceeded 600 percent, driven largely by free or low-cost access to tools such as ChatGPT, Gemini, and Perplexity.
Sreedhar Potarazu: What it will take for doctors to transform US healthcare in a widening gap (
This surge is not evidence of Americans subsidizing Indian users; it is evidence of extraordinary demand and engagement in one of the world’s most culturally diverse populations. That diversity matters because it exposes models to things called edge cases and failure modes (exceptions) that would never surface in a narrow, high-income, English-only user base.
In order for LLM’s to be of value that can’t be developed on the behaviors of one race, language and culture to be of value to eight billion people.
This is especially important in an era where bias mitigation and safety are central concerns in generative AI. Broader global usage allows developers to observe how systems behave across languages, educational levels, and cultural norms, making it easier to identify blind spots and reduce bias. In this sense, global adoption does not dilute value; it enhances it.
Navarro’s argument also ignores how every major technology platform has historically achieved dominance. Google, Meta, and other digital giants did not become global leaders by restricting access or charging aggressively at the outset. They prioritized ubiquity, knowing that scale and engagement would eventually translate into monetization. AI companies are following the same trajectory. OpenAI has already indicated that advertising and other revenue models are likely in the future. The notion that these firms are naïvely giving away value indefinitely is shortsighted. Free or subsidized access initially is a strategic investment in platform dominance to enhance algorithms — not a subsidy paid by American taxpayers.
Sreedhar Potarazu: The dilemma of destiny as our own prisoners: What MLK would tell us (
From a geopolitical perspective, Navarro’s concern is also misplaced. The United States continues to hold a decisive advantage over China in access to advanced AI chips and high-end compute infrastructure. Export controls, dominance in cutting-edge semiconductor design, and control over critical parts of the supply chain have constrained China’s ability to scale frontier AI models at the highest level. India, despite its massive talent pool and explosive AI adoption, remains a bottleneck in advanced chip manufacturing and compute capacity. These realities, not where a chatbot user happens to be located, define the balance of power in the AI race.
India’s growing influence in AI further underscores why Navarro’s framing is backward. India is set to host the Global AI Impact Summit, bringing together governments, industry leaders, and researchers to shape conversations around AI governance, inclusion, and innovation. This is not incidental. It reflects the fact that countries with large, engaged user bases will play an increasingly important role in determining how AI systems are evaluated, regulated, and trusted worldwide. Excluding or marginalizing those users would weaken, not strengthen, American leadership.
If the United States wants to remain at the forefront of artificial intelligence, it must abandon the idea that global usage is a cost to be minimized. The real strategic objective should be to build models that are globally relevant, resilient to bias, and efficient to scale. In the AI era, the platforms that win are those that are used everywhere, tested by everyone, and refined through the broadest possible engagement.
Navarro’s claim may resonate with a familiar protectionist instinct, but it is badly suited to the realities of digital platforms and generative AI. Americans are not “paying” for the world to use AI. They are helping to build the most widely adopted, most rigorously tested, and ultimately most valuable AI systems for global utility.