Nearly four years after OpenAI’s ChatGPT first launched, one in six people worldwide is now using generative AI tools, according to Microsoft’s 2025 AI Diffusion report. That includes 28 percent of the U.S. working-age population. But there is one area where America risks falling behind — so-called “open-weight” AI systems, where the model is free for the world to use.
Even with a heavy U.S. focus on closed-model AI, such as ChatGPT, investor interest in AI has grown alongside its widespread adoption, with research and advisory firm Gartner projecting a 44 percent year-over-year increase in artificial intelligence investment to about $2.5 trillion. The technology is so significant that it’s difficult to talk about AI’s market share without drifting into geopolitical discussions.
The U.S. and China stand clear above the rest of the world in adoption, capital expenditure, and regulation. As the Trump administration declares in its AI Action Plan, U.S. firms are racing for market dominance. Pursuing “unquestioned and unchallenged global technological dominance,” the White House has turned to protectionist policies, including export controls and state ownership of private firms.
Yet these policies have done less to advance adoption of American models than a simple, time-tested concept — sharing. American dominance in AI will depend on access to American AI models. That means we need more open-weight models.
American frontier models still lead their Chinese counterparts. Analysis by Epoch AI found the best Chinese open-weight models trail leading U.S. releases by roughly seven months, with the top closed AI model, ChatGPT, outpacing the leading open-weight model, DeepSeek, on AI leaderboards.
“Weights” are the specific variables an AI model learns through training. Unlike closed AI models from industry leaders like OpenAI’s GPT-5 and Anthropic’s Claude, the parameters of open-weight or open-AI models are publicly available, meaning anyone can access the model’s weights and run, change, or deploy it as they choose.
With open-weight models, you can achieve 90 percent of the same performance as leading frontier models at nearly 87 percent less cost. For companies like Uber, Airbnb, or DoorDash, which handle client interactions in the millions, switching from American closed AI to Chinese open AI saves tens of millions of dollars annually. Beyond cost savings, open-weight models give companies the control to fine-tune models on proprietary data and host them privately, without being beholden to third-party pricing or policies.
As global demand for accessible AI technology grows, the market has already shown the best path forward for American dominance in a strategy that is already paying off for Beijing. China’s global market share spiked from 3 percent to 13 percent in two months, thanks to a shift to prioritize open-source and open-weight AI models. That growth is a trend, not a blip.
A 2025 OpenRouter study of over 100 trillion tokens of real-world AI usage found Chinese models — led by DeepSeek and Alibaba’s Qwen — accounted for nearly 30 percent of global AI usage, up from roughly 1 percent a year earlier. A lack of private investment led Chinese companies to adopt an open-model approach as a necessity, with models like DeepSeek, Qwen, and Moonshot gaining global traction and performing at a level comparable to U.S. models at a fraction of the cost.
Now, nine of the top 10 open-weight models globally are Chinese. Chinese models have overtaken their U.S. competitors in total downloads with derivative open models ostensibly built by cribbing American technology. Tellingly, if Chinese labs can already replicate the capabilities of closed American models at lower costs, the case for expensive, restricted models fails.
Alibaba’s Qwen is now more popular than Meta’s Llama, which was previously the industry benchmark for open models. Now, startups and businesses around the globe — including Silicon Valley — are choosing to build on cheap Chinese models rather than paying a premium for American subscription-based, closed AI models like GPT or Gemini. American companies like Airbnb are pivoting to Chinese open-weight models, with few American options available. According to Reuters, “around 80 percent of U.S. AI startups now use Chinese open-source AI models.”
The recent deal between American open-weight developer Reflection AI and Korean conglomerate Shinsegae Group is evidence of the demand for American open-weight models and a clear sign of what’s possible. Shinsegae chose Reflection specifically because its open-weight models allow South Korea to “control, audit and evolve” its AI infrastructure on its own terms — a sovereign AI capability that American models can’t offer.
Already, 70 percent of organizations worldwide use generative AI in their operations. As businesses continue to integrate AI into their day-to-day operations, the choice will often come down to cost and flexibility, giving open-weight models an edge. China has rightly recognized this and built its global AI strategy around it.
The American private sector already recognizes that global AI infrastructure will be built on open-weight models. Allowing the open-source community to operate without government interference would preserve America’s position as the leader in the field and ensure U.S. models remain the foundational infrastructure of the global AI economy.
Businesses will choose American open-weight models over their Chinese competitors when American options are available and affordable. We just have to make sure they are.
Tosin Akintola is an assistant editor at Reason Magazine and a writer with Young Voices. His work has appeared in The American Spectator, RealClearPolicy, RealClearHealth, Short Circuit, and more. He’s a graduate of the University of Maryland, Baltimore County.