The Real AI Race: It's About Diffusion
US vs. China: Who Wins the Critical AI Diffusion Battle?
When comparing the United States and China in artificial intelligence, the spotlight is on the development of foundation models. At first glance, America maintains a comfortable numerical lead, producing 40 notable models in 2024 compared to China's 15. However, looking deeper reveals a more complex story: China's best models, particularly from DeepSeek, Alibaba (Qwen), and Tencent, have dramatically closed the performance gap on key benchmarks, achieving near parity in capabilities despite using fewer computational resources. Notably, Chinese firms have chosen open-weights strategies, making advanced models like DeepSeek freely accessible, while their American counterparts generally restrict top-tier models behind subscriptions or enterprise licenses.
Enjoy what you read? Support us & unlock more! 🤝
However, focusing solely on foundation models risks overlooking a perhaps more critical dimension of the competition: the diffusion of AI capabilities throughout the broader economy. By diffusion, I mean the speed and extent to which AI technologies move beyond elite labs and are adopted by businesses across diverse sectors – the very teams building practical applications – to enhance productivity and drive innovation. Understanding this is crucial because widespread, effective adoption, not just peak model performance, often determines long-term economic impact and competitive advantage.
China's open-weights strategy, contrasting sharply with the US approach, could potentially accelerate this diffusion, creating a different competitive dynamic. In this article, I explore how effectively AI technologies are spreading through enterprises in both countries, examining patterns and implications that may ultimately prove more decisive than headline-grabbing foundation model races.
The Engines of AI Adoption
But what precisely fuels this diffusion? It isn't simply a matter of possessing cutting-edge algorithms or vast computational resources. Rather, the successful integration of AI across an economy hinges on a confluence of enabling factors – the underlying infrastructure, talent pipelines, and market conditions that allow businesses and individuals to readily adopt, adapt, and benefit from these powerful tools. For teams building real-world applications, recognizing these catalysts is crucial for understanding the true pace and potential of AI's economic transformation. Several key drivers consistently emerge as critical:
Why the US May Win the AI Adoption Race
While China's rapid advancements and open-weights strategy capture significant attention, the United States possesses distinct advantages that may foster faster and deeper diffusion of AI capabilities throughout its economy. From my perspective, America's edge lies less in centrally directed initiatives and more in its inherently decentralised, market-driven ecosystem. Here, the impetus for adopting AI often arises organically from individual firms seeking competitive advantages and tangible returns on investment, rather than from top-down mandates. This bottom-up dynamic, while perhaps less dramatic than state-led initiatives, could prove more effective at embedding AI deeply within the diverse workflows and specific needs of businesses across myriad sectors.
China's Diffusion Advantage: Beyond the Frontier Models
While the US possesses significant advantages that could fuel AI adoption, and I certainly hope diffusion accelerates rapidly there, my sense, reinforced by observations over the years, is that the pace of practical AI implementation might currently be faster in China. Having chaired numerous AI and data conferences in China some years ago, I witnessed firsthand the intense competitive pressure driving companies to execute efficiently and rapidly – a dynamic that seems particularly suited to the current AI deployment race.
Nowhere is that momentum clearer than in healthcare. DeepSeek’s open‑weight models, for example, were integrated into many hospitals in China within weeks of their January release, underpinning everything from pathology‑slide triage to automatic discharge summaries—use cases that in the United States would still be winding through pilots and privacy reviews. Broad surveys tell a similar story: IBM’s Global AI Adoption Index found that half of large Chinese enterprises already run AI in production, versus roughly a third of their U.S. counterparts. In short, America may hold the frontier‑model high ground, but China is proving notably faster at turning those capabilities into everyday tools—an edge that matters to those of us focused on real‑world implementation.
Several key factors appear to be accelerating AI diffusion within China, creating a potentially more fertile ground for rapid, widespread adoption compared to the US:
Integrated Digital Foundation. Decades of building a pervasive digital infrastructure, exemplified by ubiquitous platforms like WeChat and Alipay, provide pre-existing rails upon which AI functionalities can be rapidly deployed and scaled across consumer and industrial applications.
Cost Dynamics & Open Access. Aggressive price competition among domestic cloud providers, coupled with a strategic embrace of open-weight models by leading firms like DeepSeek and Alibaba (Qwen), significantly lowers the cost and technical barriers for businesses experimenting with and implementing advanced AI.
Pragmatic Application Focus. There's a discernible emphasis on applying AI to solve concrete, sector-specific problems – particularly in manufacturing, healthcare, and public administration – rather than solely pursuing general intelligence, leading to faster real-world impact.
State Encouragement & Public Sector Pull. Significant government adoption, particularly at local levels for administrative tasks, creates substantial demand, signals confidence, and encourages broader private sector uptake, acting as a powerful catalyst.
Conducive Regulatory Environment (for now). While evolving, the current regulatory landscape, particularly concerning data access and initial deployment hurdles, appears less immediately restrictive than in the US, allowing for faster iteration and implementation cycles. China's developers appear to have largely unfettered access to data, including copyrighted material, without consistently respecting international intellectual property regimes. This provides a significant advantage in training (and fine tuning) data volume and diversity.
High Consumer Receptivity. Chinese consumers generally exhibit greater enthusiasm for, and familiarity with, AI-powered products and services, fostering a market more readily accepting of new AI integrations.
From Competition to Coordination? Navigating the AI Superpower Divide
While it’s tempting to frame the AI race purely through the lens of foundational models, I suspect we’re only at the very start of a more consequential contest: the widespread adoption of these technologies in everyday business practices. China's strategy of releasing capable, frequently updated open-weight models is already influencing how teams, including some in the US, adopt these tools. Attempts by Washington to throttle China's progress, notably via export controls on key hardware, increasingly look porous, unable to prevent China's steady march towards broader global AI influence.
Unlock Exclusive Content: Upgrade Your Subscription! 🔑
This technological diffusion unfolds against a backdrop of rising US-China friction, complicated by AI's undeniable military applications. As we edge towards more powerful systems, perhaps even AGI, the stakes – both competitive and existential – escalate sharply. Navigating this means walking a tightrope: pursuing national interests while recognizing the urgent need for shared guardrails, particularly around safety and responsible deployment. My hope is that the sheer scale of the risks and potential rewards compels a move towards pragmatic coordination on standards and protocols, steering this transformative technology away from becoming purely a tool in a zero-sum geopolitical contest.
Ben Lorica edits the Gradient Flow newsletter. He helps organize the AI Conference, the AI Agent Conference, the NLP Summit, Ray Summit, and the Data+AI Summit. He is the host of the Data Exchange podcast. You can follow him on Linkedin, Mastodon, Reddit, Bluesky, YouTube, or TikTok. This newsletter is produced by Gradient Flow.