MILAN — The ongoing AI revolution hinges not only on the development of increasingly powerful models but also on the effective adaptation and adoption of these models across various sectors of the economy. This adaptation is crucial for reducing the cost of existing products and services while also creating new or improved offerings that can drive economic and social progress. While model development is primarily concentrated in the United States and China, the diffusion of AI technology must occur on a global scale to realize its full potential.
AI development is expected to follow a J-curve pattern. Initially, there is substantial investment in areas such as infrastructure, software, business-model adaptation, data consolidation, and human-capital development. However, these investments may not yield immediate benefits, leading to downward pressure on productivity, which includes benefits not captured by conventional national income accounts. Eventually, as the technology’s value-creation potential materializes, the curve will begin to slope upwards. Currently, it is unclear what this upswing will entail, as it remains an unknown how high or steep the curve will be. Investors generally seem to be banking on significant returns, although uncertainty lingers, with some experts cautioning against potential overestimations that could result in a bust. The outcome will largely depend on how effectively diffusion occurs.
So far, AI diffusion has been uneven across sectors. Industries such as technology, finance, and professional services have rapidly embraced the technology, while larger employment sectors like health care and construction have been slower to adapt. If these disparities persist, the J-curve may flatten, leading to muted returns on investments and delayed growth and productivity gains. This raises questions about whether AI is in a bubble, with the eventual determination resting on the pattern and pace of diffusion over the coming years.
Several channels facilitate AI diffusion, with software-as-a-service (SaaS) providers being among the fastest. Companies like Google, Microsoft, Notion, Salesforce, and Adobe are embedding AI into their offerings. Moreover, AI can be integrated into scientific processes relatively swiftly. With major developers of language models and multimodal models offering application programming interfaces (APIs), tailored AI models can be created more efficiently, potentially accelerating progress in various fields.
Open-source models, more prevalent in China than in the U.S., present further opportunities for specialization and competition, especially for smaller firms and countries that may lack the extensive computing infrastructure necessary for the largest AI models. However, barriers to entry still exist, including reliable electricity, robust computing capacity, and accessible mobile-internet connectivity—key prerequisites for broad adoption.
Trade dynamics, particularly regarding advanced semiconductors, significantly influence AI adoption. Additionally, the availability of human capital—from advanced AI engineering to user-related skills—plays a crucial role. An economy must ensure access to diverse capabilities through education, reskilling, and labor mobility. Data accessibility is another essential component; fragmented, incomplete, or inaccurate data systems can severely hinder the training of effective AI models.
While the private sector is pivotal in driving AI diffusion, policy frameworks and regulatory structures also play a critical role. China’s government, for instance, has adopted a pragmatic approach to leverage AI for real-world development and economic challenges. Huawei founder Ren Zhengfei noted that while developing advanced models is a priority, broad AI deployment is equally essential for achieving rapid gains in service quality, efficiency, and productivity, particularly in the context of an aging population.
China’s government actively directs innovators towards these objectives, encouraging major tech platforms to build open-source models and develop applications in specific sectors such as autonomous driving, health care, robotics, supply-chain management, and green technologies. This policy engagement has yielded results; China accounts for over 30 percent of global manufacturing output and, in 2024, is projected to represent 54 percent of all robot installations worldwide. The country now houses nearly half of the world’s installed robots, exceeding 2 million.
In contrast, the U.S. policy framework is less engaged, allowing tech giants and well-funded AI startups to focus on pushing the boundaries of large models, often aiming for artificial general intelligence and superintelligence. While diffusion channels are open, the responsibility for their utilization primarily falls on the private sector. This approach may work in tech-centric sectors with the resources for experimentation but is unlikely to address the broader issues inhibiting AI adoption in areas like data fragmentation, capacity limitations, regulatory hurdles, and scalability challenges.
The result could be a two-speed diffusion pattern, leading to sluggish economic growth, adverse distributional outcomes, and weakened economic foundations for national security. Recognizing the importance of state guidance, the U.S. government has long acknowledged the necessity of ensuring that private-sector innovation aligns with public objectives. A similar hybrid, active, and pragmatic approach is needed for AI diffusion across various sectors. Without it, the risks of subpar economic growth and negative social outcomes will likely escalate.
In the realm of AI diffusion, a passive approach of watching and waiting is not a viable strategy.
Michael Spence, a Nobel laureate in economics, is an emeritus professor of economics and a former dean of the Graduate School of Business at Stanford University and a co-author of “Permacrisis: A Plan to Fix a Fractured World” (Simon & Schuster, 2023).
Copyright: Project Syndicate, 2025.
www.project-syndicate.org
See also
Spot AI-Generated Images on Social Media: Key Signs and Verification Tools
Tech Firms Embrace Custom Generative AI Models to Enhance Data Security and Drive Innovation
AI Models Improve Rapidly in 2025, Raising Concerns Over Copyright and Artistic Integrity
iMini AI Launches Precise Edit Feature, Transforming AI Image Generation for All Users
AI Video Generation Evolves: Image-First Workflows Enhance Creative Control and Quality
