How to Make AI Not a Bubble
A Tale of Two Strategies
In a recent issue, The Atlantic posed a question that is quietly reverberating through the tech world: "Are we in an AI bubble?" This is not mere media speculation; it reflects a growing tension between the frenzied investment in artificial intelligence and its tangible, real-world impact.
The most potent symbol of this potential disconnect is a financial manoeuvre by Oracle. The tech giant recently booked a reported $300 billion, five-year contract with OpenAI as "Remaining Performance Obligations" (RPO). In accounting terms, this allowed Oracle to present a highly uncertain future revenue stream as a near-guaranteed backlog of business. The market reacted with euphoria: this single data point catalyzed a historic surge in Oracle's stock, briefly crowning its founder as the world's richest person.
This event perfectly encapsulates the "detachment from the real economy" that defines the American AI narrative. A speculative bet on the future of a single, non-public client was transformed, through financial engineering, into present-day market certainty. It reveals a system where narrative and accounting can temporarily supersede the fundamentals of value creation. The Oracle-OpenAI deal, therefore, serves as the critical starting point for understanding a core risk: that the American AI boom may be built on a foundation of virtual-economy finance rather than real-economy substance.
The Virtual World's Dead End
This detachment is not just financial; it's systemic. The American AI industry currently operates within what can best be described as a capital vortex. Venture capital pours billions into AI startups, which then spend a significant portion of that funding on computing infrastructure from a handful of tech giants like Nvidia and Oracle. The soaring revenues of these infrastructure providers are then cited as proof of the AI boom's viability, justifying another cycle of investment into the startups. It's a closed loop, a self-sustaining cycle of capital churning within the virtual economy, largely untethered from the creation of value in the physical world.
This loop might be sustainable if AI were generating revolutionary productivity gains within its native digital domain. Yet, the evidence here is also shaky. As The Atlantic's report highlights, even in coding—arguably AI's most mature application—the promise of massive productivity boosts is running into a harsh reality. A rigorous study found that experienced developers using AI tools were actually slower than those without. This suggests an imagination bottleneck: when AI's applications are confined to the virtual world of generating code, text, and images, its capacity to create transformative value is surprisingly limited.
The fundamental reason for this virtual-world confinement is structural. The disconnect is starkly illustrated by two figures: while the academic world produces AI research at a blistering pace, with more than three preprint papers submitted to arXiv every hour, U.S. manufacturing companies invest on average a mere 0.1% of their revenue in the field. The U.S. has experienced a decades-long decline in its manufacturing base, leaving a deficit of complex, real-world industrial scenarios where AI could be applied. Without a robust physical economy to engage with—factories to optimize, supply chains to manage, physical assets to maintain—AI development is inevitably pushed towards the digital realm where it risks becoming a solution in search of a problem. The technology is forced to "involute" within the virtual economy because the fertile ground of the real economy has eroded.
The Real Economy's Imagination
In contrast, China is charting a different course. Its "AI+" strategy, formally initiated by the State Council's "Guo Fa" (State Council Issuance) [2025] No. 11 directive, is not about funding a standalone AI sector but about systematically embedding AI as a foundational tool across the entire real economy. This top-down vision was quickly followed by concrete action plans from key ministries, such as the Ministry of Industry and Information Technology (MIIT), which immediately began work to expand "AI+" application scenarios. This approach unlocks a far broader horizon of imagination.
Consider the energy sector, where the National Development and Reform Commission (NDRC) and the National Energy Administration (NEA) jointly released detailed implementation guidelines for "AI + Energy." Here, AI is not generating programming scripts but re-engineering the national power grid. It is being deployed to create hyper-precise forecasts for wind and solar output, enabling intelligent dispatch that balances supply and demand in real-time. In the coal industry, the plan calls for AI to facilitate the remote, unmanned operation of mining equipment, dramatically improving safety while boosting efficiency. This is AI as a tool for national energy security and transition, a task of immense physical-world complexity.
The vision extends to the maritime economy. In Zhejiang, a major maritime province, the Provincial Department of Ocean Economy Development recently held a symposium to accelerate its "AI+ Ocean" development, guided by a three-year action plan. With a focus on "small-slice, big-impact" applications, the plan specifically targets using AI models to monitor fishing vessels for safety compliance and to issue early warnings for risky behaviour. In aquaculture, AI is slated to help create optimal breeding conditions for the large yellow croaker, a key local product. This is AI moving from the server rack to the fishing trawler and the fish farm, enhancing traditional industries and improving livelihoods.
This imaginative application of AI is also reshaping transportation and new energy formats. The development of intelligent connected vehicles, a priority in pilot zones like Chongqing and Changsha-Zhuzhou-Xiangtan, relies on AI to process vast amounts of real-world sensor data for autonomous driving and traffic management. In the power sector, national energy policy explicitly promotes AI as the brain behind "virtual power plants," which aggregate distributed energy resources like rooftop solar and EV batteries to act as a unified, flexible grid resource. These applications are not confined to a screen; they are the building blocks of smart cities and future energy networks.
What these diverse cases share is a common focus: they apply AI to solve core, tangible problems in the physical world. They address genuine pain points in national industries and people's lives. This is the true blue ocean for AI's potential, creating value that vastly exceeds what can be generated by circling within the virtual economy alone.
The Deeper Engine: Data as a Factor of Production
Underpinning this entire strategy is a deeper, often overlooked institutional innovation: the market-oriented allocation of data as a new factor of production. A recent State Council directive ("Guo Han," or State Council Letter) [2025] No. 86 approved comprehensive pilot reforms in 10 key regions across the country—from the Beijing Urban Sub-center to the nine mainland cities of the Greater Bay Area—with the explicit goal of building a market system for data.
The most fundamental measure, seen across all pilot zones, is the push for open sharing of public data. This involves breaking down traditional "data silos" between government departments. For instance, Beijing's plan calls for leveraging its Municipal Public Data Open Platform to release high-value datasets. Similarly, Guangdong's initiative for the Greater Bay Area supports a provincial-municipal integrated "one network for sharing" platform. This is the essential first step: making the raw material available.
But the reforms go further, aiming to cultivate a market for data circulation and trading. The plan for Zhejiang's Hangzhou-Ningbo-Wenzhou cluster involves conducting big data transactions through existing venues. Chongqing's pilot goes a step further, focusing on creating a comprehensive trading system complete with "data brokers" and other intermediary services. In Anhui, the provincial data exchange is designated as a central hub for this activity. These policies are designed to turn data from a static asset into a dynamic, tradable resource.
The most forward-thinking aspect of these pilots is the exploration of data assetization. This involves treating data as a tangible, bankable asset, much like land or capital. Henan's pilot in Zhengzhou, for instance, is tasked with establishing systems for public data asset registration and valuation. This goal is echoed in Beijing's plan to develop statistical accounting methods for data as a factor of production, a concept that has been elevated to the national level since the 4th Plenary Session of the 19th Central Committee of the CCP. This is the crucial final piece of the puzzle: formally recognizing data's economic value on a company's balance sheet. Crucially, this model ensures that corporate value is not manufactured through financial market manoeuvres, but is instead a reflection of genuine value created by first empowering the socio-economy and then assetizing the resulting data.
This creates a powerful, virtuous cycle. The "AI+" strategy's push into the real economy generates immense demand for high-quality, real-world data. In turn, the data-factor market reforms create the supply, incentivizing industries to unlock and productize their dormant data assets. AI empowers the real economy; the real economy produces vast amounts of data; this data, now a marketized factor, provides the essential fuel to train more powerful and relevant AI models, which then better empower the real economy.
Conclusion
In the end, AI is just a tool. It is neither inherently magical nor malevolent. Whether a technological wave culminates in a speculative bubble or a productivity boom depends less on the technology itself and more on how it is wielded. The crucial variable is policy guidance. A country can choose to let capital chase short-term returns in the virtual economy, or it can steer that same technology towards solving fundamental problems in the real world. These two paths determine whether AI becomes a magnificent bubble or a powerful, durable engine of progress.