After years of explosive growth, experts are now pointing to potential cracks in AI’s foundation.
Astronomical industry valuations, vast levels of investment and a promise of exponential progress have driven the AI sector to the heart of the US economy and helped send the stock prices of technology companies soaring.
But in recent months, voices from within the technology and finance sectors have suggested that we could currently be in an AI bubble.
OpenAI’s CEO has warned that currently ‘investors as a whole are overexcited about AI’ and the head of Alphabet recently acknowledged ‘elements of irrationality’ in the sector, while both backing AI’s long-term transformative potential. The Bank of England warned of a ‘sharp correction’ in the value of major tech companies, with the IMF comparing today’s valuation levels to the bullishness over internet companies that resulted in the 2000 dot-com crash.
These concerns signal a fundamental shift in the AI landscape, one with potentially profound implications for global technology competition and AI adoption. To understand these potential implications, we need to think beyond market correction and consider how an AI bubble burst could reshape the industry as players shift positions and adopt new strategies.
The core issue
Many still back AI’s exponential potential. Investment has continued to pour in and vast amounts are being spent on tangible AI infrastructure, such as cloud storage and semiconductor chips. Nvidia CEO Jensen Huang said he sees something ‘very different’ to an AI bubble. Besides private sector players, senior government officials working on AI have also expressed belief in the exponential potential of AI capabilities to the author.
But bubbles are not uncommon in the technology sector, from the 2000 dot-com crash to the metaverse hype.
Current speculation may stem from two intersecting pressures on AI: the faltering promise of constant improvement and an unclear route to profitability.
Large language models (LLMs) are hitting a performance plateau – with diminishing returns in capability improvements despite increasing training costs. If they continue to plateau, it would contradict the fundamental premise that justifies some of the copious investment and trillion-dollar valuations: that AI capabilities will continue improving at an exponential rate.
Hundreds of millions of people enjoy LLMs such as ChatGPT, but far fewer have reportedly subscribed monthly to an advanced version. This encapsulates the challenge many generative AI companies face: how to make profit.
Given this core issue, venture funders could lose faith and consumer-facing AI startups could contract as investors scrutinize pitches, demanding clearer paths to profit. Other funders are turning to cost-effective open-source alternatives. Regardless of the numerous technical breakthroughs in the past few years – and there have been many – proprietary generative AI is not yet financially sustainable.
Potential impact on US AI
The bursting of a potential AI bubble could have severe consequences for the US economy and financial stability beyond. But a collapse would also likely reshape the AI sector itself.
So, what could this scenario look like?
In the US, companies would potentially face a severe financial squeeze, forcing a strategic pivot across the technology sector away from uncertain consumer markets. This would likely see the rebalancing of corporate finance strategy: AI companies may seek to downsize while increasing the price of their products. Aside from layoffs for highly skilled employees, this may also result in a technology that is too expensive to significantly scale across society.
A second potential outcome would be AI companies increasingly pursuing defence and national security contracts over civilian application.
In summer, the Pentagon awarded contracts to Google, xAI, Anthropic and OpenAI for the expansion of military AI applications. A securitization shift – from civilian chatbots and image generators to autonomous weapons systems and intelligence analysis tools – would represent both an economic survival strategy and an acknowledgment of where reliable government spending lies.
How China could capitalize
With interdependence and global supply chains, an AI bubble bursting will be felt worldwide. But China is in a unique position to endure – and potentially capitalize.
Though not watertight from shocks, from the outset China has adopted a different AI approach that has created an environment that is more insulated from potential financial turbulence experienced by Western competitors.
China’s technology ecosystem benefits from diverse funding through state investment, subsidies and private capital. Innovation has spread beyond big firms to universities, SMEs and hedge funds. Rather than promising revolutionary breakthroughs, these entities emphasize incremental improvements and practical applications. Crucially, they have favored smaller, adoptable and cheaper open-source models.
In this scenario, if Western companies retreated to premium markets and government contracts, they would potentially cede vast portions of the global market for AI. Meanwhile, China would be better positioned to pursue an aggressive strategy of cost-effectiveness and market penetration, capturing developing economies and sectors underserved by Western providers with cheaper Chinese alternatives. This precedent has already been set with mobile phones, internet connectivity, solar power and electric vehicles.
China’s low pricing strategy is particularly effective when coupled with aggressive integration into critical infrastructure projects, like the Digital Silk Road. In the Global South, diminishing international assistance means the appetite for budget-friendly AI integration will likely increase, with more countries opting for China’s cheaper offering. Given that switching costs multiply with each layer of integration, dependencies on China could extend beyond immediate AI to all layers of AI infrastructure.
The result: Chinese AI systems would become the standard across much of the world. This would have implications for sovereignty and security that extend far into the future – including potential espionage and data harvesting, hidden backdoors, and kill-switches.
Policy response
President Donald Trump recently made the striking announcement that the US would allow Nvidia’s advanced H200 chips to be exported to ‘approved customers’ in China. However, this policy stance could easily and rapidly change. In the scenario of an AI bubble bursting, the US may seek to impose new export controls on advanced semiconductors to attempt to reduce rival technological advantages.
But such measures are blunt instruments with significant limitations in enforcement due to black market trading and dual use technologies. Additionally, restricting technology exports can inadvertently incentivize competitors to pursue their own innovative R&D where restrictions apply.