The Shallow Moat of Superintelligence

The AI industry is racing forward at breakneck speed, with investors and critics grappling with the impending hyper-profitability of America’s top labs. Valuations of Anthropic and OpenAI have soared to around $1 trillion, rooted in assumptions that may prove increasingly tenuous.

One reason for this optimism lies in the vast potential market size of superintelligent machines. Frontier AI systems promise to revolutionize white-collar industries, slashing costs and boosting performance across sectors. Companies like Airbnb have already demonstrated the transformative power of innovative technologies; a universal tool capable of remaking virtually every industry would be worth exponentially more.

However, this rosy scenario overlooks a fundamental flaw in the business model of America’s AI giants: their crippling expense structure. Developing state-of-the-art AI models requires an outlay of tens of billions on semiconductors, data centers, power plants, and other capital investments. Training a single new Claude model can cost hundreds of millions of dollars, with fine-tuning these systems adding yet another layer of expense.

The notion of a “moat” – where Anthropic’s immense costs serve as a safeguard against competition – is being challenged by recent breakthroughs from Chinese labs. Beijing’s Z.ai has debuted a model nearly as powerful as Claude and ChatGPT’s second-tier systems, while Moonshot unveiled “Kimi K3,” which outperforms all American rivals except the very latest versions of Claude and ChatGPT. Alibaba’s Qwen3.8 Max even outclasses OpenAI’s most advanced systems, trailing only Claude’s Fable in capabilities.

These developments suggest that Anthropic’s moat may be shallower than previously thought. Chinese AI labs’ rapid progress – achieved through aggressive investment and clever engineering – has set off alarm bells within the industry. If cheap, open-source AI becomes a reality, the prospect of vast profits for America’s top labs begins to look increasingly uncertain.

The implications are far-reaching. In a world where new advances can be leapfrogged by cheaper upstarts, hoarding technology – and its profits – will become harder for any one company to do. The notion that building a machine God might be lucrative is being challenged, as the AI industry’s high-stakes competition may ultimately lead to a world of low-margin business.

The sudden emergence of Chinese AI models has exposed vulnerabilities in Anthropic’s business model. America’s top labs had staked their claim on being sole gatekeepers of superintelligent machines, but this vision is now under siege. The implications extend far beyond the industry itself, as policymakers and regulators will need to reassess their approach to regulating a field that may no longer be dominated by a small handful of giants.

As AI development becomes increasingly democratized, the focus shifts from individual companies’ bottom lines to broader societal benefits. The prospect of cheap, open-source AI raises questions about how these systems will be used – and by whom. In a world where anyone can build a state-of-the-art model at a fraction of the cost, what safeguards can ensure that these technologies serve humanity’s interests rather than simply enriching their developers?

The future of AI research is inherently tied to its economic viability. As costs come down and competition increases, labs will need to reevaluate their priorities and strategies. The age of superintelligence may require a new breed of researchers – ones who prioritize collaboration over competitive advantage, open-source innovation over proprietary dominance.

The Chinese AI labs’ rapid progress is not merely a challenge to Anthropic’s profit expectations but also a harbinger for a more inclusive and decentralized future. As the industry converges on a world of low-margin business, companies will need to adapt by developing new revenue streams, exploring novel applications, or pursuing entirely new fields.