Leaders at frontier labs like OpenAI and Anthropic indicate that RSI—AI models that self-improve—is closer than anticipated. The arrival of RSI would trigger unprecedented demand for compute, as models consume vast resources to develop and improve themselves autonomously.

Counter-intuitively, as AI models become more efficient, the total consumption of compute resources will rise. This economic principle, Jevons Paradox, states that increased efficiency lowers costs, which in turn unlocks more applications and drives greater overall demand.

As compute costs rise, driven by AI's ability to perform high-value tasks like automating scientific research, many current AI applications will be priced out. AI labs will be willing to outbid consumer use cases to allocate scarce compute for their own R&D, shifting the landscape of viable AI services.

While AI compute demand seems limitless, its price is not infinitely elastic. As inference becomes a core cost of goods sold (COGS) for AI products, excessively high compute prices will break the business models of infrastructure customers, ultimately limiting demand.

Contrary to fears that efficient models will curb computing needs, lower costs will attract more users and enable complex applications, leading to higher overall consumption. This is a classic example of Jevon's paradox, where increased efficiency drives greater demand for a resource.

The current compute crunch isn't just a supply issue. It's because new AI models are so much more capable that they unlock a total addressable market (TAM) of valuable tasks that grows exponentially, far outpacing the linear or geometric growth of compute supply.

The value unlocked by frontier AI models is expanding so rapidly that there isn't enough hardware to meet demand. This scarcity ensures that not just the top lab (like OpenAI), but also second and third-tier competitors, will operate at full capacity with strong margins.

A counterintuitive view of Moore's Law is that for it to hold, the economic value of computation must halve every 18 months because we historically run out of uses for it. The recent rise in H100 GPU rental costs suggests AI is the first application where demand is growing faster than supply, breaking this trend.

The common goal of increasing AI model efficiency could have a paradoxical outcome. If AI performance becomes radically cheaper ("too cheap to meter"), it could devalue the massive investments in compute and data center infrastructure, creating a financial crisis for the very companies that enabled the boom.