Firms like OpenAI and Meta claim a compute shortage while also exploring selling compute capacity. This isn't a contradiction but a strategic evolution. They are buying all available supply to secure their own needs and then arbitraging the excess, effectively becoming smaller-scale cloud providers for AI.

George Hotz outlines a contrarian AI infrastructure strategy. Instead of expensive enterprise hardware, Tiny Corp plans to use upcoming consumer AMD GPUs, pair them with extremely cheap power in Oregon (~$0.03/kWh), and sell compute tokens on existing platforms. This low-overhead model aims to undercut traditional cloud providers.

Contrary to the "bubble pop" narrative, a market shift away from high-margin frontier models toward cheaper alternatives could boost overall AI usage. This would redirect revenue from labs like OpenAI to infrastructure players who provide the most efficient, low-cost compute.

The AI compute market has stratified into a pyramid. Hyperscalers serve top frontier labs, forcing NeoClouds and inference platforms to build their own data centers. This trickles down, compelling AI startups to seek GPU capacity from an increasingly fragmented landscape, including providers that repurpose crypto mines.

When power (watts) is the primary constraint for data centers, the total cost of compute becomes secondary. The crucial metric is performance-per-watt. This gives a massive pricing advantage to the most efficient chipmakers, as customers will pay anything for hardware that maximizes output from their limited power budget.

The market undervalues chips from vendors like AMD because their software stack and kernel libraries are less mature than NVIDIA's CUDA. A team with deep expertise in low-level software and kernel optimization can extract significantly more performance from these chips, creating a powerful arbitrage opportunity by buying them at a discount.

China compensates for less powerful domestic AI chips by leveraging cheaper energy. By placing data centers in regions like the Gobi Desert with abundant, low-cost solar power, it can economically operate more hardware to achieve the necessary compute scale, turning an energy advantage into a technological workaround.

Model performance isn't just about architecture; it's also about compute budget. A less sophisticated AI model, if allowed to run for longer or iterate more times, can often match the output of a state-of-the-art model. This suggests access to cheap energy could be a greater advantage than access to the best chips.

China is compensating for its deficit in cutting-edge semiconductors by pursuing an asymmetric strategy. It focuses on massive 'superclusters' of less advanced domestic chips and creating hyper-efficient, open-source AI models. This approach prioritizes widespread, low-cost adoption over chasing the absolute peak of performance like the US.