AI compute is centralized & expensive.
A handful of hyperscalers control the vast majority of GPU supply, and they price it opaquely. Spot rates swing by the hour, Reserved Instance discounts lock developers into multi-year commitments, and the underlying cost of running inference is impossible to audit. The result is a market where the supplier always wins and the builder always pays.
For teams shipping AI products, this is more than an annoyance — it is structural leverage working against them. A single provider can deprecate an instance class, throttle throughput, or hike pricing with thirty days notice. There is no fallback, no transparency, and no recourse. Builders absorb the risk and pass the cost to users.
Worse, idle GPU capacity sits everywhere — in gaming rigs, in research labs, in data centers off-peak — but it cannot reach the market. The plumbing to discover, price, and settle compute jobs between strangers simply does not exist in the centralized stack.