Topic
The economics and build-out of AI computing: GPU supply and rental pricing, data centers, power constraints, and the cost of training and serving models.
Exponential View puts annualized AI revenue at $276bn, enough to cover 182% of the yearly cost of data-center capital already in service. The tally finds today's build covered and the hurdle rising with every dollar committed from here.
Reality
Goldman Sachs says AI labs renting SpaceX's GPU compute pay rents well above the revenue the industry needs to turn a profit. The same bank underwrote SpaceX's June listing and projects its AI revenue rising about 100-fold by 2030.
Reality
Silicon Data has raised $35.2m since April 2024 and says its own venue has traded hundreds of millions in notional. CME Group intends to reference its benchmarks for cash-settled GPU futures this fall.
Reality
Vinod Bijlani argues in Forbes that a token price measures only what an application consumed, while the idle accelerators and rising power draw behind it never reach the invoice a CIO reviews. He works for HPE.
Reality
GPU rental prices fell more than 60% in 2025. SemiAnalysis data says H100s have since climbed about 40% to $2.35 per GPU-hour, with 12 to 18 month waits for clusters.
Reality
Muse Glimmer ships as Apache 2.0 weights sized for a 24GB card. Muse Spark 1.2 stays on Muse Code and the Meta Model API. Plan capacity for two tiers, not one.
Perspective Coverage
6 publishersReality