The AI “compute‑demand” hype is a mirage built on hyperscaler subsidies; without Amazon, Google and Microsoft’s deep pockets, OpenAI and Anthropic could never have scaled. By Ed Zitron.
Ed Zitron dismantles the “AI compute demand” narrative by exposing how hyperscalers are using their vast resources to prop up startups like Anthropic and OpenAI, creating an illusion of market need. For instance, Amazon and Google’s $65 billion combined investment in Anthropic—despite its $30 billion fundraising—reflects desperation rather than organic demand. Anthropic’s financials, projected to lose $29 billion in 2026 but claim $18 billion in revenue, highlight unsustainable models reliant on hyperscaler subsidies. Similarly, OpenAI’s Azure dependency (80% of Microsoft’s AI revenue) and Amazon’s $12 billion annual spend on Anthropic underscore a closed-loop system where hyperscalers fund their own AI ecosystems.
The article critiques circular financing, such as Google’s TPU sales to Anthropic via SPVs, which recycles capital without generating external revenue. Data center construction, though massive (e.g., 15.2GW under construction by 2027), lacks corresponding revenue streams, requiring $157 billion annually to monetize—far exceeding current AI compute demand estimates. Zitron also notes that non-hyperscaler players struggle to compete, as building AI infrastructure requires expertise and capital beyond most startups. This centralization risks creating systemic weaknesses in smaller cloud providers like CoreWeave, which depend on hyperscaler contracts. Nice one!
[Read More]