DEVOURED
OpenAI and Anthropic Could Dominate Global AI Compute
Anthropic and OpenAI are on track to control most global usable compute by 2028, potentially triggering massive shifts in sovereign debt and interest rates.
What: Dylan Patel (SemiAnalysis) discusses how rising lab revenues ($50M+ per megawatt) allow OpenAI and Anthropic to outbid other sectors for compute, leading to extreme centralization that may crash non-AI equity valuations and strain global debt markets.
Why it matters: This reveals the potential macroeconomic consequences of AI reaching 'super-profitability,' suggesting that capital might be sucked away from traditional industries into compute infrastructure.
Deep dive
- Compute centralization: AI labs are expected to capture 70-80% of incremental global compute by 2028.
- Economic re-allocation: The return on capital for AI compute is so high that it is forcing up the cost of borrowing for the entire economy.
- Sovereign risk: High-debt nations not involved in the AI supply chain face significant default risks if interest rates rise to match AI-driven capital demands.
- Training vs. Inference: Labs are increasingly shifting compute budget toward R&D and training rather than inference to accelerate toward AGI.
Decoder
- FLOPs: Floating point operations per second; the fundamental unit for measuring AI compute performance.
- RSI: Recursive self-improvement; the point at which an AI model can autonomously improve its own architecture or training processes.
- Capex: Capital expenditure; money spent to acquire or maintain physical assets like chips, power plants, and data centers.
Original article
Full article content is not available for inline reading.
Explore all 56 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi