The AI Hype Triumvirate: Nvidia, OpenAI, and Oracle's Circular Bet (pshi.substack.com)

NVIDIA, OpenAI and Oracle have announced a tightly interlinked set of deals that look more circular than catalytic: NVIDIA has pledged a roughly $100B equity infusion into OpenAI (a non‑binding LOI), OpenAI signed a reported $300B long‑term compute contract with Oracle, and Oracle has committed at least $40B to buy NVIDIA chips to power the data centers that will run that compute. NVIDIA’s press release also frames a plan to deploy roughly 10 GW of NVIDIA systems, with the first gigawatt not expected until H2 2026 and Oracle’s deal ramping in 2027 — timelines that make much of the headline value distant and conditional. For the AI/ML community this matters because it’s effectively vendor financing: NVIDIA is capitalizing a major customer, that customer is committing to a big cloud vendor, and that vendor in turn is buying back NVIDIA hardware. That creates the risk of artificial demand, distorted capacity planning, and obscured ROI (think the late‑90s telecom equipment playbook). Technical implications include potential overprovisioning of high‑end accelerator capacity, compressed pricing signals for chip makers and cloud compute, and reduced transparency around real, incremental compute needs for research and startups. The announcements will accelerate infrastructure buildout, but practitioners and investors should distinguish durable demand from circular capital flows that could mask misallocated compute investment.