Welcome to the much bigger, messier era of 'too big to fail' (www.cnn.com)

Nvidia announced a roughly $100 billion letter-of-intent to invest in OpenAI, effectively swapping cash for committed chip purchases: Nvidia funds data-center buildout and OpenAI buys Nvidia accelerators (at a discount). The arrangement — still vague on timing and terms — is emblematic of a new “too big to fail” era where a handful of firms are financially entwined. OpenAI currently pulls in about $13 billion in annual revenue, is projecting massive cash burn ($115 billion through 2029 per reports), and plans roughly 10 gigawatts of compute capacity (about the power of 8 million homes), which could require an additional ~$40 billion per gigawatt beyond Nvidia’s cash. For AI/ML, this is significant because it concentrates systemic risk and further monetizes infrastructure dependency: chipmakers, cloud providers, and model developers are recycling capital in circular deals (e.g., Amazon–Anthropic), obscuring whether demand growth or internal financing is driving purchases. The economics are daunting — Bain estimates AI firms will need ~$2 trillion in annual revenue by decade’s end to fund proposed data-center expansion, leaving an ~$800 billion gap — and product-side monetization is uncertain. OpenAI must convert massive user engagement into sustainable enterprise revenue despite mixed results (e.g., ChatGPT-5 backlash) and studies showing limited revenue impact from current AI deployments. The episode spotlights that funding compute scale may be less a technical than a solvency problem.