A wave of AI data center spending is set to keep pouring money into chipmakers, which is good news for Nvidia, TSMC and Broadcom, but the payoff comes with real risks.

• Global spending on AI infrastructure could reach $1.4 trillion a year by 2030, with the biggest cloud companies alone spending close to $800 billion next year.
• Nvidia is the main supplier of AI computing power, and its new Vera Rubin systems started shipping in August — each data center now needs far more expensive chips than a year ago.
• TSMC makes the advanced chips for Nvidia, Broadcom, AMD and Apple, and its most cutting-edge processes now bring in most of its chip revenue.
• Broadcom is taking a different route, designing custom AI chips for cloud giants like OpenAI, Meta and Anthropic so they rely less on Nvidia.
• The risks are real: Nvidia depends on smooth production and memory supply, TSMC faces thinner profits from new factories and China tensions, and Broadcom leans on a handful of big customers.

Outlook: The money keeps flowing toward chips for now, but power, cooling and networking suppliers may end up sharing the windfall.