AI 遠未結束——英偉達仍在引領遊戲

$英偉達(NVDA.US) Nvidia’s relentless pace of innovation is now the industry standard: a new generation every year. While the Blackwell platform is still in high demand and supply can’t keep up, CEO Jensen Huang has already confirmed that the next-gen “Ruben” platform’s six core chips have all completed tape-out at TSMC.

This isn’t just a blow to competitors—it’s a clear message to customers: Nvidia’s speed of innovation is its strongest shield. Any strategy of “waiting” or “finding alternatives” risks falling behind quickly.

Jensen also raised the long-term market forecast for AI infrastructure from the previous USD 1 trillion to an epic USD 3–4 trillion by the end of the decade. He backed this up by pointing out that top cloud service providers have doubled their capital spending to USD 600 billion in just two years. As an analyst, this is a strategic reset, anchoring Nvidia’s valuation to a much bigger future.

What’s driving demand now is shifting from “generative AI” to “reasoning AI”—Jensen repeatedly emphasized that “Reasoning AI” and “Agentic AI” will be the next wave, requiring 100 to 1,000 times more computing power than simple chatbots. This narrative shift is crucial: it answers the market’s question about whether AI compute demand will saturate. The story has moved from “teaching AI to talk” to “teaching AI to think and solve problems,” laying the foundation for continued exponential growth in compute demand.

Nvidia also shared a striking figure: US data center construction is expected to grow at a 46% CAGR over the next five years, potentially reaching 12% of US GDP. And it’s not just Nvidia—companies like AVGO, MRVL, AMD, and TSM are all projecting similar numbers.

The AI boom is far from over, and the real competition may just be getting started.