Blackwell out, in with Rubin

Blackwell is old news, stand by for headlines about Rubin. I'm talking about AI chips from Nvidia. They've only just started shipping the former in a big way, and are already revealing details about the latter.

Nvidia's annual developers conference kicked off yesterday in San Jose, California. It's called the "AI Woodstock" and will have 25 000 in-person attendees.

Nvidia CEO Jensen Huang will launch a souped-up version called Blackwell Ultra at his keynote address later today. But the real interest is in the news about the next generation of AI chips called Rubin. They are expected to show very impressive performance improvements over Blackwell. Keep in mind that Blackwell provided 30 times faster performance than the company's previous generation on AI inferencing.

According to an article I read in the Wall Street Journal, Nvidia's data-centre business will hit about $237 billion in revenue in 2026, and then keep climbing by 30% per year over the next four years. And that's after catapulting by more than seven times over the last two.

This exciting scenario requires that Nvidia's biggest customers, Amazon, Google, Microsoft and Meta keep barrelling ahead with capex spending. To do that they'll have to see meaningful demand from consumers and businesses for premium generative AI services.