Dear Readers,

Jensen Huang didn't launch a chip in San Jose last week; he launched a factory blueprint for manufacturing intelligence at an industrial scale. GTC 2026 was NVIDIA's most ambitious conference yet, unveiling the Vera Rubin platform as a seven-chip, five-rack supercomputer ecosystem while quietly absorbing Groq's LPU technology to split the inference pipeline into specialized lanes. The real power move, though, was software: Dynamo 1.0 as an open-source inference operating system and OpenShell as a security runtime that treats AI agents like processes in an OS, not chatbots with extra features. NVIDIA now sells the complete production line — from silicon to orchestration to safety — and wraps it in open-source packaging that runs best on its own hardware, a strategy as elegant as it is binding. The trillion-dollar revenue projection through 2027 isn't a forecast; it's a price tag for the dependency the industry is choosing to accept.

Superintelligence was there live and is giving you a deep dive into everything that made it special. Enjoy!

All the best,

When Jensen Huang took the stage at the SAP Center in San Jose on March 16, 2026, the crowd wasn’t waiting for another GPU, they were waiting for a vision. And what they got was something far more ambitious: a blueprint for the industrialization of intelligence itself. “This conference is going to cover every single layer of the five-layer cake of artificial intelligence,” Huang promised at the top of his two-hour keynote (Jensen Huang, GTC 2026 Keynote). What followed was not a product launch in any traditional sense, but more of a strategic declaration. NVIDIA no longer wants to be known primarily as a chipmaker. It wants to be the operating system, the factory floor, and the supply chain of the AI era.

The GPU Technology Conference (”G-T-C”), now in its seventeenth year, brought together more than 30,000 attendees in person, with millions watching virtually - and we, as Superintelligence, attended in person as well. There were 450 sponsors, over 1,000 sessions, and 2,000 speakers. But the sheer scale of the event belied the precision of its message. At GTC 2026, NVIDIA introduced an entirely new vocabulary for artificial intelligence infrastructure, one centered on “tokens per watt” and “cost per token”, as if the output of a language model were now a commodity to be manufactured, optimized, and shipped. Data centers, in Huang’s framing, are no longer data centers. They are “token factories.” The question is: does this framing hold up under scrutiny? Is NVIDIA building the indispensable foundation of the AI age, or constructing a beautifully engineered dependency trap?