Nvidia CEO Jensen Huang took the stage at GTC Taipei on Wednesday to unveil the Vera Rubin platform, a new computing system the company says has already racked up $1 trillion in projected orders. The massive figure — roughly 60 times Nvidia's annual revenue — signals just how aggressively hyperscalers and AI labs are betting on the next generation of hardware.
What the Vera Rubin platform is
Huang described Vera Rubin as a purpose-built architecture for AI workloads, though the company has not released detailed specs. What is clear: the platform is designed to handle the massive compute demands of training and running frontier models. The $1 trillion order book, if realized, would dwarf any previous product launch in the semiconductor industry.
Why inference costs matter
One of the biggest bottlenecks in deploying AI at scale today is the cost of inference — the process of running a trained model to generate outputs. Huang said Vera Rubin is expected to significantly lower those costs, making it cheaper for companies to run chatbots, coding assistants, and other generative AI tools. That could accelerate adoption, especially among businesses that have been priced out by current GPU rental rates.
Market expectations shift
The announcement could reshape how investors and analysts think about AI economics. For months, Wall Street has debated whether the enormous capital spending on AI infrastructure will ever pay off. A platform that substantially cuts inference costs might change that calculation — if it delivers. The $1 trillion in orders suggests customers believe it will.
But questions remain. Nvidia hasn't said when Vera Rubin will ship, how much each unit will cost, or whether the projected orders include firm contracts or letters of intent. The company faces stiff competition from AMD, Intel, and a growing crop of AI chip startups, as well as cloud giants building their own custom silicon. Huang's presentation at GTC Taipei was light on benchmarks; the real test will come once the hardware reaches customer data centers.
