NVIDIA GTC 2026 Opens: Jensen Huang Unveils Vera Rubin GPU and Groq 3 LPU, Projects $1T Demand and Declares 'Inference Inflection Point'

On March 16, 2026, NVIDIA's annual AI conference GTC 2026 opened in San Jose, with CEO Jensen Huang delivering a keynote that unveiled the next-generation AI platform 'Vera Rubin.' This platform comprises seven new chips: the Vera CPU, Rubin GPU, NVLink 6 Switch, ConnectX-9 SuperNIC, BlueField-4 DPU, Spectrum-6 Ethernet switch, and the newly integrated Groq 3 LPU, forming a comprehensive AI computing solution.

Huang stated that 'Vera Rubin is a generational leap, built to supercharge every stage of AI -- seven groundbreaking chips, five racks, one massive supercomputer.' Of particular note is the 'Groq 3 LPX' rack equipped with the inference-specialized Groq 3 LPU chip. Featuring 256 LPU processors, 128GB of on-chip SRAM, and 640TB/s scale-up bandwidth, it is designed to handle the low-latency, large-context demands of agentic AI. Combined with Vera Rubin, it delivers up to 35x improvement in inference throughput per megawatt.

Huang projected at least $1 trillion in demand by 2027 and declared that AI has reached an 'inference inflection point.' He highlighted the potential to expand revenue opportunities for trillion-parameter models by up to 10x, underscoring the rapid growth of the AI industry. OpenAI CEO Sam Altman commented that 'NVIDIA Vera Rubin will enable us to run more powerful models and agents at scale, delivering faster, more reliable systems to hundreds of millions of people.' The Vera Rubin platform is expected to become available through partner companies starting in the second half of 2026.