Keep pulling the thread on Jensen Huang.
NVIDIA has built an installed base of hundreds of millions of GPUs and computing systems that run CUDA over the past 20 years.
NVIDIA's confidential computing capability in its GPUs is now able to support protected deployment of models from OpenAI and Anthropic in various cloud regions.
NVIDIA's CEO estimates that the demand for computing has increased by one million times in the last two years, driven by generative, reasoning, and agentic AI.
NVIDIA's CEO projects at least one trillion dollars in demand for its products through 2027, an increase from the $500 billion projected through 2026 a year prior.
NVIDIA's Grace Blackwell architecture with NVLink 72 delivers 35 times higher performance per watt for AI inference compared to the previous generation.
According to a Semi-analysis report, NVIDIA's Grace Blackwell architecture is actually 50 times more performant per watt than the previous generation, exceeding NVIDIA's own claim of 35x.
NVIDIA is releasing NemoClaw, an enterprise-secure reference design for OpenClaw that includes a security technology called OpenShell.
NVIDIA is introducing the Vera Rubin platform, an AI supercomputer architecture featuring 72 GPUs connected by NVLink, delivering 3.6 exaflops of compute and 260 terabytes per second of all-to-all bandwidth.
NVIDIA is integrating Grok's LPU as a token accelerator into the Vera Rubin platform, creating the GroK3 LPX rack.
NVIDIA is in full production with its Spectrum X switch, the world's first switch with co-packaged optics (CPO).
NVIDIA's Vera CPU delivers twice the performance per watt of any other CPU in the world and is projected to become a multi-billion dollar business for the company.
NVIDIA is announcing the Rubin Ultra platform, which uses a new Kyber rack to connect 144 GPUs into a single NVLink domain.