Jensen Huang, the charismatic CEO of NVIDIA, took center stage at CES 2025 for a 90-minute keynote address, unveiling a series of groundbreaking innovations and products that set the tone for the future of technology.
The RTX 5090 GPU
Huang opened with the highly anticipated RTX 5090 GPU, NVIDIA's latest flagship graphics card. This powerhouse boasts an incredible 92 billion transistors, delivers 4000 AI TOPS (trillion operations per second), achieves 380 RT TFLOPS (trillion floating-point operations per second) in ray-tracing performance, and features a 1.8 TB/s memory bandwidth. It also offers 125 Shader TFLOPS, enabling seamless shading performance for high-end applications.
The pricing for the RTX series was a pleasant surprise, coming in lower than anticipated: RTX 5090: $1,999, RTX 5080: $999 and RTX 5070: $549
Huang highlighted the RTX 5070's performance, claiming it rivals last year's RTX 4090, which retailed for $1,600. He attributed this leap in efficiency and cost-effectiveness to advancements in artificial intelligence.
In addition, NVIDIA unveiled a range of laptops featuring the new RTX GPUs, priced between $1,299 and $2,899.
Blackwell GPU: Production in Full Swing
Huang then shifted the spotlight to NVIDIA's much-anticipated Blackwell GPU, confirming it is now in full production. This product delivers unparalleled performance: 1.4 ExaFLOPS TE FP4 computational power across 72 Blackwell GPUs, 130 trillion transistors, 2592 Grace CPU cores, 72 ConnectX-8 NICs (network interface cards), 576 memory chips with a total capacity of 14TB and 1.2PB/s bandwidth and 18 NVLink switches, offering a bandwidth of 130TB/s.
Huang described market demand for Blackwell GPUs as insane, driven by the ever-growing need for computational power in the AI era.
AI Agents: The Next Trillion-Dollar Opportunity
The keynote progressed to AI Agents, which Huang positioned as the next trillion-dollar market opportunity. He demonstrated how AI agents can assist in tasks ranging from search assistance and factory operations to employee management and financial analysis.
Huang boldly declared that AI agents will soon become coworkers, requiring specialized training to perform specific tasks.
However, the audience's reaction to this vision appeared lukewarm. Observing this, Huang quickly pivoted to more tangible applications like autonomous driving and robotics.
Autonomous Driving: Thor Processor and Key Partnerships
NVIDIA's advancements in autonomous driving took the stage next. Huang revealed details about the Thor processor, NVIDIA's next-generation automotive chip, which delivers 20 times the performance of its predecessor, Orin. Thor also has applications in traditional robotics.
Huang announced a major collaboration with Toyota, which will use NVIDIA chips and its DRIVE OS platform to produce next-generation autonomous vehicles. DRIVE OS is the first autonomous driving platform to achieve ASIL-D certification, the highest safety standard in the automotive industry. Huang emphasized that with improved training data, the development of autonomous vehicles will accelerate rapidly in the coming years.
Robotics: Pioneering Human-Like Machines
Turning to robotics, Huang showcased over a dozen NVIDIA-developed robots. He acknowledged that training robots is significantly more complex than developing autonomous vehicles, as robots must adapt to highly varied human tasks. NVIDIA leverages its robust AI models and extensive datasets to train these robots effectively.
Huang also announced the open-sourcing of NVIDIA's Cosmos, a commercially viable foundational AI model trained on 20 million hours of driving and robotics video data. Cosmos is designed to accelerate the development of autonomous driving systems and next-generation robots.
Early adopters include robotics companies such as 1X and Agility Robotics, along with autonomous driving leaders Uber, XPeng, and BYD.
Project Digits: Compact AI Supercomputing
Huang concluded the keynote by unveiling Project Digits, a compact AI supercomputer roughly the size of a smartphone. Powered by the new GB110 chip (NVIDIA's smallest Blackwell GPU, developed in collaboration with MediaTek), this device is expected to launch in May. It can serve as a small workstation or be integrated with existing PCs.
Market Impact and Closing Remarks
The CES 2025 keynote highlighted NVIDIA's aggressive focus on AI, with nearly every product and innovation tied to advancing artificial intelligence. The official production launch of Blackwell GPUs is set to drive unprecedented growth for NVIDIA in the coming years.
As of press time, NVIDIA's stock price in after-hours trading fell slightly by 0.5%, to $148.7 per share.