Nvidia Surpasses Moore's Law: Driving Down AI Costs
CEO Jensen Huang Unveils Breakthroughs in AI Chip Technology at CES 2025, Exceeding Traditional Growth Metrics
Nvidia CEO Jensen Huang announced groundbreaking advancements in artificial intelligence (AI) chip technology at the Consumer Electronics Show (CES) 2025, claiming that the evolution of Nvidia's AI chips is outpacing the predictions set forth by Moore's Law.
This development could lead to a significant reduction in costs associated with AI model usage.
Moore's Law, formulated in 1965 by Intel co-founder Gordon Moore, posits that the number of transistors on a chip doubles approximately every two years, substantially increasing performance without a significant rise in cost.
Intel identifies this principle as foundational to modern technological progress.
At the event, Nvidia introduced the GB200 NVL72, a superchip boasting 30 times the performance efficiency in handling large language model (LLM) workloads compared to its predecessor, the H100.
This performance leap is attributed to Nvidia's ability to optimize across the architecture, chip design, and algorithms, thereby overcoming some constraints imposed by Moore's Law.
Huang proposed the development of a specific law for AI systems' growth, focusing on pre-training, post-training, and inference.
Among these, inference—essentially the ability to generate precise responses through reasoning—remains the most costly, but Nvidia's innovations aim to mitigate these costs.
Furthermore, Nvidia anticipates that advancements akin to the GB200 NVL72 will make sophisticated models, like those developed by OpenAI, more financially accessible.
By democratizing access to high-quality AI, Nvidia seeks to enable both everyday and professional tasks such as text composition and supply chain management.
Looking toward the future, Huang imagines a world where superintelligent AI could revolutionize industries including robotics and chip design, enhancing daily life for users.
However, he insisted that the deployment of such technologies must be accompanied by human responsibility, emphasizing that "machines are machines" and should be designed to serve humanity's interests.
This development could lead to a significant reduction in costs associated with AI model usage.
Moore's Law, formulated in 1965 by Intel co-founder Gordon Moore, posits that the number of transistors on a chip doubles approximately every two years, substantially increasing performance without a significant rise in cost.
Intel identifies this principle as foundational to modern technological progress.
At the event, Nvidia introduced the GB200 NVL72, a superchip boasting 30 times the performance efficiency in handling large language model (LLM) workloads compared to its predecessor, the H100.
This performance leap is attributed to Nvidia's ability to optimize across the architecture, chip design, and algorithms, thereby overcoming some constraints imposed by Moore's Law.
Huang proposed the development of a specific law for AI systems' growth, focusing on pre-training, post-training, and inference.
Among these, inference—essentially the ability to generate precise responses through reasoning—remains the most costly, but Nvidia's innovations aim to mitigate these costs.
Furthermore, Nvidia anticipates that advancements akin to the GB200 NVL72 will make sophisticated models, like those developed by OpenAI, more financially accessible.
By democratizing access to high-quality AI, Nvidia seeks to enable both everyday and professional tasks such as text composition and supply chain management.
Looking toward the future, Huang imagines a world where superintelligent AI could revolutionize industries including robotics and chip design, enhancing daily life for users.
However, he insisted that the deployment of such technologies must be accompanied by human responsibility, emphasizing that "machines are machines" and should be designed to serve humanity's interests.
