At the recent GTC (GPU Technology Conference), Jensen Huang reexamined what artificial intelligence truly is. He constructed a concept of a "five-layer cake," pointing out that energy powers advanced computers, which form AI factories; the models running in these factories generate tokens, or digital intelligence, which are then integrated into various applications. He also mentioned that "to participate in the AI economy, one must become an AI producer; consuming AI is only part of it."
Huang noted that there are significant differences between the AI industry landscapes of Europe and the US and China. AI in Europe is more focused on industrial applications, such as automotive design, manufacturing, and drug development, which are important scenarios for its AI development. With a solid industrial foundation, Europe has unique characteristics in its AI usage. However, in terms of large models and cloud platforms, Europe relies more on American companies, using external models to meet the demand for rapid AI efficiency improvements. While this can solve short-term issues, the high costs of using external models and the fact that American companies have access to local companies' data create inherent conflicts, making security hard to guarantee.
The Impact of Huang's AI Producer Concept on Europe
The idea of becoming an AI producer that Huang mentioned precisely hits the urgent need for change in Europe's AI landscape. Each layer of his "five-layer cake" proposes corresponding solutions, especially in the construction of AI factories. He emphasized that the primary task of becoming a producer is to build upstream systems, namely infrastructure. In the AI industry, this includes chips, computing power, and cloud services. Once these infrastructures are in place, companies can utilize local computing power to train models with their own data, and for sensitive data, companies can confidently upload it to local cloud platforms.
This way, companies can enhance their control over data and computing power, reduce reliance on external services, and lower model usage costs in future operations. However, building these infrastructures is no easy task, involving multiple aspects such as energy, cooling, connectivity deployment, and local debugging, which takes time. Starting from scratch, from purchasing equipment to installation and commissioning, could take several months. Huang pointed out, "If your expected annual revenue is several billion dollars? Even if calculated at $40 billion/year — this infrastructure produces $40 billion a year — if there's just a 3-month delay, that's a loss of $10 billion."
NVIDIA's proposed DSX POD aims to enable users to generate revenue sooner. The DSX POD moves some on-site engineering into the factory, integrating seven liquid-cooled computer racks and supporting systems into a transportable container, ensuring all facilities are operational before on-site delivery. According to his predictions, the time from equipment arrival to "producing tokens" can be shortened from "4 to 5 months" to "about 1 month," which not only saves costs from idle equipment but also reduces losses from delayed revenue. Additionally, the official statement indicated that by combining DSX's power management, the same AI output could reduce power demand by up to 30%.
Notably, power management is also one layer of the "five-layer cake." As chip computing power continues to improve, power supply, cooling, and deployment speed may become new bottlenecks, which is also why NVIDIA is considering building factories. Huang emphasized that helping Europe transition from an AI consumer to an AI producer requires starting from scratch and gradually addressing all aspects of effectively utilizing computing power.
In terms of energy, Europe has established energy companies such as Schneider and Siemens. Schneider Electric is responsible for power distribution and liquid cooling, Siemens evaluates thermal, noise, and grid impacts, while Dassault Systèmes connects racks with factory design. NVIDIA is responsible for providing power management solutions. Such collaboration not only allows NVIDIA to better understand its products but also brings new business opportunities, while European companies can integrate their existing products and engineering capabilities into new infrastructure development. This is not an "invasion" by a US company, but a multi-party win-win efficiency revolution.
Furthermore, Huang proposed the concept of industrial ecosystem collaboration, such as model, simulation, and agent development tools. If companies require further development starting points, NVIDIA will consider and provide finished products. Of course, European companies are not starting from scratch; existing industrial software, self-developed models, and other suppliers' more suitable products can continue to be retained in the system, and companies ultimately need to compare performance, cost, and compatibility with existing businesses.
In this way, NVIDIA's ecological position in Europe has a solid foundation. Various constructions centered around NVIDIA's infrastructure will establish a very broad ecosystem in Europe. Once these ecosystems and trust are gradually built, proven products and experiences will become reasons for continued adoption of this solution. For users, localized AI deployment not only reduces usage costs but also brings new business opportunities, with regulatory responsibilities managed locally.
Local cooperation, customer trust, and joint investment in brand ecosystems will gradually accumulate. In a future where Moore's Law slows down, technology may catch up, but with such a close cooperation foundation, this will become an unshakeable business fortress.
In his speech, Huang pointed out, "After decades of development in the computer industry, we are now at the beginning of a new industrial revolution." He recalled that during the first industrial revolution, the spinning jenny was an early representative invention that allowed one worker to operate multiple spindles simultaneously. With the improvement in spinning efficiency, the expansion of production scale, and the increased demand for raw materials, a series of changes were triggered, initially in the textile industry, and later gradually expanded to the entire production mode of society. Changes in productivity often lead to widespread effects.
Today's large models are like the spinning jenny, enhancing the work efficiency of ordinary people to varying degrees. As AI's participation in actual production continues to increase, the scale of use will also expand. The demand for raw materials — energy and computing power — will spur the upstream industrial chain; as technology advances, more convenient and efficient applications and products will emerge. The interdependence between industries will strengthen, and industrial ecosystem collaboration will become a necessary requirement.
However, whether these links can progress together will affect the final output. Goldman Sachs recently pointed out in discussions about China's AI supply chain that "AI technology inherently has a multiplier effect — if multiple components in the supply chain lag by one generation, it may lead to total computing power output/efficiency lagging by more than one generation." Therefore, the development level of each link is crucial, highlighting the importance of industrial collaboration. Even if chip computing power is fast, if network transmission, power supply, or cooling cannot keep up, performance cannot be fully realized; once computing power is ready, software and business processes must align to translate into actual efficiency gains for companies.
Technological leadership in a single link can only translate into actual productivity with the cooperation of other links; industrial collaboration will affect how much return AI investment ultimately generates.
NVIDIA's "European AI Development Strategy" is a microcosm of this industrial revolution. NVIDIA's approach is to connect industries that originally belonged to different fields with a well-integrated full-stack technology solution, combined with localized cooperation. This transforms it from a chip supplier into a glue for industrial collaboration in Europe. The value of the entire solution lies in transforming the gradual integration process between industries into a collaboration that can be planned and advanced in advance.

