The Contest for AI: It’s More Than Just Microchips, It’s About Energy
Over recent years, the focus in the AI sector has revolved around semiconductors. However, the landscape has shifted, everyone. The true challenge? It’s no longer a matter of chip production quantities, but rather how to power them. What many are contending with now is access to power grids, interconnection delays, and aging infrastructure that has seen better days.
The Energy Transformation: AI’s Emerging Domain
Envision’s head honcho, Lei Zhang, made headlines at VivaTech in Paris this June, discussing how AI is just as much about revolutionizing energy as it is about computing. Think of the steam engine converting coal into energy? Today, GPUs are performing a similar feat for AI by transforming electricity into intelligence. The key takeaway? Energy efficiency reigns supreme, much like James Watt revolutionized the industry in his era.
The Genuine Threat: AI Consuming the Power
Statistics don’t lie, my friend. Goldman Sachs predicts US datacenter power demand will reach 31 GW by 2025, soaring to 66 GW by 2027. However, reality check—only about 72 percent of these facilities are expected to be up and running on schedule due to electricity being a finicky factor. As AI demand is set to triple, we’re looking at a possible 3 percent slice of worldwide electricity consumed by AI by 2030.
Grid Pressure: The Expanding Discrepancies
There’s an ongoing mismatch: while AI models are updated every six months and chip technology advances yearly, power grids have remained static for a long time. With AI server power density escalating elevenfold from 2020 to 2025, and another fourfold increase anticipated by 2027, our reliable power electronics and transformers are under increasing strain.
Mission Gobi: AI Pursues Energy Sources
Introducing Mission Gobi. Launched by Envision at VivaTech, it plans to establish 5 GW of green AI computing capacity across deserts by 2030. The clever strategy? To have computing follow energy sources, reversing the traditional model where energy flows to technology. Deserts, with their broad expanses of inexpensive land and abundant solar and wind resources, present an ideal setting for this endeavor.
Computing Should Follow Energy
The concept is straightforward: rather than tapping into power from homes and industries, Mission Gobi aims to create new renewable energy infrastructures specifically for AI, expanding the overall energy pie instead of dividing it further.
Establishing Energy-Centric AI Framework
Envision’s strategy comprises three layers: an intelligent operational hub, Physical AI supported by Tianji and Dubhe models, and advanced power infrastructure. The main hurdle? Harmonizing renewable energy, since AI facilities demand stable power, and to be frank, solar and wind power can be just as unpredictable as British weather.
From Theory to Practice: Envision’s Real-World Experience
Envision has already acted decisively. In Chifeng, Inner Mongolia, they have a 2 GW system entirely powered by renewable sources. Alongside it, a gigawatt-scale AI campus is being constructed in Ulanqab, serving as a demonstration of what energy-centric computing could resemble.
SpaceX Gazes Upward, Mission Gobi Looks Ahead
Envision isn’t the sole entity recognizing energy as AI’s vulnerable point. Elon Musk’s SpaceX is eyeing orbital datacenters fueled by continuous solar power from space. Both initiatives seek plentiful energy, whether in outer space or across arid deserts.
A New Model for AI Infrastructure
If the industrial era revolved around coal and the electrical era centered on power grids, the AI era could be defined by energy abundance. The future of AI infrastructure doesn’t merely rely on GPU quantities; it concerns new energy sources, reducing community burdens, and fostering technological advancements without monopolizing the existing energy resources.
Provided by Envision.
What Lies Ahead for AI?
In the broader perspective, it’s evident that the AI competition is heating up, not just regarding technology but also energy. Stay tuned to discover if deserts emerge as the future’s computing hubs or if another contender has a surprise in store. Regardless, the next significant AI revolution might just require a bit more sunlight and wind to thrive.
