For years, the global AI race fixated on semiconductors—the quest for faster GPUs, the scramble for advanced fabs, and the billions poured into colossal data centers. Yet, a fundamental shift is underway: the hardest problem in AI is no longer the chip, but the megawatt. The binding constraint has transitioned from raw compute power to the energy required to run it, pushing grid access and aging infrastructure to the forefront of challenges.
This pivotal observation came from Envision founder and CEO Lei Zhang at VivaTech, framing AI's ascent as an energy revolution paralleling its computing one. Just as James Watt optimized steam engine efficiency to transform industry, AI now demands an energy-first approach to sustainable growth.
What Happened
The insatiable energy appetite of AI is creating a structural mismatch between rapidly evolving AI models and chips, and static, decades-old power grids. The numbers paint a stark picture:
- Surging Demand: Goldman Sachs projects US data center power demand to more than double from 31 GW in 2025 to 66 GW by 2027. This assumes that a significant portion of planned facilities will face delays due to electricity, not construction, issues.
- Global Impact: The International Energy Agency (IEA) estimates data centers consumed roughly 1.5 percent of world electricity in 2024, a share expected to rise to 3 percent by 2030 as AI-specific demand triples.
- Density Explosion: Rack densities are soaring from 5 kW towards an astounding 200 kW. AI server power density itself increased elevenfold between 2020 and 2025, with a further fourfold rise anticipated by 2027, severely straining power electronics and transformer supply chains.
This unprecedented demand is leading to difficult societal questions. Communities are increasingly concerned about AI infrastructure drawing power that homes, factories, hospitals, and public services rely on. The potential for higher consumer bills, limited grid capacity for manufacturers, and the overall burden on public infrastructure are becoming pressing non-technical challenges.
In response, Envision unveiled Mission Gobi at VivaTech. This ambitious initiative aims to develop 5 GW of green AI computing capacity across deserts and arid regions by 2030. The core premise is a reversal of traditional logic: instead of energy following computing, computing must now follow energy. Deserts offer unparalleled solar and wind resources, vast low-cost land, and minimal competing residential or industrial demand, providing an opportunity to build entirely new, dedicated renewable energy systems for AI.
Why It Matters
For developers, IT professionals, and enterprise decision-makers, this shift profoundly alters the landscape of AI infrastructure. The focus moves beyond merely acquiring the latest GPUs to a holistic consideration of the entire energy ecosystem:
- Infrastructure Planning Redefined: IT architects and operations teams must prioritize power availability, grid resilience, and energy efficiency above all else. Location intelligence for data centers will increasingly factor in renewable energy potential and grid independence, rather than just proximity to fiber or population centers.
- Cost Implications: The rising cost and scarcity of power will directly impact the operational expenditure of running large AI models. This could drive innovation in more energy-efficient algorithms and hardware, but also necessitate entirely new budgeting approaches for AI projects.
- Sustainability as a Core Constraint: Environmental impact is no longer a 'nice-to-have' but a fundamental technical constraint. Solutions like Mission Gobi highlight a future where AI's carbon footprint is proactively managed at the infrastructure level, pushing developers and platforms to optimize for lower energy consumption.
- Supply Chain Strain: The surge in power density puts immense pressure on the supply chains for power electronics, transformers, and cooling systems. IT teams need to anticipate potential delays and increased costs for these critical components, not just for server hardware.
- New Geographic Opportunities: The concept of 'AI follows energy' opens up new possibilities for building AI centers in previously unconventional locations. This could lead to regional decentralization of AI compute, potentially impacting latency and data sovereignty considerations for global applications.
What To Watch
Mission Gobi represents a bold vision, but its execution and scalability will be key. We should watch for:
- Technological Innovation in Energy: Beyond just solar and wind, look for advancements in energy storage, microgrids, and highly efficient cooling solutions tailored for extreme data center densities.
- Policy and Investment: Governments and investors will need to support such large-scale renewable energy infrastructure projects specifically for AI, including streamlining grid interconnection and land use.
- AI Model Efficiency: Continued pressure for developers to create more energy-efficient AI models and training methodologies. This includes advancements in quantization, sparse models, and hardware-software co-design to minimize power draw per inference or training step.
- Alternative Cooling Technologies: As rack densities climb, traditional air cooling becomes less viable. Expect accelerated development and adoption of liquid cooling, immersion cooling, and other advanced thermal management techniques.
The era of cheap, abundant power for AI is rapidly drawing to a close. The industry's ability to innovate not just in compute, but in energy, will define the next phase of AI development.