The explosion of artificial intelligence is doubling data center power demands every two years, yet the public has almost no visibility into what actually runs inside these colossal facilities. A new Forbes analysis pulls back the curtain on the myths and realities of AI data centers, revealing a sector grappling with unprecedented energy hunger, cooling crises, and chip shortages.
Forbes' tech council article, published June 12, 2026, argues that most discussions about data centers are not grounded in fact. Many people have limited visibility into how these facilities operate or what AI workloads require. The piece aims to correct misconceptions as AI infrastructure spending skyrockets.
AI data centers differ fundamentally from traditional cloud data centers. Traditional centers handle varied workloads like web hosting and email, with moderate, bursty demand. AI data centers, however, are designed for massive parallel computing—training and running large language models like GPT-5 or Gemini Ultra. These workloads require thousands of GPUs running at near-peak power around the clock, generating immense heat and consuming up to 10 times more power per rack than conventional servers.
Key details are stark. Hyperscalers—Microsoft, Amazon (AWS), Google, and Meta—spent over $150 billion on AI data center construction and equipment in 2025 alone, according to industry estimates cited by Forbes. Each facility can draw 500 megawatts or more, equal to a small city. The International Energy Agency projects that AI data centers could consume 8–10% of global electricity by 2030, up from roughly 2% today. Chip demand is equally voracious: NVIDIA shipped nearly 4 million H100 and B200 GPUs in 2025, yet allocation remains tight. Cooling is a critical bottleneck—liquid cooling, including direct-to-chip and immersion, is replacing traditional air conditioning, but retrofitting legacy centers is expensive and slow.
Analysis from energy and tech observers highlights a deeper conflict. AI data centers' insatiable power appetite threatens corporate net-zero pledges. Microsoft, for instance, saw its emissions rise 30% since 2020 due to AI infrastructure, despite its carbon-negative promise. Local communities are pushing back against new facilities over water usage and grid strain. The Forbes piece suggests that the industry is at a crossroads: either innovate on energy efficiency and clean power procurement, or face regulatory clampdowns and public backlash.
What happens next is critical. Hyperscalers are investing in next-generation cooling, small modular nuclear reactors, on-site solar and battery farms, and grid-interactive data centers that can shift workloads to times of lower demand. Meanwhile, governments are drafting AI energy reporting mandates, and chip designers are racing to create lower-power AI accelerators. The next 24 months will determine whether AI growth can be sustainable or whether energy constraints will throttle its potential.
AI
→ Neutral
The Truth About AI Data Centers And Why It Matters
Not every discussion about data centers is grounded in fact, and many people have limited visibility into how these facilities operate or what AI workloads require.
Expert Panel®, Forbes Councils Member
Forbes
3 min read
7/10
Key Takeaways
- AI data center power demand is projected to reach 8–10% of global electricity by 2030, up from roughly 2% today, per IEA forecasts.
- Hyperscalers Microsoft, Amazon, Google, and Meta invested over $150 billion in AI data center infrastructure in 2025 alone.
- NVIDIA shipped approximately 4 million H100 and B200 GPUs in 2025, yet supply still falls short of demand from AI training clusters.
- Liquid cooling adoption in hyperscale AI data centers grew from under 10% in 2022 to over 40% in 2025, driven by thermal density exceeding 100 kW per rack.
- Microsoft's total greenhouse gas emissions increased 30% since 2020, largely attributed to the buildout of AI data center capacity.
Frequently Asked Questions
An AI data center is a specialized facility designed to handle the massive computational demands of artificial intelligence workloads, including training and running large language models. Unlike traditional data centers, AI data centers rely on clusters of high-performance GPUs running at near-peak power 24/7, requiring much more electricity and advanced cooling systems.
Individual AI data centers can draw 500 megawatts or more—equivalent to a small city. The International Energy Agency projects that AI data centers could consume 8–10% of global electricity by 2030, up from approximately 2% today. Power demand is doubling every two years due to the rapid growth of AI workloads.
Traditional air conditioning is being replaced by liquid cooling methods such as direct-to-chip cooling and immersion cooling. These are necessary because AI racks generate over 100 kW of heat per rack, far beyond what air cooling can handle. Adoption of liquid cooling among hyperscalers grew from under 10% in 2022 to over 40% by 2025.
AI data centers have a significant environmental footprint due to high electricity consumption, water usage for cooling, and associated carbon emissions. Microsoft's emissions rose 30% since 2020 largely due to AI infrastructure. Local communities are raising concerns about grid strain and water scarcity near new facilities.
The largest builders are hyperscalers: Microsoft, Amazon Web Services, Google, and Meta. These companies spent over $150 billion on AI data center construction and equipment in 2025 alone. Other players include Oracle, IBM, and specialized providers like CoreWeave.
The industry is moving toward on-site clean energy solutions like solar and battery storage, small modular nuclear reactors, and grid-interactive data centers that shift workloads to off-peak hours. Chip designers are also developing lower-power AI accelerators. Government mandates for energy transparency are expected to accelerate these trends.
Topics
Original source
www.forbes.com
Discussion
Join the discussion
Sign in to post a comment or reply.
No comments yet. Be the first to share your thoughts!