EP 21 — ASG’s Dan Golding on CRE’s Fake Data Center Problem

The AI boom has triggered the most dramatic transformation in data center history, but it's also created an epidemic of “fake” projects that threaten to undermine the entire sector. Daniel Golding, Chief Technology Officer, Appleby Strategy Group, has coined the term "fake data center" to describe the 90% of proposed AI infrastructure projects that exist only on paper. In his conversation with Sean, Dan reflects on how AI workloads have shattered 25 years of predictable data center development, with single racks now consuming 120 kilowatts — equivalent to powering 30 houses — compared to 10 kilowatts for traditional cloud computing.

 

This extreme power density has forced operators to abandon air cooling systems entirely in favor of liquid coolants, while project costs have skyrocketed to $5 billion for a single AI-capable building, eliminating speculative development and requiring 100% pre-leasing before ground breaking. Dan also touches on why fiber connectivity often determines project success more than power availability, and why small modular nuclear reactors represent the only sustainable solution for AI's massive energy requirements that only hyperscalers like Google, Meta, and Microsoft have the capital to fund.

 

Topics discussed:

 

• The “fake” data center epidemic where 90% of proposed AI infrastructure projects lack actual land control, power allocation, or network connectivity despite seeking significant media attention.
• How AI workloads have increased power density from 10 to 120 kilowatts per rack, forcing the abandonment of air cooling systems in favor of liquid coolant solutions.
• The network effect phenomenon in data center land valuation, where proximity to existing facilities and internet exchange points creates exponential rather than linear value increases.
• Why the donut model of internet interconnection eliminated traditional hierarchical ISP structures and enabled massive dark fiber deployments between data center campuses.
• The elimination of speculative development in AI data centers due to billion-dollar price tags that hobble elasticity.
• Why network connectivity represents the most overlooked aspect of site selection, with fiber infrastructure often determining project viability more than power availability.
• Small modular nuclear reactors as the essential long-term solution for AI's energy requirements, with hyperscalers positioned as the only entities capable of funding initial deployment.