At the Data Center Knowledge News Desk during Data Center World 2025, SemiAnalysis technology analyst Jeremie Eliahou Ontiveros provided insights into how AI infrastructure is evolving, highlighting rack-scale computing’s emergence as a key architectural approach for AI workloads.
“We’re exploring unprecedented territory in terms of complexity, especially on the networking side,” Eliahou explained, addressing the industry’s push toward 1-MW racks.
He pointed to the staggering scale of current solutions, noting that Nvidia’s NVL72 racks already contain thousands of copper cables – a complexity that will require revolutionary solutions like co-packaged optics to overcome.
The analyst identified two distinct timeframes shaping the industry’s trajectory. In the immediate term, power access remains the critical bottleneck, with providers struggling to secure hundreds of megawatts or even gigawatts of capacity in strategically valuable locations.
However, looking toward 2030’s projected $1 trillion data center capex milestone, Eliahou emphasized that economics will become the defining factor.
“If you want to justify a trillion dollars of capex, you’re going to have to come up with return on investment,” he stated, suggesting that financial considerations will ultimately drive decisions around emerging technologies like fuel cells and behind-the-meter natural gas generation.
▶️ Watch the full interview here:
Deutsch 
English