AI data centers are growing to staggering sizes

AI data centers are rapidly expanding to unprecedented physical sizes, with some facilities now stretching over 1.24 miles in length, according to Ciena CEO Gary Smith.

Scale of expansion: The growing demand for GPU-powered AI computing is pushing data centers to expand both horizontally and vertically, creating massive multi-story facilities that challenge traditional networking approaches.

• Cloud providers are building data centers that span more than two kilometers in length
• These facilities are increasingly utilizing multi-story designs, adding vertical scale to horizontal sprawl
• Current campuses are blurring traditional boundaries between wide-area networks and data center infrastructure

Power requirements: The energy demands of these new AI facilities are reaching unprecedented levels, highlighting the infrastructure challenges of scaling AI operations.

• At least twelve new AI data centers are being planned or constructed that will each require one gigawatt of power
• By 2026, global AI processing is expected to consume 40 gigawatts of power, equivalent to eight times New York City’s daily power usage
• These power requirements underscore the massive scale of infrastructure needed to support AI development

Networking challenges: Traditional direct-connect GPU networking technologies are struggling to meet the demands of these expanding facilities.

• Current direct-connect solutions like Nvidia’s NVLink are facing limitations due to increasing distances
• The physical scale of new data centers is pushing beyond the capabilities of existing networking technologies
• Corporate campuses are experiencing strain on their network infrastructure as GPU clusters grow larger

Technology adaptation: Fiber-optic networking equipment, traditionally used in long-haul telecommunications, is being modified for use within these expanding AI data centers.

• Ciena is developing specialized fiber-optic equipment for intra-data center connectivity
• These solutions will be similar to long-haul telecom technology but optimized for GPU communication
• The transition to fiber-optic networking represents a significant shift in data center architecture

Looking ahead: Infrastructure evolution: The unprecedented scale of AI data centers is forcing a fundamental rethinking of data center design and networking technology, with implications for power infrastructure, networking solutions, and facility architecture. The industry appears to be approaching a critical juncture where traditional data center designs may no longer be sufficient for AI computing needs.