Underwater cables are a vital piece of the AI buildout and internet — investment is booming (www.cnbc.com)

Global investment in subsea cables is surging as cloud and webscale firms race to link sprawling AI data centers: spending on new projects is forecast to hit about $13 billion between 2025–2027, driven by major bets like Meta’s 50,000 km Project Waterworth, Amazon’s Fastnet (>320 Tb/s capacity), Google’s Sol and numerous Microsoft investments. Undersea fiber already carries more than 95% of international data—hundreds of terabits per second across nearly a million miles of cable—so these privately owned routes are becoming a strategic part of the AI stack, enabling the high-throughput, low-latency connections needed to distribute training, inference and cloud services at global scale. That concentration raises engineering and security implications for the AI/ML community: capacity and latency are critical limits for multi-region model training and real-time inference (satellite links are higher‑latency and lower‑capacity), so more fiber increases practical scope for distributed compute. But subsea cables remain vulnerable—accidental damage from anchors/fishing and a recent uptick in suspected sabotage have prompted NATO patrols, tighter U.S. FCC rules on foreign involvement and closer supply‑chain scrutiny. For AI practitioners and ops teams this means planning around network redundancy, cross‑region routing, capacity provisioning and geopolitical risk as core parts of infrastructure design.