AI's Unsung Heroes
In 2025, an estimated $320 billion will be invested in the AI sector, with a significant portion allocated to infrastructure. This level of investment is unprecedented. The last time such a massive amount of capital flowed into a single sector was at the peak of the dot-com boom in 2000, when most of the investment went toward software startups rather than infrastructure.
Unlike the dot-com boom—when dial-up connections were the norm—the AI revolution requires immense computing power to process and transmit vast amounts of data. While AI's software advancements dominate headlines, the vital infrastructure powering it—data centers and interconnects—often goes unnoticed. NVIDIA is the most frequently mentioned company in AI infrastructure-related news, largely due to its dominant GPU market share.
Data centers may soon become the new global symbols of power. The US leads the world in the number of data centers (~45%), followed by Western Europe and China. Given China’s unmatched infrastructure development capabilities, the world should pay close attention to its rise in data center technology, particularly its innovative approach to undersea data centers like the one deployed in Hainan. Their data center advancements have likely played a role in the rapid emergence of GPT players like DeepSeek AI and Manus AI.
Despite its strengths in data center development, China may struggle in the interconnect sector—not due to a lack of capability, but rather the high barriers to entry in the global interconnect market. The US maintains a competitive edge in global interconnect infrastructure, with US telecoms such as AT&T, Lumen Technologies, Zayo Group, and Verizon holding a significant foothold in the global dark fiber business. Around the world, there is ample opportunity for regional dark fiber players to benefit—provided they can align their strategies with their data center ecosystem. Beyond this, full automation of dark fiber networks—following Google's recent push for AI-driven network optimization—will set the stage for dark fiber companies to compete in the AI era.
One of the biggest challenges in capitalizing on AI’s infrastructure boom is the sheer capital requirement. Building a hyperscale data center requires billions of dollars, while deploying a mile of dark fiber can easily cost hundreds of thousands, if not millions. On top of these expenses, navigating the intricate web of local regulations, permits, zoning, and resources—such as power and cooling—demands expertise in large-scale infrastructure projects.