Forthright Insight

Spring 2026 update

As we head into ITW 2026 in May, I’m intrigued by the contrast between hyperscale AI networks and the traditional WAN, with the wholesale market in between. 

The hyperscalers are building out their own AI networks. No one data center has sufficient space, power, or resiliency, so they link multiple data centers with custom fiber networks. These include multiple ducts and high-count fiber cables. This extension of the AI backend network is poised to eclipse cloud-era data center interconnects by an order of magnitude or more. 

Most of these latency-sensitive links are regional in scope, within the reach of pluggable optics, rather than dedicated transport systems. Hyperscalers have global network demand, but AI backend networks are geographically concentrated. Even a pan-USA megacluster is aspirational. 

How much of this captive demand leaks out to the wholesale market? Probably not much, directly, but even a small proportion could be substantial relative to the WAN. Meanwhile, we have demand from AI neoscalers that are not quite commissioning their own fiber builds, other DCI synchronization functions, and frontend, DC-to-user connectivity that stands to grow and traverse the WAN—especially as inference grows in proportion to training. Questions are, how much, and when. 

I hope to catch up with you at ITW to explore these dynamics relative to your business. I’m mainly doing market research for LightCounting, but also representing clients in the submarine cable business. I’m happy to advise on any aspect of the global connectivity business on the spot, or point you in the right direction if your interests extend beyond my expertise. 

For more about me, go to LinkedIn, or reach out through the contact form below. Thanks!

Erik Kreifeldt, Principal