The Silent Bottleneck: Why AI’s Future Hinges on Networks We’re Ignoring
If you’ve been following the AI boom, you’ve likely heard the hype about GPUs, models, and data lakes. But here’s a truth that’s flying under the radar: the real bottleneck for AI isn’t compute power—it’s the networks connecting it all. Personally, I think this is one of the most overlooked stories in tech right now. While everyone’s obsessing over who has the biggest GPU farm, the plumbing that moves data between those farms is quietly becoming a crisis.
The Neocloud Mirage: Compute Without Connectivity
Let’s start with the rise of neocloud providers—those GPU-as-a-service platforms that popped up to meet the insatiable demand for AI compute. Companies like CoreWeave and Gcore have scaled their hardware impressively, but here’s the catch: many of them built their networks as an afterthought. What makes this particularly fascinating is how their origins betray them. CoreWeave started in crypto mining, where network latency was a nuisance, not a dealbreaker. Gcore? They were in content delivery, where speed mattered but not in the same way it does for AI.
From my perspective, this is where the AI gold rush gets messy. Neoclouds are selling raw power, but their networks are a patchwork of legacy systems and rushed upgrades. Omdia’s warning that networking will “make or break” these providers isn’t hyperbolic—it’s a cold dose of reality. If you take a step back and think about it, AI workloads aren’t just about crunching numbers; they’re about moving data across continents, clouds, and edges in milliseconds. A network built for yesterday’s workloads won’t cut it.
The Nervous System Analogy: Why Kate Johnson is Right
Lumen CEO Kate Johnson’s open letter is a wake-up call wrapped in a metaphor. She compares networks to a nervous system, and I couldn’t agree more. What many people don’t realize is that AI isn’t a static process—it’s a dynamic, distributed ecosystem. Bots, agents, and models are constantly interacting, generating, and consuming data. If the network is the nervous system, latency is the equivalent of a neurological disorder.
Here’s where it gets interesting: Johnson claims that over 50% of internet traffic is now generated by autonomous workers. This isn’t just bots spamming websites—it’s AI models training, inferencing, and collaborating. If networks can’t adapt to this new reality, AI investments will stall. It’s like building a supercomputer and connecting it with dial-up.
The Sovereignty Factor: A Hidden Network Challenge
One thing that immediately stands out is how data sovereignty complicates this picture. Camille Mendler from Omdia points out that AI workloads don’t just move—they move across borders. This raises a deeper question: How do you build a network that’s not only fast and secure but also compliant with a dozen different regulations? What this really suggests is that the network isn’t just a technical problem—it’s a geopolitical one.
Personally, I think this is where the next wave of innovation will happen. Networks will need to be programmable, consumption-based, and border-aware. It’s not enough to throw more bandwidth at the problem; you need intelligence baked into the infrastructure.
The Future: Networks as the New Cloud
If you ask me, the network is about to become the next cloud. Just as we stopped thinking about servers and started thinking about services, networks will shift from being invisible pipes to strategic assets. This isn’t just speculation—it’s already happening. Providers like Lumen are positioning themselves as AI-first network operators, and enterprises are starting to pay attention.
But here’s the kicker: this transition won’t be smooth. Legacy networks, siloed architectures, and outdated mindsets will slow things down. What many people don’t realize is that upgrading a network isn’t like swapping out a GPU—it’s a multi-year, multi-billion-dollar endeavor.
Final Thoughts: The Unseen Foundation of AI
As we marvel at AI’s capabilities, let’s not forget the networks that make it all possible. In my opinion, this is the most underappreciated story in tech today. Networks aren’t just the backbone of AI—they’re the nervous system, the circulatory system, and the immune system all rolled into one.
If there’s one takeaway, it’s this: AI’s future isn’t just about smarter models or bigger data. It’s about building networks that can keep up. Because without them, even the most advanced AI will be stuck in the slow lane.
