No network, no AI evolution

The AI conversation is obsessed with models, chips and benchmark leaps. The truth is simpler: AI is only as powerful as the connectivity that links data, compute and users.

If AI is the brain, then connectivity is the nervous system. It’s what turns “intelligence” into something that can sense, coordinate and act. 

A model in isolation is just ones and zeros with opinions, the value appears when it’s stitched into feedback; fed by live data from people, machines and systems, and able to push decisions back out quickly. It turns islands of experimentation, into an AI-powered organization, or even a country.

Nokia’s Build first, lead forever research makes the gap hard to ignore.

Most respondents said AI is transforming their business, with three in every four running or piloting AI today, however 88% believe the network threatens to inhibit future growth of AI. In plain terms, the brain is ready, but the nervous system isn’t.

That is the real challenge for the U.S. in the AI supercycle.

Infrastructure can look “good enough” right up until the moment it isn’t.

Connectivity networks were designed for the way we have used digital services over the past 20 years. We downloaded files, navigated the internet and streamed. It is all downlink-heavy usage.

AI flips that logic, it needs to be fed. When was the last time you checked your home WiFi’s upload speed? 100 MBPS download speed with 10 MBPS upload is average today, but this won’t work for the AI evolution. 

The flow of data becomes more bi-lateral and distributed, with far less tolerance for delay, jitter, packet loss and downtime. As consumers, we become frustrated with buffering, but AI won’t tolerate it.

AI changing the geometry of the internet and when the geometry changes, the network design must change with it.

Not just a faster network, but a new type of network

In Nokia’s research, it is clear that the current standards aren’t working. 

Three out of four decision makers expect they’ll need sub-30ms latency for AI in the next 2–3 years, including 13% targeting sub-10ms. This is a new consideration for network design and delivery.

Recent analysis from Ookla suggests the U.S. network has work to do. Not a single state in the U.S. has latency below 30ms. At the high range is Hawaii at 108ms and the low range is D.C. at 37ms. Only 15 states had readings below 50ms.

Latency is the time in which networks take to respond. Generally speaking, the lower the better as when the delay grows it starts to feel like a bad video call where people talk over each other. 

We’ve not had to be significantly concerned with latency to date as real-time responsiveness hasn’t been critical; we tolerate a small amount of buffering when streaming for example. 

For AI systems making decisions in the moment, this is not acceptable. 

Fraud detection and payment authorization, autonomous transport safety systems or keeping the lights on when the grid is under stress, these are examples where low latency becomes mission critical. 

That’s why connectivity keeps surfacing as the real choke point. In Nokia’s research, 58% of decision-makers say the network is the single biggest barrier to scaling AI, not the models, but the ability to access them everywhere, in real-time. 

The result is a widening gap between AI ambition and AI delivery. Organizations can buy compute and can license models, but without stronger, more trusted connectivity between data, compute and users, AI stays impressive in demos and frustrating in real life.

If this new status quo is not built into networks now, the supercycle does not just slow down, it stalls.
 

Connectivity doesn’t scale on innovation alone. It scales on certainty and trust; certainty that network upgrades won’t be trapped in permitting limbo, and trust that the infrastructure being deployed is secure, accountable and resilient.

That’s where policy matters most. 

Washington can’t create the ecosystem and build the foundation to decide whether the U.S. builds AI on trust. The job is twofold: make it faster to modernize networks and make “trusted connectivity” a clear end-to-end standard, not a label reserved for one visible layer.

And the risk isn’t theoretical. 

Nokia’s research shows that when constraints persist, companies shift behaviour. Many invest in private infrastructure, some narrow to only the highest-ROI AI use cases, and some move workloads to wherever infrastructure is stronger and more predictable. 

That is how leadership erodes quietly, not with a single failure, but with lots of individual decisions that add up to a shift in where AI innovation, jobs and capital concentrate.

If the US wants to stay ahead, it must do two things at once:

• Build the connectivity capacity that AI demands, across fiber, backbone, optical backhaul and data-center interconnect, engineered for distributed, east–west AI traffic and real-time performance expectations.
• Ensure the vendors building it are trusted, secure and accountable across the entire stack, not just the parts that are most visible.

In the AI era, the fastest way to sabotage innovation is to underbuild the network, or to build it on components you cannot trust.