Power, connectivity and the next phase of the AI supercycle

Thirty years ago, the internet was dismissed by some as little more than a super-charged fax machine.

Three decades and trillions of dollars of value later, we know that 'today-forward' thinking does not work for technology supercycles. They arrive in waves and reshape everything they touch.

We risk making the same mistake with the AI supercycle.

Impressive as today’s AI agents are, with a Massachusetts Institute of Technology (MIT) study showing productivity gains of 60%, they are not the endpoint of the AI supercycle.

The next frontier moves AI from our laptops and phones into the real world. We already see early signals of that future with Waymo's robotaxis, Amazon's industrial robots and Meta and Google's innovations in smart glasses.

From an infrastructure perspective, this shift demands a 'future-back' view of what AI will require next. One thing is clear: AI is outgrowing the infrastructure built for the internet in terms of compute, connectivity and power.

The pressure shows up first in AI factories, the data centres where models train and deploy. These facilities continue to grow larger and denser, based on the assumption that concentrating compute in a small number of locations remains viable.

Power challenges that assumption. AI workloads are already consuming tens of gigawatts globally, according to industry experts. By the end of the decade, some estimates suggest consumption could approach hundreds of gigawatts, pushing against the limits of the existing power infrastructure. Grid connections take years to secure. Critical electrical components face long lead times. Scaling yesterday’s data centre model forward will not keep pace.

This is more than a power problem. It is an architectural one.

The industry often asks how to build bigger data centres. The more important question is how to build different systems. AI requires architectures that place compute where power exists and connects them so they operate as one.

Connectivity plays a central role. High-performance optical and IP networks allow AI workloads to disaggregate, synchronize and move efficiently across locations. They turn distributed capacity into usable infrastructure.

The next phase of AI depends less on raw compute capacity and more on how efficiently intelligence moves.

Research from Bell Labs points to high-teens annual growth in inter-data-centre and metro traffic over the next decade. This reflects a structural shift towards distributed AI architecture, rather than a short-term surge.

Training increasingly spans multiple sites. Inference moves closer to users. Metro and regional locations become active parts of the AI system, rather than simple access points.

In this model, networks stop acting as passive transport layers. They form the intelligence fabric that synchronizes compute, data and decisions across core, metro and edge environments.

Connectivity no longer just moves data. It determines where intelligence can exist.

This next phase is often described as physical AI — intelligence embedded in machines, infrastructure and systems that interact directly with the real world, where decisions occur in fractions of a second and errors can be catastrophic.

Physical AI changes what networks are required to do. Networks evolved around human behaviour. Traffic followed predictable patterns, flowed mostly downstream and tolerated delay.

AI-native traffic behaves differently. It varies more widely, pushes more data upstream and demands strict latency and reliability.

Today, only a small fraction of mobile network traffic qualifies as truly AI-native. As that rises, network requirements change fundamentally.

Latency must remain deterministic. Reliability must be engineered. Availability approaches what is known in the industry as 'six nines' – near-perfect system reliability levels, guaranteeing 99.9999% uptime – because physical systems depend on consistent outcomes.

These demands require AI-native networks. Software-driven platforms observe conditions, anticipate demand, optimize performance and secure themselves in real time. Ultimately, more of the value in connectivity will come from the intelligence embedded within the network itself.

This shift fundamentally reshapes how critical systems are built and run.

Factories, transportation systems, energy grids, hospitals, public safety networks and defence environments all depend on connectivity that is predictable, resilient and trusted under extreme conditions.

As AI shapes physical outcomes, network integrity becomes part of the safety system itself — shaping reliability, resilience and national security.

Trust starts with infrastructure that performs predictably, protects data and remains secure under all conditions. It also depends on providers that governments, industries and partners can rely on for critical networks. In the AI era, trust determines whether intelligence can operate safely at scale.

AI workloads demand seamless and persistent connectivity. Infrastructure must support intelligence moving reliably from data centre to edge to device, across enterprise, mobile and satellite networks and across partners and borders. Interoperability enables scale, resilience and speed. Open, standards-based systems reduce dependency and allow intelligence to move where power, data and demand exist. In a more contested world, interoperability becomes a source of strength.

No company or country can build the AI future alone. Progress depends on collaboration and co-innovation across cloud, silicon, software, communications and industry ecosystems. Leadership in the AI era means convening partners and driving from innovation into deployment rapidly. Those who bring ecosystems together will shape how AI drives growth, security, and long-term competitiveness.

The internet supercycle reshaped the global economy by connecting people. The AI supercycle will reshape it again by connecting intelligence across industries and the physical world.

The question now is not whether AI can deliver transformative productivity gains; it is how quickly those gains arrive and how broadly societies can capture them.

Infrastructure will determine the answer. Trusted, interoperable connectivity will shape where intelligence flows, how safely it operates and how widely economies benefit.

The task ahead is clear. We need infrastructure that moves intelligence securely, efficiently, and reliably to wherever it is needed. When we get this right, AI becomes a durable driver of productivity, competitiveness and long-term economic progress.