AI Doesn’t Need More Mega Data Centres. It Needs a Smarter Network.
Artificial intelligence has sparked one of the largest infrastructure races in modern history.
Every month brings announcements of new hyperscale data centres, facilities that cost billions of dollars, consume enormous amounts of electricity, and require years to build.
The prevailing assumption is simple, more AI requires more data centres.
But what if that’s the wrong assumption?
According to the International Energy Agency (IEA), data centres consumed approximately 415 terawatt-hours (TWh) of electricity in 2024, around 1.5% of global electricity demand. By 2030, that figure is expected to more than double to approximately 945 TWh, driven primarily by AI workloads. That’s roughly equivalent to Japan’s current annual electricity consumption. (IEA)
The world already owns millions of powerful GPUs sitting inside gaming PCs, creator workstations, and developer machines. Many of these systems spend a significant portion of their lives underutilised, despite being more than capable of serving AI inference workloads.
Instead of building another warehouse full of hardware, perhaps the future lies in connecting the hardware we already have.
That is the philosophy behind distributed AI infrastructure.
The World’s Largest Untapped Compute Resource
Every day, millions of computers around the world sit idle or significantly underutilised.
Gaming PCs are idle while their owners are at work.
Creative workstations wait overnight.
Development machines spend hours doing nothing between workloads.
Individually, these computers don’t look significant.
Collectively, they represent one of the largest untapped compute resources ever assembled.
Distributed AI networks transform those independent machines into a global inference network capable of serving real AI workloads.
Rather than concentrating compute inside a handful of buildings, the infrastructure exists wherever people already own capable hardware.
The Problem With Building Bigger
Hyperscale data centres solve one problem while creating several others.
They require vast amounts of land.
They place enormous pressure on electrical grids.
They consume significant resources for cooling.
They concentrate economic value into a relatively small number of infrastructure owners.
The challenge is becoming increasingly local. While data centres currently account for a relatively small share of global electricity consumption, the IEA notes that they are often heavily concentrated in specific regions, creating significant pressure on local power grids and infrastructure. Modern hyperscale AI data centres commonly require around 100 MW of capacity, while some projects currently under development are planned to consume as much as 2,000 MW, with proposed future developments reaching 5,000 MW. (IEA)
Communities often experience increased demand on local infrastructure while seeing only a fraction of the long-term financial benefit.
As AI adoption accelerates, this centralized model becomes increasingly difficult to scale sustainably.
A distributed approach asks a different question.
How much existing infrastructure are we failing to use?
Community-Owned Infrastructure
One of the biggest advantages of distributed computing isn’t technical.
It’s economic.
Instead of AI infrastructure belonging exclusively to large cloud providers, anyone with compatible hardware can participate.
The owner of a gaming PC.
A freelancer with a high-end workstation.
A small business with idle GPUs after business hours.
These people become infrastructure providers instead of passive consumers.
AI becomes something communities help power, not simply something they purchase.
Demand Is the Missing Piece
Historically, decentralized compute platforms have struggled with one major challenge.
Get Mohammed El Hassan’s stories in your inbox
Join Medium for free to get updates from this writer.
Supply existed.
Demand did not.
People could contribute hardware, but finding customers willing to run workloads on those machines was difficult.
That’s where NexusCompute takes a different approach.
Rather than waiting for developers to discover individual providers, NexusCompute connects contributors directly to existing AI inference demand from platforms such as OpenRouter and Together AI. Providers can install the lightweight agent, join the network, and begin receiving real inference requests without needing to market their own hardware or negotiate with customers.
That changes distributed computing from a speculative marketplace into infrastructure that begins working immediately.
Developers Benefit Too
Distributed infrastructure isn’t only valuable for hardware owners.
Developers gain access to a decentralized pool of GPU resources through an OpenAI-compatible API, allowing existing applications to connect with minimal code changes. Instead of long-term commitments or minimum capacity requirements, workloads can be routed across available providers with transparent, per-token pricing.
This creates greater flexibility while reducing dependence on a small number of centralized cloud providers.
Resilience Through Distribution
Centralized infrastructure has centralized risks.
Power failures.
Regional outages.
Network disruptions.
Supply shortages.
Distributed infrastructure naturally spreads those risks across thousands of independent nodes.
If one provider goes offline, another can take its place.
This is the same architectural principle that made the internet resilient in the first place.
Instead of depending on a handful of massive facilities, capacity exists everywhere.
Sustainability Through Better Utilisation
Perhaps the greatest opportunity isn’t replacing data centres entirely.
It’s reducing the need to build so many new ones.
Every GPU already sitting inside a home or office represents embodied manufacturing, existing electrical infrastructure, and installed computing capacity.
Making better use of those resources is often more sustainable than constructing entirely new facilities.
As electricity demand from AI continues to rise, improving utilisation of existing hardware becomes just as important as deploying new hardware. Smarter use of existing GPUs can reduce pressure on constrained electricity grids, delay unnecessary infrastructure expansion, and allow AI capacity to grow more efficiently. (IEA)
Distributed computing isn’t about replacing hyperscale infrastructure.
It’s about ensuring new infrastructure is built only where it’s genuinely needed.
The Next Chapter of AI Infrastructure
Training frontier AI models will continue to require some of the world’s largest supercomputers.
But inference, the work of serving models to millions of users, is fundamentally different.
Inference is distributed by nature because users are distributed.
As AI becomes part of everyday life, the infrastructure supporting it should reflect that reality.
Instead of concentrating opportunity, it should distribute it.
Instead of leaving capable hardware idle, it should put it to work.
Instead of asking communities to make room for ever-larger data centres, it should allow communities to become the infrastructure themselves.
The future of AI won’t be defined solely by who builds the biggest data centre.
It will also be shaped by who builds the smartest network.
If global data centre electricity demand is expected to more than double before the end of the decade, building larger facilities alone cannot be the only solution. The future of AI infrastructure must also focus on making dramatically better use of the computing resources that already exist. (IEA)
At NexusCompute, that’s the future we’re building, a decentralized AI infrastructure where existing hardware becomes productive, developers gain flexible access to compute, and individuals can participate directly in powering the next generation of artificial intelligence.
