Looking ahead: The data center may be undergoing a fundamental redesign – not through expansion, but through contraction. While global demand for computing power continues to soar, the physical footprint of the infrastructure behind it is increasingly up for debate.
For most of the last decade, data centers have been synonymous with size. These sprawling, warehouse-scale installations consume vast amounts of power to support everything from online banking to the surge of artificial intelligence processing. Nvidia's Jensen Huang has described them as "AI factories," underscoring how critical they have become in powering the machine-learning economy.
But a new generation of technologists is questioning whether bigger is always better. The emergence of micro and edge data centers – some so compact they fit under desks or inside swimming pool facilities – is reframing what counts as high-performance infrastructure.
In Devon, England, a company called Deep Green made headlines for powering a public swimming pool using excess heat from a data center the size of a washing machine. Its founder, Mark Bjornsgaard, sees the model as the future. "Small is definitely the new big," he told the BBC.
Bjornsgaard argues that every public building could host a miniature data center, creating a distributed network that shares workloads and recycles thermal output. "London is just one giant data center that hasn't been built yet," he adds.
Other innovators have taken the idea into mainstream living spaces. In late 2025, a British couple revealed that a compact server in their garden shed was heating their home. A month later, a university professor described how a GPU used for AI computation under his desk now doubles as his office heater. These examples illustrate how the physical and functional boundaries of data infrastructure are beginning to blur.
Even as hyperscalers invest billions in vast new facilities, momentum is growing for compact, localized alternatives. Telecommunications analyst Benedict Evans contends that there is "a case for smaller 'edge' data centers near large populations," as proximity can reduce latency and improve responsiveness for compute-intensive applications.
Amanda Brock, who leads the business advocacy group OpenUK, sees a broader shift coming. "The data center myth will be a bubble that will burst over time, I think," she said, noting that disused urban spaces – derelict buildings and shuttered shops – could be repurposed into small, connected data hubs.
Brock also believes that in the longer term, much of the processing currently centralized in data centers will migrate to local devices: "Processing will move to a handheld device, or a set-top box, or a router in your home."
The trend toward downsizing appears to extend beyond computing hardware to the AI models themselves. The industry's infatuation with massive, general-purpose LLMs is being tempered by a turn toward smaller, task-specific systems.
AI and climate lead Dr. Sasha Luccioni of Hugging Face observes that bespoke, locally trained models "tend to perform more accurately, and can require less computing." She adds: "We are already seeing a paradigm switch between large models taking huge resources, to smaller models being more bespoke and running more locally and tailored to business uses."
That shift also aligns with growing environmental concerns. According to Luccioni, large data centers "are taking more and more resources," making it increasingly sensible "to not use them all of the time."
Security researchers also point to resilience advantages. Alan Woodward, a professor at the University of Surrey, notes that "small targets have less impact if they are penetrated," whereas "larger centers can be big points of failure, as we've seen recently with huge AWS centers going down."
Not all future visions are earthbound. Avi Shabtai, CEO of Ramon Space, believes that "space offers a unique opportunity to rethink data structure, where small, scalable data centers in orbit can deliver efficiency, performance and flexibility." His company is developing technology to test that concept, extending the notion of distributed computing literally off-planet.