At Tech Show London, a panel from the US data centre ecosystem laid out how North America is rapidly rewriting the global playbook for digital infrastructure. From hyperscale AI ‘factories’ and power scarcity to regulation and culture, the discussion provided a candid, honest snapshot of what it takes to deploy AI at scale.
From warehouses of servers to AI factories
The central theme of the conversation was clear: AI isn’t just another workload; it is fundamentally transforming the design, scale, and operating model of data centres.
From Google’s perspective, successful AI deployment at scale is about meeting unprecedented demand while staying efficient, responsible, and reliable. As Anahita Mouro, Global Leader, Google Data Centre – Field & Reliability Quality, Google explained, this requires a system-level mindset.
She argued that hyperscalers have a unique advantage because they can vertically integrate “from silicon to software,” allowing them to break the old equation of more compute = more power. Google’s Ironwood GPU, she noted, is designed specifically for inference and training and is “about 30 times more optimised from the previous revision,” enabling far greater training output per unit of power.
The move to AI factories demands not just smarter chips, but a holistic rethink of the entire stack: integrated building systems, modular and resilient supply chains, and an ecosystem approach where product portfolios are treated as “systems of systems.”
Power: from consumer to co‑developer
If AI is the engine of this new era, power is the fuel – and it’s increasingly scarce. Panellists repeatedly returned to the reality that the grid, as it stands, cannot keep up.
Jeff Ivey, VP Data Centre Construction, Crusoe framed his company’s challenge bluntly: it’s about “gigawatts of scale and gigawatts of speed.” The industry can no longer assume that utilities will simply deliver the power when and where it’s needed.
Mouro echoed this from the hyperscaler side: utilities alone can’t support the growth curve of AI. Operators are evolving from data centre builders into power infrastructure co‑developers, investing in renewables, grid modernisation, and protection of the grid itself.
This marks a sharp shift in industry identity. As one panellist noted, 20 years ago, you could “find a corner, slap a data centre up, the power would just show up.” That era is over.
Scaling fast
Buddy Rizer, Executive Director, Loudoun County Department of Economic Development, highlighted how quickly models and requirements are changing. Developers file a site plan, then come back two months later with a completely new design as their AI and Cloud strategies evolve.
Ivey illustrated this with a single building that grew from 200 to 345MW in less than two months – a 70% increase in planned capacity before construction was even complete. In one multi‑building project, he noted, every single building now has a different design because chips, cooling requirements, and densities are changing so quickly.
For Mouro, the real problem is that the rate of obsolescence is outpacing the rate of construction. By the time a facility is built, elements of it are already outdated. Her answer is not to freeze the design, but to engineer for resilience and rapid iteration: “Most of the focus should be really on, how can you shorten the duration from design change or innovation launch to deployment?”
Standardisation, modular building, and decoupling innovation cycles (e.g., not redesigning foundations every time chips change) are key levers to stay in the game.
Open collaboration: standardising the AI era
The Open Compute Project (OCP) plays a central role in this story. Cliff Grossner, its Chief Innovation Officer, described OCP as a nonprofit foundation that gives companies a safe, legally structured environment to solve shared problems and standardise where it doesn’t differentiate their business: “Our job is to provide a collaborative environment where our members can come together and solve their shared problems for the benefit of everyone, accelerating the market.”
OCP’s work on common rack designs in the Cloud era took about seven years to converge, but it delivered substantial efficiencies. Now, with AI, things have “come a little bit undone” as many rushed to bespoke solutions. The foundation is pushing back towards common AI data centre reference designs – both inside the data hall and in the power architecture outside the hall, including moves such as bringing 800V AC to the racks.
A recent IDC study commissioned by OCP estimates that by 2029, $295 billion of equipment will be purchased and deployed – evidence of just how central open standards are becoming in this space.
Regulation
The panel also explored how policy and regulation are struggling to keep pace. Rizer underscored that the bottleneck is increasingly on the energy side: there is “no fast path” today for new nuclear or major new power plants, even as AI drives electricity demand from flat to “hockey stick” growth.
Adrian Mountstephens, Strategic Business Development, Banking, Equinix, argued that, globally, widespread regulation of data centres is inevitable: “Ultimately, globally, the data centre industry is going to end up being regulated completely in all jurisdictions … [it is] just too important.”
He pointed to the EU’s DORA (Digital Operational Resilience Act) as a sign of what’s coming: not just financial institutions, but major data centre providers supporting them are now under direct financial regulation, treated alongside banks and payment systems.
This regulatory shift is inseparable from public awareness. Governments and citizens now understand that “what runs in data centres is our whole lives,” making digital infrastructure a matter of national and societal resilience.
Community and culture
Beyond technology and power, the panel returned repeatedly to people – both inside and outside the data centre.
On the community side, Rizer’s advice was succinct: “Communicate early, communicate often, be fact based, and be responsible.”
Others stressed that data centre operators must ensure their presence is a net positive for local communities, not just in tax revenue, but in jobs, education, and infrastructure.
Mouro highlighted a glaring gender gap – around 90% of the workforce in North America and Europe is still male – and the broader shortage of skilled talent. Programmes like women‑in‑construction initiatives, paid bootcamps in under‑represented communities, and youth outreach via organisations such as Nomad Futurist are attempts to close what one estimate pegs as a 400,000‑person workforce deficit by 2030.
“At the end of the day, at least so far, data centres are designed to be operated by humans,” noted Mouro.
The message was clear: the AI era is as much a human and cultural challenge as it is a technical and financial one.
