Executive Summary

• This article is adapted from our fireside chat, “Energy, AI and the Hyperscaler Playbook”, featuring Dr Sayuri Moodliar, a sustainability consultant and infrastructure project finance lawyer. The session was a core part of our virtual summit, “DCi Horizons: Building the AI-Ready Data Centre.”
• Large-scale AI infrastructure can be deployed in months, but expanding utility grid infrastructure needs 10 to 15 years to obtain permits and build.
• Major hyperscale operators are imposing legally blinding, strict supply chain codes of conduct; Microsoft now mandates annual third-party audited emissions tracking, a 55% reduction by 2030 and a full transition to carbon-free electricity.
• Investors and clients are shifting to use CUE, WUE and PUE combined for real-time data tracking.

The key to winning hyperscale contracts is sustainability; it’s not a marketing checkbox anymore, it’s actually a hard commercial barrier for companies wanting to work with hyperscalers. The explosion of AI has destroyed and is reshaping traditional data centre forecasting. Sayuri mentioned AI workloads are completely different to traditional cloud compute:

“Training large models and running inference at scale drives fairly high power densities compared to conventional IT. And this obviously forces us to rethink our traditional facility design. And we’ve also noticed that AI-driven demand is compressing timelines dramatically. So what used to be a 10-year energy demand forecast is now being consumed in two to three years.”

The permitting mismatch

The rapid scaling of AI as it moves to gigawatts and multi-data centre campuses has created a construction mismatch, with slow multi-year timelines for building and upgrading the electrical grids. AI infrastructure can be deployed in a matter of months, but then the utility grid infrastructure expansion needs 10 to 15 years just to get permits and build. Grid congestion is rife in the key tech hubs, which is making the authorities nervous, leading to restrictions in new data centre developments in areas across the US and Europe, like Dublin and Amsterdam.

Hyperscaler playbook

Companies like Microsoft, Google, Amazon and Meta are using direct PPAs (power purchase agreements) to power their new AI facilities and bypass the grid limitations. However, the burden of proof does fall on colo operators when they sign hyperscale agreements; there’s a multi-layered trap when those contracts don’t factor in the enormous operational costs of compliance with the hyperscaler’s strict requirements.

Sayuri details the three commitments needed from a colo operator to win a hyperscale contract:

• Annual audit costs: “…you are required to measure and disclose annually detailed greenhouse gas emissions data and calculations. And this must be verified through third-party assurance, in other words, audited at your own cost.”
• The 55% scopes reduction: “…you also need to commit to a minimum 55% reduction in your greenhouse gas emissions across all scopes, scope one, scope two and scope three of your emissions, from your baseline year to 2030.”
• 100% carbon-free power integration: “…if you’re a large-scale supplier, the third requirement is you are required to adopt 100% carbon-free electricity by 2030 for any products and services provided to Microsoft.”

The Investor Ecosystem

In the past, ESG was almost a tick-box exercise and that was that; investors and lenders have moved well beyond that now and there are frameworks for green loans, sustainability-linked financing so the industry improves its ESG to gain access to more funding. There’s also covenant structures (agreements) tied to PUE ratings – so if you were to acquire a loan, you’ll get a reduced interest rate if you reduce your PUE or increase the percentage of renewable energy you use.

“Investors want to see board-level accountability recently, not just an ESG report – a glossy report produced by a team,” states Sayuri.

The evolution of operational metrics

PUE has been a great metric, but we need more than that now; liquid cooling and compute demands need far more than PUE to measure efficiency, which is why there’s a shift to real-time data capture. PUE is still useful, but the industry long ago recognised it’s not good enough as a stand-alone metric – it certainly isn’t appropriate as a measure of clean energy or sustainability – so the industry is moving toward using more metrics alongside PUE, such as CUE and WUE.

Global geopolitics vs UK reality

In global markets, the US is creating momentum and is moving fastest in terms of the sheer scale, but it’s also running into the same constraints as the other markets. There are interesting developments over in the Nordics, particularly Sweden and Norway, as they’re rather attractive for their abundance of land, renewable energy and their cool climate, not to mention the use of waste heat for efficiency metrics. Singapore and Southeast Asia are reopening since the moratorium was imposed, with the Middle East also deploying infrastructure. Australia and Japan are becoming incredibly important markets for AI infrastructure.

For all developed markets, we’re seeing speed to market and rapid growth, and if we look East, Malaysia and Indonesia are attracting investment and sub-Saharan Africa is emerging as an attractive location for infrastructure. In sub-Saharan Africa, the regulatory landscape is still in development, which makes it incredibly appealing to companies wanting to expand into that market.

Across all markets, there are concerns of water stress and grid carbon intensity; both material concerns that responsible operators must address. In every market, you now have to deliver and fortunately, we’re in a highly innovative era where solutions to the bottlenecks will be created.

The UK’s reality is that the National Grid’s connection queue stretches into mid 2030 and beyond in many areas and comes with a backlog reform process that’s underway but still doesn’t unlock the power bottleneck at a pace needed by the market. Grid carbon intensity has been falling over the last decade in the UK because they have strong renewable energy alternatives. However, the UK still faces other issues, such as planning issues, competition for land and community resistance in some locations; the planning system hasn’t fully adapted to speed adn scale the sector needs to move at. There’s scope for the UK to position itself as a well-governed market for AI infrastructure if it leans into renewable credentials, regulatory clarity and financial centre status, but that needs coordinated action between the government, grid, planning and the industry itself.