This article summarises the Forvis Mazars webinar Investing in Data Centers, hosted with Aurora Energy Research.
Neethu Ram moderated, with panellists Anton Yunussov (opens in a new tab) (Cybersecurity Practice Lead, UK), Alice Strevens (opens in a new tab) (Director, Sustainability Consulting), Marten Ford (opens in a new tab) (Project Leader, Aurora Energy Research) and Ryan Humphrey (opens in a new tab) (Partner, Financial Modelling).
Data centres have moved well beyond their origins as tech infrastructure. They now sit firmly in infrastructure investment, drawing capital from institutional funds, development finance institutions and project finance lenders alongside the hyperscalers and colocation operators that have traditionally driven the sector.
The scale of the shift is clear. Today’s connected world generates roughly 3 trillion gigabytes of data through smartphones, digital platforms and fintech applications. Add AI workloads and demand is expected to nearly triple in the next five years. For investors, the question now is how to manage the risks of an asset class that is changing faster than the infrastructure around it.
How AI and cloud adoption are driving data centre demand
Three forces drive the growth in data centre demand. The first and most visible is the rapid growth of artificial intelligence workloads. Training large language models, running advanced analytics and delivering AI‑powered services all need immense processing power. AI workloads are expected to account for 50% of cloud demand in the coming years.
The second is the broader wave of cloud adoption and digital transformation. Organisations in every sector are moving to cloud infrastructure, and that move needs compute capacity at scale.
The third is regulation. Data sovereignty requirements under frameworks such as GDPR are driving demand for local data centre capacity, so operators cannot simply consolidate into a handful of global locations.
“AI training, large language models, advanced analytics require immense processing power. That’s the reason why we’re seeing the data centre space exploding in terms of demand.”
As Marten Ford observed, what matters most is that these use cases are now being monetised and turned into investable, long‑term revenue streams. That is what drives some of these very large pieces of infrastructure.
Where capital is flowing in global data centre investment
North America and Europe remain the mature markets for data centre investment, but the pipeline is global. As Ryan Humphrey noted, our team has worked on data centre transactions in the US, Europe, Africa, APAC and Latin America.
Within those regions, where to build is increasingly decided by access to power. The traditional European clusters, often called the FLAP‑D markets (Frankfurt, London, Amsterdam, Paris and Dublin), lead the way but are increasingly constrained by grid capacity. Developers are moving further afield to find the power connections they need to deploy quickly.
“Having easy access to power means that you can build and deploy rapidly and meet that exponential growth in demand for AI services as quickly as possible.”
The investment model has also shifted. What started as venture capital and equity investment now sits firmly in infrastructure funds, which has turned data centres into real estate assets rather than just tech hubs. Finding land or capital is no longer the constraint. The real constraint is power generation and, more specifically, power transmission.
Energy strategy and grid constraints for data centre projects
Power is the largest operational cost for a data centre. As Ryan Humphrey put it bluntly, roughly half of the opex is power, probably more. That makes energy strategy central to the investment case.
Marten Ford set out three dimensions of the energy challenge. The first is access to power, which in most markets means securing a grid connection. The second is the energy procurement strategy, which largely decides long‑term operating costs. The third is managing the wider effects a data centre has on the grid around it.
The grid challenge is particularly acute because data centres are a step change in demand. In many Western countries, overall power demand has been falling for decades, and grids have not had to accommodate new power‑intensive industries at this scale. Networks need to reorient themselves to support data centre connections, and that takes time.
“Transmission is the key short‑term issue. That is what is constraining projects today.”
In the medium term, coordinating government‑led investment in generation with market‑driven data centre demand becomes critical. Operators need to engage with policymakers early, not wait for the grid to catch up.
Practical responses include behind‑the‑meter power, bridging power while grid upgrades are completed and flexibility in power demand. New forms of power purchase agreement, such as a recently announced deal in the UK between a technology company and an energy company, are early examples of operators solving these problems commercially.
The question of whether nuclear power could play a meaningful role drew a cautious response. Extending the life of existing nuclear assets through PPAs, as in several US transactions, is a near‑term opportunity. New nuclear build, including small modular reactors, is a medium‑term aspiration that needs to show it can be delivered before investors place much weight on it.
For modellers working through these energy assumptions, our tutorials on CFADS and the features of a cash flow waterfall in project finance cover how power costs flow into the cash flow metrics lenders and investors test.
Cybersecurity risks investors must assess before buying data centres
Data centres are high‑value targets for cyber attackers, and the risk rises during acquisitions and expansions. Anton Yunussov set out three categories of risk investors need to understand.
The first is resilience. Data centres face sustained denial‑of‑service attacks, and withstanding them is a basic requirement. The second is hidden vulnerabilities. When acquiring or merging data centre operations, unknown security gaps must be found and fixed before the deal closes. The third is the larger attack surface that comes from integrating IT systems from different environments, which adds complexity and more ways in for attackers.
Anton Yunussov’s advice to buyers: carry out in‑depth cyber due diligence before the deal closes, so that any hidden vulnerabilities, resilience gaps and compliance gaps are known before the deal goes through.
Pre‑deal cyber due diligence usually has two workstreams: a cyber risk assessment against the industry standards that apply to data centres, and penetration testing, often including red team exercises that mimic real attacker techniques to test controls in practice.
Beyond the deal itself, operators face continuing challenges from technology obsolescence, talent shortages and regulatory compliance. The EU’s NIS2 rules now class data centres as critical infrastructure providers, which means continuous evidence of compliance.
ESG and sustainability considerations for data centre investments
ESG regulation is in flux. Deregulation is on the cards across parts of Europe and at US federal level, while sustainability requirements are tightening at state and national level. This creates uncertainty, but it does not reduce the importance of building sustainability into investment decisions from the outset.
As Alice Strevens noted, if you focus only on minimum compliance, you may not build sustainability principles into planning and investment decisions, and then you may not be able to safeguard your investments for the longer term.
Data centres are among the most resource‑intensive asset classes. Investors increasingly look for low carbon footprints, renewable energy where possible, efficient water use and transparent reporting. ESG is moving from voluntary to mandatory, driven by local compliance requirements and disclosures.
Social licence to operate is an emerging risk that deserves more attention than it usually gets. The Dublin moratorium on new data centre grid connections showed what happens when data centre growth crowds out other uses of grid capacity, including housing. Communities may expect broad economic benefits from a data centre, only to find that once it is running it creates relatively few local jobs.
“Engaging with communities to understand what’s needed within that local context could help to enhance your local community engagement and help you tackle risks with communities as they arise.”
Practical community initiatives include funding local skills and training, improving broadband for local institutions and reusing waste heat for nearby homes.
On greenwashing, investors should question claims of carbon neutrality or 100% renewable energy that lack third‑party audits. Check that certifications such as LEED, BREEAM or ISO 14001 are genuine, and treat heavy reliance on carbon offsets as a red flag.
Financial modelling challenges for data centre transactions
From a financial modelling perspective, data centre transactions mix familiar infrastructure modelling challenges with sector‑specific ones. Ryan Humphrey observed that many of the financial risks are the same as in other asset classes: uncontracted revenue is riskier than contracted revenue, construction costs are significant and supply chain risks are real.
The sector‑specific elements, such as power usage effectiveness (PUE), power cost modelling and churn assumptions, often concern investors most. But they tend to be straightforward to model: in effect, price times volume.
“Where we’ve seen issues and actually found problems in models is because they haven’t been planned and designed well, or they’ve tried to use a previous model as their business has evolved.”
The more common problems the team finds are structural. Aggregating several data centres in a portfolio, splitting a single site by service type and handling different revenue mechanics across sites all add complexity. Each is well understood on its own, but together they become challenging, particularly for investors making their first data centre transaction.
The recommendation is to spend more time on model planning and design. Ryan Humphrey suggested that the usual guidance of spending 10 to 20% of the time planning a model should be higher for data centres, because the calculations are not the hard part. The structure is. For teams exploring how AI tools can support model review and documentation, our AI in Project Finance Modelling course covers practical ways to use AI across the modelling lifecycle, from data extraction and integrity checks to model documentation.
Flexibility matters just as much. Many operators use the same model across several transactions, moving from equity‑funded greenfield builds to refinancings as cash flows mature. Building in the ability to break down and restructure the model from the start avoids retrofitting a model that was not designed for the next stage. Our Financial Modelling for Digital Infrastructure Projects course covers these structural challenges: portfolio aggregation, revenue mechanic design and the flexibility needed as data centre businesses move through successive financing stages.
Data centres also bring some distinctive sensitivities. Run‑off cases, which assume no customer growth and test existing cash flows adjusted for churn, are a relatively new type of sensitivity the team now sees more often. Construction delay and capex sensitivities remain relevant for greenfield projects, as for any infrastructure rollout.
Our tutorials on average DSCR, the loan life coverage ratio (LLCR) and cash sweep analysis cover the credit metrics and debt mechanics lenders apply across infrastructure asset classes, data centres included.
The bubble question
The question came from the audience, and the panel answered it directly. Ryan Humphrey’s view was that the interest is underwritten by real money from big tech, governments and lenders, while monetisable AI use cases are growing. That combination suggests it is not a bubble, even if the volume of commentary creates that impression.
Marten Ford added two qualifications. First, power availability may naturally limit the rapid growth, which could temper some of the very large proposed investment figures. Second, advances in chip design are reducing the power needed per unit of compute, so the link between data demand and physical infrastructure is not linear.
Our Best Practice Project Finance Modelling and Advanced Project Finance Modelling courses cover the financial statements, debt sizing and scenario analysis behind infrastructure transactions like these.