Data Centre Business and Finance Models - Evolution and Revolution

After two years of almost weekly ‘$10bn  - $30bn funding and project announcements from the usual sources, alongside Government ‘mine is bigger than yours’  AI plans, 2026 has seen a reality check. Financing instruments are  being examined in detail at the same time when execution is key and delivery risks are becoming carefully monitored. 

Make no mistake, data centres are still the hottest commercial property deal in town. But there has been a realisation, especially among those with little or no experience of data centre development, that this a complex industry, made more complex by a variety of factors: the rapid scale that has thrust the sector into the public eye, an AI revolution road map where few know what’s over the horizon and a traditional power industry landscape that is hardly fit for delivering multiple GW expansion in the necessary timeframes. 

From financing to property to power – the questions build

How is the AI infrastructure being financed? (Now a major discussion topic within and beyond the industry.) 
Has the industry shifted from being property centric to power centric? 
Has the ability to finance and build sequencing changed? 
How are market foundations and fundamentals being viewed? 
What’s driving the demand? How stable is demand?

Are these just the growth pains of a revolution? If so how are capacity risks being measured – from the equipment OEMs making production capacity investments to the end customers who need data centre capacity to drive revenue now? 
All this alongside the difficult to quantify elements such as AI demand growth, shifting regulations, power access constraints and public acceptance. 

Not quite a Witch’s brew. But a complex picture none-the-less. And one in which finance continues to surge.

Financial Structures

As recent fund news shows debt and capital continues to flow. Vantage Data Centers: Secured $17.7 billion in funding from private investors. DayOne Data Centers: Raised $4.5 billion in a Series C round. T5 Data Centers: Raised a full disclosed $8.0 billion in a single base funding round. Blue Owl Capital & Meta: Formed a joint venture for the Hyperion AI campus, where Blue Owl will fund its share of $27 billion in development costs. Principal Financial: Closed its $3.64 billion Data Center Growth & Income Fund and announced an additional $3 billion across two new data center funds.

From where are such enormous volumes of debt and capital continuing to originate?

Financing currently comes from multiple sources:

• Commercial banks 
• Private credit funds 
• Institutional asset managers 
• Insurance companies 
• ABS (Asset Backed Securities) markets 
• Institutional loan markets 

In a market which is characterised by abundant capital competing for attractive digital infrastructure assets there is no shortage of liquidity.

However, depending on who you speak with - there are new risks.

Says Cole Crawford, long time industry stalwart and more recently founder of NativX, a compute capacity trading platform (see below.): "The top five hyperscalers are sitting on $1.6 trillion of debt off their balance sheets. Why? Because everyone feels this is a race to some end state that'll give them an edge. But it's a bit of the blind leading the blind, financially. Every frontier model release now gains maybe 10%, not the 20-30% we used to see, at $36 billion per gigawatt. Meanwhile China's GLM, DeepSeek, Kimi K2 are compressing that delta, building models 85-90% as good with $20 to $40 million training rounds.” 
What is driving the issuance? "This entire industry is currently driven by equal portions of hope / patience and FOMO as much as by fundamentals,” he says.

Integrated Power Is Creating New Financing Structures

Some see financing of combined Data Centre / Power Generation projects as the only future. 

However this brings complexities associated with mixing the risk of financing combined data centres and power generation facilities – so called  of Project on Project risk. 

Much financing activity today remains focused on greenfield hyperscale developments characterised by  construction loans ranging from $500 million to several billion dollars; Three-year build periods; Refinancing into long-term capital markets once operational as refinancing after commercial operation remains an essential part of the financing lifecycle.

The new move in data centres is AI developers building their own dedicated power generation. As data centres increasingly incorporate behind-the-meter generation, financing evolves from funding a single infrastructure asset to financing a hyperscale data centre plus an integrated power plant.

This creates an entirely new category of project finance.

This is because historically lenders financed either power plants or data centres —but rarely both together.

Now both assets are being increasingly financed within a single transaction.

This points to shifting underlying commercial models – from power taker to power maker.

Data centers as power plants — are the business models already blurring?

There is a shift in mindset: rather than treating data centers purely as power consumers, the industry (and DSOs/TSOs) is starting to explore data center operators as active grid participants — potentially even grid-balancing assets and commercial power providers.

The active conversations around hybrid investment structures have begun. Here data centre operators and utilities are discussing partnerships so a facility can function partly as a power generation or management asset that can sell back into the grid via a TSO/DSO.

Thierry Chamayou, EMEA VP for Cloud and Service Providers at Schneider Electric says that though neither an operator or an energy distributor over the last two years the company has hosted multiple forums workshops (via its "Data Center of the Future" unit) between utility innovation teams and operators, TSOs and DSOs at regional level with the full spectrum of operators — colocation players, neoclouds, real estate players  and co-developing partners including Nvidia as “stakeholders building a "virtuous" sustainability methodology that earns faster access power security.”

“Very humbly at our level, we've tried to make those worlds connect. We're talking on how we work together by looking at the DC operator as a grid regulator rather than a consumer. There is a natural convergence where one player can actually push the other to go faster – meaning trends connect and to become a bigger flow in the right direction.

“All the players are asking the same questions: How do you see [power and DC convergence] rendering the permit faster? What are the right methods?” says Chamayou.

AI and the delivery models of those involved has clearly altered the geography of data centre investment.

While not exactly swapping clothes what is clear is that utilities to talk technology – AI for energy and DCs want to talk power for AI.

Specific operator go to market

Micharl Cohen is Head of Data Centre delivery at OVHCloud a French headquartered cloud provider.  

He says just as elsewhere AI is transforming both the technical and commercial models of cloud and data centre operations. But here Cohen argues that while much of the industry has focused on power shortages and GPU demand, OVHcloud sees a greater challenge in managing uncertainty around AI infrastructure lifecycles, evolving customer demand and the rapid pace of innovation. 

As AI workloads require dramatically increased server power density this is fundamentally changing data centre design to deliver more electrical capacity within a smaller footprint making access to power the critical factor governing future expansion. 

But he says unlike many operators, having developed proprietary direct liquid cooling technology over more than two decades, OVHcloud did not need to reinvent its strategy to support AI and is positioned to accommodate higher-density computing.

While demand for GPU capacity is expected to remain strong, neither providers nor customers yet know which AI services will deliver the greatest long-term commercial value.

As a result, OVHCloud sees vertical integration as a competitive advantage and is investing across multiple layers of the AI technology stack while maintaining flexibility as customer requirements continue to evolve.

Integrating AI still requires dedicated high-density zones, revised power distribution and new engineering approaches for future facilities. Existing data centres can absorb some AI workloads in the short term, but longer-term growth will require redesigned infrastructure optimised for much higher power densities. 

The Challenge Is No Longer Infrastructure Cost

As indicated, Cohen sees infrastructure investment itself as not being the primary financial challenge. He argues that the cost of power systems, cooling and electrical infrastructure remains relatively small compared with the rapidly rising cost of GPU hardware. Instead, the greatest uncertainty lies in forecasting the useful life of AI hardware, which is becoming obsolete far more quickly than traditional servers. 

In pure ‘how infrastructure is being funded’ terms Crawford warns that the high cost of hardware and rapid GPU architecture refreshes are the major financial consideration.

He says: “The reality is an H100 is worth less today than it was a year ago, and less then than the year before that. This is a depreciating asset, full stop, yet everyone's trying to treat it like a fungible commodity. Think about this from a financing standpoint: futures markets allow an entity to build ETFs, bonds, and swaps. I don't know how you build those financial tools on the chip itself when it's depreciating, that doesn't make sense to me.”

He says today the industry finally has is a useful industry metric in the form of tokens per watt per dollar.

“In the early 2000s, if you'd asked 'performance per watt per dollar' of a SQL database or an Apache web server, nobody could answer that. Now everybody gets measured on tokens per watt per dollar. That's an actual scorecard, the first real one this industry has had."

He adds: “My advice to institutions: there's financial tooling in public markets — swaps, bonds, ETFs, indexes, forward curves — that give you real insight so you don't drive off that debt cliff. At NativX we built the NxCI (NativX Compute Index) based on a digital asset called COIL. This is based on Energy Normalised Compute Accelerated Compute Unit contracts which are fungible and can be traded. Intercontinental Exchange has licensed to build a derivatives market on top of COIL. A bank can issue swaps against it, Others can issue an ETF against it. The financial markets should use this instead of a pay-and-pray approach to lending NeoClouds money. Oil is volatile, sure, but it's not depreciating — scarcity can make it more valuable. Silicon is the opposite. You cannot build sound financial engineering around a depreciating asset. It just doesn't work."

Blurred lines

Says Alex Cordovil, research director at market analysts Dell’Oro Group, business model boundaries are dissolving and financiers have noticed.

“Traditional lines between hyperscaler, neocloud, and colo are blurring — Meta and xAI are acting as colocators; crypto miners have pivoted to neoclouds. Even hyperscalers with historically "safe" business models are under strain as their free cash flow is redirected from shareholder returns into unprecedented Capex. We may see future failures and headline-generating collapses, particularly among newer entrants who signed contracts without adequately pricing in delivery liability,” he says.

Execution capability, not capital, is now the scarce and richly priced asset

2025 was "the year of the big announcement" — headline-grabbing, unfunded-in-practice Capex pledges ($10bn here, $30bn there). Entering 2026, the market is separating genuine execution capability from paper commitments.

New entrants without deep experience (permitting nuances in specific jurisdictions, GPU cluster build complexity, power procurement) are falling behind. Investors should treat demonstrated execution track record as a key underwriting factor, not just balance sheet size.

The "tripod" investment strategy is emerging

Dell’Oro’s view is because no single element of a project (power, land, permitting, off-take) can be secured in isolation — each depends on the others being locked down first — sophisticated investors are now taking simultaneous positions across power generation, real estate/land banks, and operators. Controlling more of the stack reduces reliance on third parties and materially improves a project's odds of actually getting built.

Capital Remains Highly Competitive

Capital is flooding in, and it's chasing the whole stack, not just data centres.
In 2026 there is no sign of investor appetite waning.  For example, Partners Group recently raised roughly $15bn specifically to invest across data center operators, vendors, and adjacent infrastructure.

Despite the enormous financing requirement markets remain exceptionally competitive with large syndicated loans regularly oversubscribed. Banks are actively recycling exposure through syndication, portfolio sales, retail bank participation and refinancing into institutional markets.

Rather than a shortage of capital, lenders continue searching for additional opportunities to deploy funds.