Stack talks signal AI data centers’ move into core infrastructure

From growth story to infrastructure asset class

Data centers have long sat between real estate, technology and infrastructure. AI is tilting the balance toward infrastructure.

Training and inference workloads require dense, energy-intensive campuses with large upfront capital commitments and long lead times. In the largest projects, the bottleneck is not simply servers. It is power availability, permitting, equipment procurement, construction sequencing and the ability to match capacity with multi-year customer commitments.

That is why the reported Stack valuation range matters. A $20 billion to $25 billion transaction would imply that institutional buyers are willing to underwrite data-center platforms at a scale typically reserved for essential infrastructure networks.¹

The appeal is straightforward. If AI and cloud demand continue to compound, owners of powered, developable and leased capacity may occupy a strategic position similar to owners of ports, pipelines, towers or renewable-power platforms.

The Asia-Pacific angle adds another layer. Regional cloud and AI adoption are expanding, but power, land and regulatory constraints vary sharply by market.

A portfolio with operating assets and development optionality can offer institutional capital diversified exposure without requiring buyers to assemble sites one by one. It also gives global infrastructure investors a foothold in a market where hyperscalers, sovereign capital and local operators are competing for the same scarce inputs.

Capital is following contracted demand

Recent AI infrastructure deals show why the underwriting model is shifting. Akamai disclosed in a September 24 SEC filing that it entered a material cloud-infrastructure agreement with Anthropic involving an $11.6 billion, seven-year commitment and a warrant structure.⁶

In a related announcement, Akamai said Anthropic would use Akamai Cloud infrastructure to support AI workload growth and that associated capital expenditures were estimated at $5.5 billion.⁷ Akamai’s investor materials described the relationship in infrastructure-like terms, including financial phasing, cumulative capex, revenue timing and megawatt requirements.⁸

For dealmakers, the message is that AI demand is becoming more contractible. Long-term customer commitments can support debt, lower the perceived risk of new capacity and make data-center cash flows legible to infrastructure funds, insurers, pension plans and bank syndicates.

The technology may be novel, but the financing language is familiar: contracted revenue, utilization risk, capex phasing, counterparty exposure, residual value and power availability.

That does not eliminate risk. It changes where risk is allocated. Investors are no longer underwriting only secular data growth. They are underwriting whether AI customers can honor multi-year obligations, whether developers can deliver campuses on time, whether utilities can provide enough power and whether financing structures can withstand delays.

Project risk is becoming financing risk

The same week as the Stack report, Reuters reported that a delay tied to an Oracle, Blue Owl and Stack-linked AI data-center project was affecting lender and investor appetite.³ Bloomberg separately reported that Oracle cited force majeure in connection with Project Jupiter, a large AI data-center campus involving Stack and Blue Owl, highlighting issues around leases, power and project finance.⁴

Those reports show the other side of the infrastructure analogy. Large AI campuses behave like infrastructure projects not only because they require long-duration capital, but also because delays can ripple through complex capital stacks.

If a power connection, construction timeline or customer occupancy date slips, the effects can reach lenders, landlords, developers, tenants and equity sponsors.

In traditional infrastructure, this is managed through detailed risk allocation: construction guarantees, availability milestones, liquidated damages, power-purchase arrangements, reserve accounts and covenant packages.

AI data centers are moving in the same direction. The more capital-intensive the campuses become, the more financing will depend on documentation that can allocate delay risk among customers, operators, sponsors and creditors.

Debt markets are scaling around digital infrastructure

Institutional equity is only one part of the shift. Debt markets are also building capacity for AI-era data-center growth.

VIRTUS Data Centres announced a £2.45 billion financing package to support continued expansion, including an AI-ready campus, with a 13-bank consortium backing the package.⁵ That type of multi-bank structure reflects the scale of capital now required to build digital infrastructure in Europe and other power-constrained markets.

For lenders, data centers offer potentially attractive infrastructure-credit characteristics: hard assets, repeat enterprise demand and the possibility of long-term contracted cash flows.

But AI also creates concentration and execution challenges. Campuses may depend on a small number of very large customers, high-density equipment specifications can change quickly, and power procurement can become the central credit variable.

That means debt terms are likely to become more sophisticated. Facilities may increasingly differentiate among stabilized colocation assets, pre-leased hyperscale buildings, speculative powered shells and early-stage land banks.

Lenders will also focus on how much of the revenue base is supported by investment-grade counterparties or well-capitalized AI companies, and whether customer contracts adequately cover the capital being deployed.

Ownership is moving toward platforms, not assets

The Stack process also reinforces a platform premium. In a market defined by speed, scale and scarcity, buyers may value management teams, customer relationships, power pipelines and development capabilities as much as existing buildings.

Institutional investors are not merely buying racks and roofs. They are buying the ability to originate, permit, finance and deliver future capacity.

That favors large infrastructure managers. Firms such as BlackRock-backed infrastructure platforms and IFM can aggregate capital across funds, co-investors and strategic partners.

They can also hold assets for longer periods than many real estate or growth-equity strategies, which matters when development pipelines extend across multiple years and returns depend on compounding capacity additions.

The implication for technology dealmakers is that ownership of AI infrastructure may consolidate around investors with the lowest cost of capital and the strongest tolerance for construction complexity.

For corporates, that could mean more opportunities to secure capacity through long-term agreements rather than owning every asset directly. For sponsors, it raises the bar: access to power, financing certainty and execution credibility may matter more than headline AI exposure.

A core-infrastructure market with technology-cycle risk

The central tension is that AI data centers combine infrastructure economics with technology-cycle uncertainty. Demand appears deep, but workload mix, chip efficiency, model architecture and customer concentration can change quickly.

Infrastructure investors typically prefer stable usage patterns; AI is still evolving. That makes contract quality, renewal assumptions and residual use cases critical.

Still, the direction of travel is clear. The reported Stack negotiations, Akamai’s multi-year Anthropic commitment, VIRTUS’s bank financing and the financing scrutiny around Project Jupiter all point to the same conclusion: AI capacity is becoming a corporate infrastructure race financed by institutional capital.¹⁵⁶

If the Stack transaction closes near the reported valuation range, it would be more than a large data-center deal. It would signal that AI infrastructure has entered the same capital markets as energy, transport and communications networks.

The winners may be those that can turn volatile demand for compute into investable, contracted and powered capacity at global scale.