.avif)
Written by Mizuho | Greenhill's Jennifer Fritzsche, Managing Director, Technology, Media & Telecom, Investment & Corporate Banking; Stancel Riley, Head of Financial Sponsors; and Rahul Shah, Head of Project Finance
As 1980s movie icon Ferris Bueller said, “Life moves pretty fast. If you don't stop and look around once in a while, you could miss it." Wise words, indeed, but even Ferris might be surprised just how quickly life has moved since ChatGPT burst on the scene in late 2022. The chatbot’s almost instantaneous popularity helped set off the fast-moving wave of Artificial Intelligence (AI) innovation we are seeing today, driving new entrants and massive spending to build out the digital infrastructure needed to support rising demand for AI-driven products and services.
Despite recent pullback on data center projects by major players like OpenAI and Oracle, the build-out is moving full speed ahead, with no signs of slowing. In the first quarter of 2026, over 8GW were signed in North America alone. To put this in perspective, 15GW were signed in ALL of 2025. The market has seen a shift in funding amounts and where that funding comes from, perhaps the most important factor in fostering continued growth is that funding for new projects remains robust. Mizuho estimates that more than $500 billion of capital will have been raised across various markets from 2021 through 2026 to support data center buildouts.
Build It Now, We Needed It Yesterday
Supporting data center architecture will play a critical role in the proliferation of AI because the availability of computing power will ultimately influence end-user uptake and drive additional data center demand. AI hyperscalers are positioning themselves to capitalize on that demand.
Some have likened this spending zealousness to previous cycles of innovation, but the level of capex spending is unlike anything the markets have experienced before. Major players like Amazon and Alphabet are doubling down and have said they plan to deploy $200 billion and $180 billion, respectively, to AI projects this year. And they are not alone. Total 2026 capex spend reported by hyperscalers is projected to be $650 billion to $700 billion. Astronomical figures, especially compared to 2020 when AT&T was second only to the U.S. Government in capex spending at $24 billion. Think about that. A four-fold capex spending increase in just six years.
As the AI “arms race” continues to intensify, significant challenges exist. The greatest hurdle for data center projects remains power sourcing, but there are other challenges to contend with including regulatory developments, supply chain constraints and community activism. All of which impact which projects are funded.
We’ve Got the Power…We Hope
Hyperscalers’ urgency towards and need for compute is only increasing. But powering the behemoth centers required to run AI applications continues to be the dominant concern for data centers, the hyperscalers themselves and infrastructure investors. As demand grows amid supply constraints, this will become more challenging. And growing uncertainty around the war in Iran adds noise to existing power concerns.
Access to power is increasingly shaping where capital is deployed, how projects are underwritten and what financing instruments are viable. The “NeoCloud” sector, made up of AI-first cloud providers, including names such as CoreWeave and Lambda Labs, has received much more interest from hyperscalers because of these platforms’ access to power. In the last five months over $120 billion in NeoCloud partnerships have been announced. These aren’t pilot programs or letters of intent, rather long-term infrastructure commitments.
Projects that have solved their power source are naturally more attractive financing counterparties, so data center operators and hyperscalers have been driven to explore alternative energy solutions and embrace creative financing structures. To broaden the lender base and keep pricing competitive, data center platforms and the NeoClouds are utilizing high yield bonds, green loans/bonds, sustainability linked loans and asset-backed or commercial mortgage-backed securities.
Similarly, in an attempt to ease grid bottlenecks, developers have grown their use of onsite parallel power—fuel cells, hybrids, battery storage—which is becoming a key lender diligence item. While these “behind the meter” solutions typically require significant upfront investment, and most centers continue to rely on retail purchases from the existing grid; the outlook is bright. According to Bloom Energy, 38% of facilities are expected to use some onsite generation for primary power in 2030, while 27% expect to be fully powered onsite in that time period.
This focus on self-production is a key differentiator when considering whether to finance a project. Investors want to understand who is responsible for sourcing the power to run the center—the developer, or the tenant; whether the contractor has a proven track-record of successful data center projects; and how complex the journey is from power source to center.
Community Activism and Regulatory Delays Are Another Cog in the System
CNBC recently observed that 1GW of power is needed to support 750,000 homes for a full year; yet data centers are being built to support 1GW+ of power. It is not surprising that core data center markets such as Northern Virginia and Phoenix are out of power. Developers have been forced to tweak their location strategies, expanding beyond congested traditional hubs into emerging corridors with stable grids and greater room to capture new capacity.
Although power scarcity is one of the defining data center constraints, many players quickly learn that expanding capacity is not as simple as changing their location strategy. Land availability and regulatory friction in municipalities exacerbate the challenge, as concerns are raised over utilities powering up and leaving taxpayers on the hook for sharing the bill for data center usage, while the corporations behind the centers enjoy tax breaks and discounted rates.
In a positive step, in March of this year leading hyperscalers agreed to the Trump Administration’s Ratepayer Protection Pledge to protect consumers from rising energy prices as a result of digital infrastructure projects. This means hyperscalers and other AI companies will provide and pay for the power and delivery infrastructure needed to run their data centers, not taxpayers.
However, communities are beginning to mobilize around other grievances too, including the volume of water needed for cooling to ensure optimal computer equipment performance; minimal prospects for long term local job creation; and the risks of environmental degradation. From late March through June 2025 alone, approximately $100 billion in projects were blocked or delayed, forcing big tech names to retreat, creating a bottleneck threatening AI infrastructure deployment.
As power demand grows amid existing grid limitations and heightened environmental considerations, regulators and legislators face mounting pressure from constituents. Government policy is increasingly determining how fast projects move, not just whether they are approved as regulatory and environmental constraints turn an approval timeline that once took months into a multi-year process. With environmental regulation, zoning reform, grid prioritization frameworks and energy efficiency mandates all contributing to longer and less predictable delivery schedules this adds complexity and risk to project approvals, timelines and site selection.
The growing share of announced and committed projects that are stalling mid pipeline has left committed capacity more than double what is actually under construction in key hubs. This highlights the need for hyperscalers and developers to find tangible solutions that ensure environmental sustainability and community acceptance.
To Fund or Not to Fund
Borrowing for data center buildouts ballooned in 2025, after years of steady annual issuance, with Investment Grade (IG) loan volume spiking to $126 billion, a more than 650% year-over-year increase. For banks and project finance teams it is easier to get comfortable with IG counterparties, but not all IG is created equal.
While there are plenty of deals searching for capital and abundant capital for the right transactions, banks have become more selective. Given the elevated capex needs for AI infrastructure, robust alternative financing sources (e.g. PIMCO) will play a larger role. Some deals may come at a higher price or need to seek alternative financing sources.
When it comes to financing availability and terms, every project is unique, but the developer’s record is critical. And lenders have many questions. Have they built a data center before? Do they have deep pockets and enough skilled labor to successfully complete this unique type of project? Will they have access to incremental capital to get across the finish line? Is an anchor tenant lined up and how long is that lease? And importantly, what are the termination rights on that and other leases? Lenders want to know if a project could be a road to nowhere that results in no tenant and no operational data center if there is a power procurement delay.
Data center cash flows often rely on a small number of hyperscaler tenants, creating significant concentration risk and making asset performance highly dependent on successful long term pre leasing. Securing long term capacity with creditworthy hyperscalers before delivery converts development and capacity exposure into contracted revenues and reduces vacancy risk for lenders and investors.
In fact, 73% of the North American construction pipeline is pre leased, with vacancy falling to a record-low 2.3%, and strong pre leasing expected to keep vacancy suppressed through 2027. This leaves many companies with no choice but to pre lease space in developments under construction, often waiting a year or more to take occupancy.
To further reduce exposure to customer concentration and issuer risk, strong-credit backstop agreements are coming into play. A lease backstop from an IG counterparty guarantees rent (or termination payments) if the sole tenant defaults, turning single tenant exposure into contracted, bond like cash flows that protect debt service and valuations. In a recent example, TeraWulf and Cipher Mining received a backstop from Google providing $3.2 billion and $1.7 billion in backstops, respectively, in exchange for penny warrants. Although, certain “outs” for Google may create risk to the backstop structure, this type of creative structuring is being done to get deals completed.
Abundant Opportunities and Immense Complexity Abound
The opportunities are vast around data center development and the issuance needed to fund these unique projects. And the complexities around getting the centers built are immense. Partnering with the right advisors is as critical as financing, construction, power sourcing and community support become increasingly interwoven. Having an experienced team of advisors on your side every step of the way is a difference-maker.
