The multi-billion dollar challenge of AI data centers

The price of a data center is beginning to approach that of a giant tech company.

If a data center is valued at $100 billion, people easily think of the servers, GPUs, fiber optic cables, cooling towers, and substations inside. But what Wall Street is really pricing is a much longer set of relationships: whether the land can be used long-term, whether the power supply can be stable, whether the water and cooling systems can withstand high-density computing power, whether tenants can make long-term payments, whether the debt can be refinanced in the market, and whether the next generation of AI models will still need this capacity.

(Image caption) Exterior view of a large AI data center campus in the United States, showcasing the combination of ultra-large-scale server rooms and land resources, echoing the asset-heavy nature behind the hundreds of billions of dollars valuation of companies like Vantage.

The price of a single server room is beginning to approach that of a giant tech company.

Reuters reported on August 13 that Vantage Data Centers, backed by Silver Lake and DigitalBridge, is evaluating various options, including an IPO, a sale of a stake, and other strategic initiatives, with action potentially possible as early as 2027. If it chooses to go public, the company could raise approximately $10 billion, valuing it at around $100 billion. The report also emphasized that these discussions are still in the early stages and no formal procedures have yet been initiated.

If the proposed plan is ultimately implemented, it could become one of the most watched and largest public market transactions in the global data center industry to date.

The weight of this news doesn't just come from the valuation figures. In recent years, capital markets have priced chip companies first, then cloud service and model companies. Now, data center operators are also being thrust into the public market spotlight. Bloomberg previously reported that Switch had secretly filed for a US IPO; Reuters, in its follow-up report, cited sources saying the company's valuation could be close to $80 billion, including debt, while noting that it could not independently verify Bloomberg's original report.

The $80 billion for Switch and the $100 billion for Vantage, when viewed together, show that AI data centers are forming a distinct category of heavy assets. AI demand is no longer just about chip orders in Nvidia's financial reports, nor is it simply a model race between OpenAI, Google, Meta, and Anthropic. It's now tied to land, power grids, loans, leases, and local government approvals.

When data centers began to go public with valuations approaching $100 billion, investors were not buying a simple AI concept, but rather a set of long-term infrastructure cash flow assumptions.

Behind the valuation of hundreds of billions of dollars lies the scarcity of AI capacity.

Vantage is not a newly emerging AI concept company. It has long served hyperscalers, cloud service providers, and large enterprises, and its core capability is providing large-scale, scalable data center campuses. Its business is more akin to an infrastructure platform than a software company.

According to official company data, Vantage completed a $9.2 billion equity investment led by DigitalBridge and Silver Lake in 2024 to support the expansion of its global hyperscale data center platform; Reuters reported that Vantage has raised approximately $11 billion since the end of 2023. While this figure is high in the traditional data center market, it still appears insufficient in the context of AI.

A new AI data center is no longer just a few buildings and server racks. It may require hundreds of megawatts, or even close to one gigawatt, of power capacity, as well as supporting facilities such as power distribution, cooling, water treatment, network interconnection, GPU deployment, and long-term operation and maintenance capabilities.

Vantage recently partnered with Oracle and OpenAI at its Stargate data center campus in Wisconsin, positioning itself at the heart of the global AI infrastructure funding chain. The official announcement states that the campus will include four data centers, providing nearly one gigawatt of AI capacity.

This raises a very interesting phenomenon. AI model companies may not own all their data centers, cloud service providers may not handle all civil engineering and power connections, while infrastructure operators connect the park, power supply, and customer needs together. The high valuations given to companies like Vantage by the capital market are essentially based on the belief that the demand for computing power will persist in the long term and can be supported by sufficiently stable leases and service revenues, allowing them to build massive debt and equity capital.

This is also a sign that the AI industry is becoming more asset-heavy. Models can iterate rapidly on the software interface, but the physical world behind the models requires years of planning, approval, construction, and financing. When Wall Street begins to envision hundreds of billions of dollars for these physical foundations, the capital market story of AI enters another phase.

(Image caption) The interior of a high-density AI server room and its liquid cooling system demonstrate the physical core of computing power and the impact of technological updates on infrastructure depreciation.

AI data centers cannot be simply valued based on real estate valuations.

Data centers possess real estate attributes, but they cannot be valued simply using real estate logic.

The value of an office building depends on location, rent, vacancy rate, interest rates, and tenant credit. Data centers are similarly affected by these factors, but the AI era adds several more layers of uncertainty. Power capacity becomes a core asset, cooling capacity affects data center density, network interconnectivity determines computing efficiency, GPU update cycles influence customer demand, and chip supply and grid connection can alter project timelines.

The location value in ordinary commercial real estate is broken down into the combined value of energy, land, network, and licenses in AI data centers.

Calling companies like Vantage, Switch, or CyrusOne "AI landlords" isn't quite accurate. They're not renting out ordinary spaces, but rather a physical infrastructure capable of supporting high-density computing power. This infrastructure requires investment years in advance, with returns gradually achieved through long-term contracts.

One of the core issues in traditional real estate is vacancy rate, while the core issue for AI data centers is whether their capacity can be used long-term. Office space can be re-rented after tenants move out; however, if the power density, cooling methods, or hardware of an AI data center are outdated, the cost of readjusting it may be far higher than that of ordinary commercial space.

If Vantage proceeds with a market transaction at a $100 billion valuation, investors will likely compare it to Switch in the same group: which has stronger power access capabilities, which has higher tenant credit, which has lower project concentration, and which can maintain stable returns in a high-interest-rate and high-capital-expenditure environment.

These issues all point to the same thing: the valuation of AI data centers is not as simple as multiplying the valuation of traditional IDCs by an AI factor.

Electricity is becoming a valuation center

The valuation of AI data centers will eventually become increasingly similar to that of electricity.

Without a stable, scalable, and cost-effective power supply, even the best land and buildings cannot support large-scale AI training and inference. In 2025, Vantage announced plans to build a more than $25 billion AI campus in Shackelford County, Texas, covering approximately 1,200 acres with a planned capacity of 1.4 gigawatts, making it the largest campus in the company's global portfolio.

1.4 gigawatts is not a figure typical for a typical industrial park. It approaches the load of a large power plant. As data centers grow from tens of megawatts to hundreds of megawatts, or even gigawatts, local power grids, transmission lines, backup power supplies, energy prices, and public policies all come into play in corporate valuations.

This is why financial institutions such as Bank of America, Morgan Stanley, and JPMorgan have recently begun to incorporate AI data centers into their critical infrastructure frameworks. These institutions are seeing not only the need for data center construction but also the associated investment needs in power grids, energy storage, natural gas, transportation, and other infrastructure. AI data centers are no longer isolated projects; they are reshaping the energy and land use of entire regions.

For investors, evaluating data center companies in the future should not focus solely on megawatt count. It's also crucial to consider whether those megawatts are licensed, whether there are long-term power contracts, whether transmission infrastructure is in place, whether electricity prices are predictable, whether local residents are willing to accept the technology, and whether power fluctuations will increase equipment depreciation and downtime risks.

The AI industry was initially often described as a race to create models and chips. With electricity now at the center of valuations, it's beginning to resemble a crossover of energy, real estate, credit, and local governance.

(Image caption) The large substation and high-voltage transmission facilities next to the data center highlight that power capacity has become a core element in the valuation of AI data centers.

Wall Street buys contracts, but is worried about concentration.

The valuation of AI data centers is largely based on long-term contracts.

When a large cloud customer signs a capacity agreement for ten, fifteen, or even more years, operators can incorporate future rental or service revenue into their financing models, supporting loans, bonds, and equity investments. These types of contracts make asset-heavy projects appear more like infrastructure than typical technology investments.

However, the contract itself also brings concentration risk.

If revenue is heavily reliant on a few large cloud service providers or AI model companies, investors need to assess not only the lease term but also the sustainability of customers' capital capabilities and computing power needs. AI demand is growing rapidly, but model roadmaps, chip architectures, and computing efficiency are also changing quickly. Once a generation of infrastructure becomes unsuitable for new hardware, the security of long-term contracts will be re-examined.

Oracle's Stargate project, linked to OpenAI, brought this financing pressure to the fore earlier. Reports indicate that large banks need to absorb billions of dollars in loans while controlling exposure to single clients and single AI infrastructure chains.

For operators like Vantage, the larger and creditworthy their customers, the easier it is to secure financing; however, the more concentrated their customer base, the more the market will question whether the risks are excessively concentrated.

This is similar to the old problem of "single large tenant" in commercial real estate, except that in the AI era, tenants are not renting desks, but rather the capacity to support electricity, server racks, networks, and future computing power.

(Image caption) The Wall Street financial scene and market pricing atmosphere symbolize how the public market prices long-cycle infrastructure assets such as AI data centers.

Community opposition is entering the financing model

The capital story of AI infrastructure cannot be written only on Wall Street.

Reuters reported on August 10 that major U.S. banks and asset management firms have begun incorporating local resident attitudes, permitting progress, and political resistance into credit risk assessments when reviewing AI data center financing. In the first quarter of 2026, at least 75 data center projects in the U.S., worth approximately $130 billion, faced local opposition. Banks are placing greater emphasis on whether projects have obtained the necessary permits and whether the surrounding communities support them.

This has direct implications for Vantage, Switch, and the entire AI data center market. The foundation of a valuation in the hundreds of billions of dollars is not just a lease agreement, but also a table of local governance.

Projects may encounter obstacles due to noise, water resources, electricity prices, land use, and environmental impacts. Even if customer demand is genuine, construction delays can alter cash flow; and even if financing is secured, local political obstacles can increase the cost of capital.

The community debate in Port Washington, Wisconsin, surrounding the Vantage, OpenAI, and Oracle data center projects demonstrates how AI infrastructure is increasingly becoming a part of local politics. Public subsidies, transparency, environmental impact, and community involvement have all become focal points of contention, prompting residents to demand greater public scrutiny of future large-scale tax subsidies.

The public market will pay attention to these details. Because valuing a company at $100 billion means investors believe it can replicate projects across multiple regions. If local communities begin demanding more compensation, greater transparency, and stricter scrutiny of water and electricity, the replication speed will slow down, and the return on capital will be recalculated.

There's also the human side to this. Data centers can generate tax revenue and create jobs, but they can also bring noise, water and electricity pressures, electricity price concerns, and land-use conflicts. The more AI moves into infrastructure, the more it needs to answer how it integrates into a local society, rather than just how it serves a cloud customer.

A clarification from the SEC could change the toolbox of financing tools.

For AI data centers to go public, they need not only equity, but also debt and securitization instruments.

Reuters reported on August 10 that the U.S. Securities and Exchange Commission (SEC) clarified that certain data center bonds are not considered asset-backed securities and are therefore exempt from several key securitization rules. This clarification, stemming from a request by Latham & Watkins, could make it easier for data center owners to issue infrastructure-related fixed-income securities to support rapid expansion driven by AI computing demand.

If data center bonds are not considered traditional asset-backed securities, issuers may have more flexibility in their structuring. Investors therefore need to carefully examine the underlying cash flows and project risks themselves.

Relaxing regulations won't eliminate risk; it merely broadens the financing toolbox. Investors will still need to assess tenant creditworthiness, contract terms, project construction risks, asset residual value, and technology upgrade cycles. If the data center bond market expands rapidly in the future, rating agencies, insurance companies, pension funds, and asset managers will all be involved in pricing the risks of AI infrastructure.

The volatility of the AI boom will also spread from tech stocks to the credit market and long-term investment portfolios. This represents a new transmission channel for the entire financial system.

Once the valuation reaches $100 billion, depreciation will become a core issue.

AI data centers may look like infrastructure, but they are more susceptible to technological updates than many traditional infrastructures.

A highway can last for decades, and airport runways and power lines also have long lifespans. While the buildings and power facilities of AI data centers can be used for a long time, the equipment, cooling density, and hardware configuration inside can change rapidly. The speed of GPU updates, changes in model architecture, rising inference demands, and the widespread adoption of liquid cooling solutions may all mean that today's designs will need to be completely redesigned in a few years.

Therefore, behind the $100 billion valuation, one must look at the depreciation policy and reinvestment needs.

If operators must continuously invest in new cooling systems, higher power densities, new network equipment, and new security standards, nominal lease revenue may not equate to free cash flow. How much maintenance capital expenditure is required behind every dollar of revenue? Who bears the upgrade costs for each campus? Are tenants willing to pay a premium for higher-density, next-generation data centers?

Commercial real estate can achieve relatively stable valuations through rent and asset appreciation. However, the value of AI data centers is more deeply embedded in the technology cycle. Misinterpreting "long-term leases" as "long-term low-risk" is one of the most common misjudgments in this industry.

The public market will ultimately factor these issues into depreciation, free cash flow, refinancing costs, and valuation multiples. When data center companies go public, investors need to look not only at revenue growth curves but also at the rate at which assets age.

(Image caption) Local community public hearings discuss data center projects, reflecting that residents' attitudes and political resistance have been incorporated into the credit risk assessments of banks and investors.

Several things the open market will ask

When analyzing AI data center IPOs, one cannot look at only the amount of funds raised and the valuation.

When pricing AI data centers in the public market, the first thing to look at is the assets, but the contracts will be quickly scrutinized. Land, server room, power access, cooling system, network connectivity, and security level determine whether a data center can handle high-density AI workloads; lease term, tenant credit, capacity bookings, and power contracts determine the stability of future cash flow.

Further down the chain are equity, bank loans, project bonds, and refinancing arrangements. At this level, AI data centers are no longer just technology assets, but part of the credit market. Financing terms, interest rates, collateral arrangements, and refinancing windows directly impact shareholder returns.

There are also public conditions. Local permits, community acceptance, water and electricity resources, tax subsidies, and environmental requirements determine whether a project can start construction and operate on time. If a data center is stuck in hearings, power transmission permits, or water resource reviews, the cash flow in the model will be delayed.

Finally, there's the technology cycle. GPU upgrades, liquid cooling, optical interconnects, model efficiency, and the structure of computing power requirements determine whether today's assets are still suitable for tomorrow's AI workloads. The faster the technology updates, the more important sustaining capital expenditures become.

These factors, taken together, bring us closer to the true valuation basis of an AI data center. Ignoring any one of them can easily lead to an oversimplified view of an AI data center.

AI's Wall Street Moment: It Starts in the Data Center

AI was initially understood by many as models, software, and algorithms. Now, it increasingly resembles an infrastructure revolution.

This revolution requires chips and electricity, models and land, engineers and bankers, and acceptance from local governments and residents. If Vantage does indeed go public with a valuation close to $100 billion, it will not only be a major deal in the data center industry, but also a public test for the AI capital market.

The public market is not testing what AI demand is.

It needs to examine whether cash flow can support the valuation, whether electricity and cooling can withstand the load in the long term, whether tenants are stable enough, whether the local community is willing to accept it, and whether the technology cycle will cause today's assets to age prematurely.

The answers to these questions will not only be written in the prospectus, but also in the grid connection agreements, local hearings, lease terms, and the operational data of each park.

When the future of a model is crammed into land, electricity, and debt, what investors really need to measure is whether these assets can hold up over a long period of time.

Disclaimer

This article is for news research and public discussion purposes only and does not constitute investment, legal, tax, or transaction advice. Any information regarding IPO likelihood, valuation, cooperation arrangements, and financing conditions mentioned herein is subject to company announcements, regulatory documents, and official disclosures.