It would be difficult to construct a more revealing snapshot of the AI infrastructure market than the one delivered in mid-July.
In the same news cycle, Csquare completed a billion-dollar initial public offering, Switch was linked to a potential $10 billion IPO, and Databricks reached a reported valuation of $188 billion. At the project level, developers advanced or disclosed campuses measured not in tens or hundreds of megawatts, but in gigawatts—from Meta’s expanding Louisiana complex and Google’s reported Wyoming plans to new Crusoe, QTS, MARA and Tract developments.
Yet the same week brought a state-level permitting pause in New York, a decisive project rejection in Palm Beach County, planned protests across more than 20 states, and fresh disputes over parkland, water availability and local control.
This is the data center and AI landscape in 2026: capital is abundant but increasingly discriminating; power is more valuable than the underlying real estate; and community consent has become nearly as important as interconnection capacity.
Public Markets Put Different Prices on the AI Stack
The capital-market headlines illustrated how differently investors are valuing the various layers of AI infrastructure.
Csquare priced 50 million shares at $21, raising approximately $1.05 billion and establishing an equity valuation of roughly $3.2 billion. The offering was substantial, but it priced below the proposed $23-to-$27 range, and the shares finished their first trading day slightly below the offer price. Brookfield retained approximately 67% of the company’s voting power following the transaction.
That reception contrasts sharply with the valuation being discussed for Switch. The DigitalBridge-backed operator has reportedly engaged Goldman Sachs and JPMorgan for a potential IPO that could raise as much as $10 billion and value Switch near $80 billion, including debt. The transaction remains prospective, but the figure is striking when compared with the $11 billion take-private agreement announced in 2022.
Higher still in the infrastructure stack, Databricks signed a term sheet for a strategic financing at a $188 billion valuation. The Wall Street Journal reported the investment at approximately $3 billion—an important correction to headlines suggesting the company had raised $188 billion. Databricks said the financing would support products including its Unity AI Gateway, Genie and Lakebase, as well as research and potential acquisitions.
The comparisons are imperfect. Csquare owns and operates physical data center assets; Switch is a much larger private infrastructure platform; Databricks provides the software and data foundation on which enterprise AI is built. But that is precisely the point. At its new valuation, Databricks is worth almost 60 times Csquare’s IPO equity value.
The market continues to place its greatest multiples on the intelligence and software layers. Physical infrastructure, despite unprecedented demand, must still contend with leverage, construction risk, power availability and multiyear deployment schedules.
The Campus Pipeline Moves Into the Gigawatt Era
The project announcements reinforce how thoroughly the gigawatt has replaced the megawatt as the industry’s unit of ambition.
Meta says the full expansion of its Hyperion campus in Richland Parish, Louisiana, will reach 5 GW and represent more than $50 billion in investment. Google has been identified through local and industry reporting as the customer behind the proposed 2.7 GW Project Tembo near Cheyenne. MARA’s planned acquisition in Matagorda County could support as much as 2 GW, while Crusoe and Lancium announced a 1 GW Childress campus and QTS outlined another 1 GW development on a Lancium site in Hall County.
Add Tract’s proposed 900 MW park outside Richmond, CleanSpark’s 175 MW lease, Plug Power’s 164 MW interconnection position and a preliminary 100 MW Crusoe-linked project in Franklin County, and the developments in this single group approach 13 GW of headline capacity.
That DCF calculation is not an apples-to-apples comparison. The figures include IT load, grid-interconnection capacity, campus potential and multiyear full-build estimates at dramatically different stages of development. Nonetheless, the aggregate conveys the scale of the pipeline—and the distance between announcing a gigawatt and successfully energizing one.
Crusoe and Lancium’s Childress model is particularly instructive. The companies describe a 270-acre campus capable of supporting hundreds of thousands of AI accelerators, with energy storage, closed-loop non-evaporative liquid cooling and construction beginning in the third quarter. Lancium is contributing land, interconnection and energy-orchestration capabilities; Crusoe will design, construct and operate the infrastructure.
QTS is applying a related division of responsibilities at Lancium’s Clean Campus near Turkey, Texas. QTS would build and operate as many as 11 data centers, while Lancium supplies a power-ready campus with a 1 GW grid connection. Reported investment exceeds $10 billion.
The model increasingly resembles industrial-scale platform assembly: one participant secures land and power, another supplies development capital, an operator delivers the facility, and a hyperscale or AI customer provides the contracted demand.
The Contract Is Becoming as Important as the Campus
CleanSpark’s Sandersville agreement demonstrates how developers are attempting to reduce the risks embedded in that model.
The company signed a 20-year triple-net lease with an unnamed investment-grade global technology company for 175 MW of critical IT capacity beginning in the fourth quarter of 2027. CleanSpark values the base contract at approximately $6.6 billion, potentially rising to $11.6 billion if two five-year extensions are exercised. It also granted the customer exclusivity over an additional 885 MW in its Texas portfolio.
The base contract equates to approximately $330 million in average annual revenue. CleanSpark estimates its landlord investment at $10 million to $12 million per megawatt, implying a project cost of roughly $1.75 billion to $2.1 billion for the initial deployment.
That per-megawatt range loosely aligns with the headline ratios surrounding several other AI campuses. Meta’s greater-than-$50 billion investment against 5 GW equates to more than $10 million per megawatt. QTS and Lancium’s reported $10 billion against a 1 GW grid position produces a similar result.
The scope behind each number differs, and none should be treated as a standardized construction benchmark. But collectively they illustrate the capital intensity now associated with AI-ready power, cooling, networking and resiliency.
For developers, a long-term investment-grade lease can transform that capital requirement from a speculative bet into financeable infrastructure.
Power Rights Are Becoming Tradable Infrastructure
Several announcements also showed that the scarce asset is often not the building. It is the right to deliver power to the building.
Plug Power agreed to sell Stream Data Centers a 66-acre site in Graham, Texas, carrying a 164 MW interconnection position. Potential proceeds total as much as $76.5 million. Stream is not merely purchasing land; it is acquiring a place in the power-development queue.
MARA’s agreement to acquire more than 1,200 acres from HIF USA follows the same logic at a much larger scale. The Matagorda County property could support 1 GW by October 2027 and 2 GW by April 2028. MARA plans to work with Starwood Digital Ventures and retain the flexibility to support both high-performance computing and Bitcoin mining.
In Nebraska, Monolith is exploring whether its Olive Creek carbon-black facility can host a data center, with Crusoe identified as a potential partner. The attraction is obvious: large industrial sites may already possess land, transmission access, operating infrastructure and experience managing intensive energy loads.
The $1.7 billion financing commitment involving Industrial Development Funding, Oaktree, Bloom Energy and Nebius extends the power race beyond the traditional grid. Bloom’s solid-oxide fuel cells are intended to provide onsite electricity for Nebius facilities, with an initial 328 MW previously targeted for deployment during 2026.
This is more than a backup-power story. It reflects a growing willingness to finance dedicated generation as part of the primary data center architecture when utility timelines cannot match AI deployment schedules.
The Physical AI Stack Extends to the Optical Connector
The 3M-Microsoft partnership provides a useful counterpoint to the gigawatt announcements.
Microsoft will become the first announced hyperscale cloud provider to deploy 3M’s Expanded Beam Optical technology. The connector system is designed to improve the reliability, density and serviceability of optical links inside data centers. At the same time, 3M will expand its own use of Microsoft cloud and AI tools.
As AI campuses move toward hundreds of thousands of accelerators, infrastructure constraints migrate deeper into the system. The challenge is no longer confined to securing generation, transformers and cooling plants. It also includes the fiber connections, optical interfaces and physical networking required to keep accelerators operating as one coordinated computing environment.
That makes the AI factory a genuinely integrated design problem. A campus can possess adequate utility capacity and still fail to deliver expected performance if its cooling distribution, network fabric or physical connectivity becomes the limiting component.
Developers Are Testing New Markets—and Local Patience
The pipeline is also spreading into markets that have not historically ranked among the largest data center hubs.
Tract’s proposed 900 MW Tuckahoe Technology Park would bring as many as 12 buildings to an 872-acre property in Goochland County outside Richmond. A Crusoe-linked project could bring approximately 100 MW to Franklin County, although it remains preliminary. In Mifflin County, Pennsylvania, a developer filed plans only two days before officials were due to vote on a local moratorium.
In Salem, Oregon, Verrus is attempting to position its proposed three-building development as a more responsive model. The company emphasizes closed-loop cooling, battery storage and the ability to adjust power consumption as grid conditions change. Its promise of “a better data center,” however, remains a developer proposition that will ultimately be judged through permitting, operation and community experience.
These projects demonstrate the geographic expansion of AI infrastructure. They also show developers encountering local governments and populations that increasingly understand the scale of what is being proposed.
The Social License Becomes a Development Gate
No mid-July development expressed the change in political climate more clearly than New York’s action.
Gov. Kathy Hochul ordered an up-to-one-year pause on discretionary environmental permitting for certain new hyperscale data centers while the state develops a broader environmental review and policy framework. The order is not a blanket ban: projects with completed permits can proceed, and the action is focused on large new facilities requiring state environmental approvals.
President Trump responded by portraying data centers as economic assets that New York was surrendering to competing states. His intervention placed data center development squarely within the national competition for AI investment, tax revenue and technological leadership.
In Palm Beach County, however, the opposition was local and decisive. Commissioners rejected Project Tango by a 5–1 vote following a hearing that stretched for roughly 12 hours. The proposed one-million-square-foot development could have required as much as 600 MW—approximately 12 times Florida’s 50 MW threshold for enhanced regulatory scrutiny. Even with closed-loop cooling, the developer projected potable-water demand of approximately 100,000 gallons per day.
A national protest campaign planned for July 18 added another dimension. Organizers said actions would take place in more than 50 communities across over 20 states, focusing on electricity costs, water consumption, noise, land use and the influence of developers over local zoning decisions.
Other controversies show how quickly a land-use decision can acquire symbolic significance. In Taylor, Texas, residents are fighting development on 87 acres originally transferred to the city for nominal consideration and intended for park use. The city subsequently sold the land to a data center developer for $10 million. In Trinidad and Tobago, proposed 150 MW and 300 MW developments have prompted questions about whether a country already confronting water and energy constraints should assume the additional burden of AI infrastructure.
Meta Offers the Counter-Narrative
Meta’s Louisiana expansion represents the industry’s clearest attempt to demonstrate that large-scale development can produce visible community gains.
The company says Louisiana businesses have received more than $1.6 billion in contracts associated with the Hyperion campus and that Meta is supporting more than $1 billion in local infrastructure improvements. The expanded project is expected to support over 1,000 permanent jobs. Local reporting indicates that teacher bonuses in Richland Parish have risen from approximately $10,000 to more than $50,000 as new tax revenues arrive.
These are powerful figures, but they also establish a higher standard for the wider industry. Communities will increasingly compare every new campus with projects able to demonstrate direct school funding, local procurement, infrastructure upgrades and durable employment.
A promise of regional economic development will no longer be sufficient. Developers will be expected to quantify who receives the contracts, how utility costs are allocated, what resources the facility consumes and what benefits remain if construction employment falls away.
Announced Is Not the Same as Delivered
Taken together, the mid-July headlines do not tell a simple story of limitless AI infrastructure growth. They describe a narrowing execution path.
Capital must be secured at an acceptable price. A creditworthy customer must sign a contract. Land must carry credible transmission access. Generation, cooling and optical systems must operate as an integrated platform. Local officials must believe that the benefits justify the resource demands. Residents must trust both the developer and the process through which the project was approved.
The industry’s center of gravity is consequently shifting from announcement to execution.
The winning AI campus will not necessarily be the project with the largest gigawatt number in its press release. It will be the project whose megawatts survive underwriting, interconnection studies, equipment lead times, environmental review, rate proceedings—and the next county commission meeting.
