Taken individually, last week’s major data center announcements looked familiar enough for the AI era: another billion-dollar investment, another multibillion-dollar compute agreement, another massive Texas campus, another nuclear proposal measured in gigawatts. Taken together, they told a more consequential story.

The AI infrastructure buildout is moving into a phase defined less by simple expansion than by control — control of capital, customer demand, real estate, power generation, supply chains and the increasingly complicated political permission required to turn all of those ingredients into operating data centers.

Samsung committed $1 billion to a new AI infrastructure platform. Lambda tapped investment-grade debt markets to fund contracted GPU deployments while advancing a new Oklahoma data center. Akamai signed an $11.6 billion compute agreement with Anthropic. Data center platforms 5C and Edged surfaced at opposite ends of potential multibillion-dollar capital-market transactions.

Meanwhile, developers in Texas are increasingly building around energy rather than waiting for it. Behind-the-meter generation is becoming an infrastructure category of its own. Amazon is reworking its nuclear strategy in Maryland. And in Virginia, the largest data center market in the world is moving toward significantly tougher requirements governing how projects are disclosed, approved and ultimately accepted by their communities.

The common denominator is becoming increasingly difficult to miss. To wit: At AI scale, the successful data center developer can no longer simply secure land, obtain utility service and construct a building. The emerging model requires assembling an entire infrastructure ecosystem around the workload — and demonstrating that the resulting project can survive financial, electrical and political scrutiny.

Capital Is Becoming Infrastructure

Samsung’s $1 billion commitment to Helix Digital Infrastructure offered one of the clearest examples yet of how the capital structure surrounding AI data centers is changing.

Helix was formed by KKR as an AI infrastructure platform with more than $10 billion already committed by founding investors including KKR, the Kuwait Investment Authority, NVIDIA and Vistra. Samsung’s new commitment pushes that capital base still higher.

But the composition of the partnership may be more significant than another billion dollars being added to the AI infrastructure ledger. Helix is intended to invest across hyperscale data centers, power generation and transmission, fiber and other connectivity infrastructure. Samsung, meanwhile, brings capabilities extending across advanced technology, construction, energy storage and cooling.

This is not simply capital chasing data center returns. It increasingly resembles an attempt to assemble the data center, energy and technology supply chain inside a single investment ecosystem.

That distinction is important, because one of the defining problems of the current buildout is that capital by itself does not produce capacity. Billions of dollars can be committed long before transformers arrive, transmission is constructed, generation is secured or a campus is commissioned. The increasingly valuable infrastructure platform is therefore the one capable of controlling more of those dependencies.

Lambda demonstrated another side of that evolution last week with the closing of a $1.008 billion delayed-draw term loan supporting three committed customer deployments across multiple data centers. The financing received investment-grade ratings from Morningstar DBRS and Moody’s and carries a 6.78% fixed interest rate.

More importantly, it is secured by both the GPU infrastructure being financed and contracted cash flows from two investment-grade customers. Capital is drawn as infrastructure reaches commissioning milestones rather than simply being handed to Lambda upfront. That begins to make AI compute look less like speculative technology spending and more like an underwritable infrastructure asset.

Elsewhere, 5C Group is considering entering the public markets after securing more than $1.4 billion since 2025. The company is first looking to complete another financing round reportedly worth between $5 billion and $6 billion.

At Edged US, meanwhile, Bain Capital was reported among potential bidders in a sale process that could value the data center company at around $15 billion. Edged has facilities operating or under development across a growing roster of U.S. markets, while Bain already owns data center assets including DC BLOX in the United States.

Even a comparatively modest transaction announced by DataBank last week fits the same theme. DataBank acquired the building housing its MSP2 facility in Minneapolis after years of leasing the property. Outgoing CEO Raul Martynek described ownership as a means of controlling the company’s own destiny — giving DataBank greater authority over long-term investment in a facility it considers a critical regional infrastructure node.

From billion-dollar funds to the dirt under the building itself, ownership and control are becoming strategic assets.

Contracts Are Becoming the Foundation Beneath the Buildout

Of course, increasingly sophisticated capital structures only work if someone is willing to buy the compute. That is what made Akamai’s newly announced agreement with Anthropic one of the more revealing numbers of the week.

Anthropic has entered into a seven-year, $11.6 billion arrangement to use Akamai Cloud infrastructure for CPU workloads, with provisions that could expand the commitment by another $9 billion.

Akamai estimates approximately $5.5 billion in associated capital expenditures, including increased spending this year to secure critical supply-chain components such as memory.

The numbers are enormous, but the mechanism matters more. Long-duration AI workload commitments are increasingly becoming the financial foundation underneath physical infrastructure investment.

The Lambda financing offers an almost unusually clean illustration of the sequence. A customer signs a long-term commitment. That contracted cash flow supports investment-grade debt. The proceeds finance GPUs and associated infrastructure. The physical capacity comes online according to commissioning milestones.

Contract becomes credit. Credit becomes compute. Compute becomes infrastructure. Then comes the dirt-under-the-fingernails part.

Lambda also announced plans for a new data center at MidAmerica Industrial Park in Mayes County, Oklahoma. The company estimates that the project could generate $500 million in taxes over a decade while creating as many as 1,000 construction jobs and 100 permanent positions.

The announcement also illustrates how much more is now attached to a major data center commitment than servers and square footage.

Lambda says it will pay the full cost of the energy resources and infrastructure required to serve the facility under Oklahoma’s Data Center Customer Ratepayer Protection Act. It plans closed-loop cooling to reduce water consumption and has emphasized siting, community engagement and the project's impact on schools and public services.

In other words, the financial promise supporting an AI campus is increasingly accompanied by another kind of contract — an implicit agreement with the community hosting it. That will become important again farther down the page.

The Buildout Is Following the Energy

If the capital stories showed who can finance the AI buildout, the development stories last week offered an equally clear picture of where that money increasingly wants to go. It wants to go where the energy is.

Crusoe’s new Armstrong County, Texas, campus may be the purest expression yet of what the company calls an energy-first strategy. The Google AI data center campus is directly connected to Serena’s adjacent 265.5 MW Goodnight 1 wind farm and a second 265.5 MW wind farm under construction. When generation exceeds campus demand, surplus electricity can flow back to the grid. The project will also use closed-loop, non-evaporative liquid cooling designed to sharply reduce water consumption.

Crusoe’s formulation is straightforward: build AI factories at the source of abundant energy rather than treating electricity as something to be procured after a site has been selected. That same logic can be seen in Zone Frontier’s emerging West Texas campus in Potter County.

The company has secured a long-term lease covering 4,077 acres, including land and water rights. An initial 200 MW of data center capacity could eventually expand beyond 500 MW. The property surrounds a 345-kV substation, sits near two natural gas pipelines and includes at least 1,800 acres that can support solar generation, battery storage and associated infrastructure.

Strip away the real-estate terminology and that parcel effectively represents the new site-selection stack in physical form: Land. Substation. Gas. Water. Generation. Storage. Expansion room. The data center is almost the final component rather than the first.

AB Energy supplied another striking data point. The company says it has secured approximately 2 GW of generation capacity for five onsite data center power projects across Texas and Oklahoma.

The projects will use modular gas-fired generation, with some incorporating battery energy storage. Absorption chillers can recover heat from the engines to produce chilled water, while the engineering effort includes addressing the step loads, ramp rates and voltage transients characteristic of AI infrastructure.

Two gigawatts of onsite generation spread across five projects is difficult to dismiss as an edge case. Behind-the-meter power has become part of the mainstream conversation because the grid connection itself has become one of the least controllable variables in data center development.

For years, site selection could be summarized around familiar requirements: land, fiber, taxes and utility power. At AI scale, developers are increasingly asking a different question: What can we control ourselves?

Nuclear Moves From Concept to Contrasting Models

Nowhere is the gap between AI infrastructure ambition and actual deployability more visible than nuclear power. Two developments last week illustrated almost opposite ends of that spectrum.

Amazon has abandoned plans for a data center campus beside Constellation Energy’s Calvert Cliffs nuclear plant in Maryland after the proposal encountered community opposition. Its replacement strategy is notable.

Amazon and Constellation have instead reached an agreement under which upgrades at Calvert Cliffs are expected to add 190 MW of generation that will feed into the regional grid. Constellation, in turn, will provide Amazon with as much as 690 MW of long-term electricity from across its generation fleet. The companies will also explore the possibility of additional reactors at the Maryland site.

That represents an important variation on the nuclear-data center model. Rather than physically pairing a dedicated data center with an existing reactor and effectively carving generation out of the surrounding system, the arrangement expands the generation base while Amazon contracts for long-duration supply.

The Utah proposal announced last week sits at the opposite end of the development curve. Valar Atomics has proposed Project Beehive on more than 9,000 acres near Price, Utah. Plans reportedly envision approximately 456 small modular reactors eventually producing as much as 9.6 GW of electrical capacity alongside data centers, nuclear-fuel production and waste-storage facilities.

The scale is extraordinary. So is the distance between proposal and operation. No company has yet placed an SMR into commercial operation. Valar is developing a high-temperature gas reactor and has conducted an early criticality demonstration, but Project Beehive would require an enormous progression through licensing, permitting, manufacturing, construction and deployment before anything approaching its proposed scale materializes.

That doesn't make the project irrelevant. Quite the opposite. The fact that serious infrastructure discussions now contemplate hundreds of reactors and nearly 10 GW of generation around data center demand illustrates just how radically AI has changed assumptions about future electricity requirements.

But announced gigawatts and deployable megawatts remain very different commodities. The nuclear story of the AI era may therefore develop along two tracks: upgrades and extensions to existing fleets that can provide near- and medium-term capacity, alongside more ambitious advanced-reactor architectures seeking to solve the next decade’s problem.

Power Is Not the Only Permission Required

The other constraint surfaced last week in Virginia. Governor Abigail Spanberger has unveiled a sweeping Data Center Accountability Framework aimed directly at how data center development is approved and governed in the world's most important data center market.

Among its provisions are plans to ban nondisclosure agreements surrounding commercial data center projects, eliminate by-right approval for facilities using more than 25 MW, give communities additional tools to negotiate agreements, remove large projects from certain fast-track permitting processes and impose stronger environmental and ratepayer protections.

For an industry accustomed to thinking primarily in terms of power availability, interconnection queues and construction schedules, the shift is significant. But it is hardly surprising. A 30 MW data center can largely be understood as a commercial development. A 300 MW campus begins to look like regional infrastructure.

A multi-gigawatt AI development with dedicated generation, transmission requirements, major water systems and billion-dollar public consequences can begin to resemble something closer to an industrial policy decision. At that scale, communities will expect a seat at the table.

The clues were present throughout last week's announcements. Lambda did not simply promote its Oklahoma project's compute capacity. It emphasized ratepayer protection, closed-loop cooling, tax revenue, schools and community engagement.

Zone Frontier discussed local benefits alongside power and permitting. Crusoe highlighted not only the energy supplying its Texas campus but surplus electricity returning to the grid and investments in efficiency for local homes and public buildings.

And Amazon's Maryland strategy changed after community resistance helped derail its original concept. The industry is learning that securing a megawatt is not the same thing as securing permission to use it.

This may ultimately be the most important shift revealed by the week's developments. The companies attempting to win the AI infrastructure race are assembling ever larger pools of capital, signing ever larger compute contracts and finding increasingly creative ways to secure electricity.

They are bringing generation behind the meter, moving campuses toward energy sources, considering public markets and reorganizing the relationship between data centers and the power grid itself. The technical and financial machinery is becoming extraordinarily sophisticated. But the buildout is also becoming more visible.

And that means the next generation of successful AI infrastructure platforms will have to control almost everything they can — while recognizing that one critical variable cannot simply be bought, engineered or brought behind the meter. The right to build will increasingly have to be earned.

At Data Center Frontier, we talk the industry talk and walk the industry walk. In that spirit, DCF Staff members may occasionally use AI tools to assist with content. 

Keep pace with the fast-moving world of data centers and cloud computing by connecting with Data Center Frontier on LinkedIn, following us on X/Twitter and Facebook, as well as on BlueSky, and signing up for our weekly newsletters using the form below.