Power Is Not Yet Compute

In April 2026, Core Scientific disclosed approximately 590 megawatts of data-center capacity contracted to CoreWeave. More than 185 megawatts were active and being billed as of March 13. The figures described a phased buildout, but they also made an important distinction visible: contracted capacity and revenue-producing infrastructure are different things. Core Scientific, Inc.

Between them sit engineering, equipment, construction, cooling, connectivity, commissioning and customer acceptance.

That gap is where an important part of the AI infrastructure opportunity is taking shape. Securing electricity matters. Turning it into capacity a customer can actually use is a different capability. Doing so repeatedly, without rebuilding the company’s balance sheet for every project, is another.

For venture investors, these distinctions matter more than the size of the development pipeline.

Scarcity creates the opening. It does not, by itself, create the enduring company.

A capacity gap / 01

Contracted capacity is not yet billable capacity

~590MW

net critical IT capacity contracted to CoreWeave

>185MW

active capacity being billed

Between the two

Construction · equipment · cooling · connectivity · commissioning · customer acceptance

The scarce input is usable capacity

The demand backdrop is substantial. The IEA projects global data-center electricity consumption to rise from 485 terawatt-hours in 2025 to approximately 950 in 2030. But its analysis also identifies constraints across electricity, grid connections, advanced chips and capital. Power has not simply replaced semiconductors as the bottleneck. Several constraints are tightening together. IEA

Electricity is particularly consequential because its delivery timeline can exceed that of the building it serves. JLL’s 2026 outlook puts the average wait for a grid connection in primary data-center markets above four years. JLL

Yet “available power” remains an incomplete description of what customers buy.

A connection expected in five years is not equivalent to one operating today. Sufficient annual renewable generation does not establish continuous supply. A facility with enough electricity may still lack the cooling, networking or physical configuration a particular deployment requires.

The product is capacity delivered in the right place, on the required date, at the necessary performance and reliability, for an acceptable total cost.

That creates an opening for companies that can make previously difficult capacity usable. But it also sets a demanding test: solving one constraint is valuable only if the solution does not introduce a more expensive constraint elsewhere.

New architectures compete with an improving incumbent

There are several ways to close that gap.

A company might bring generation closer to the customer. It might combine storage with controlled flexibility to accelerate a grid connection. It might make an existing facility support more compute through better cooling and power management. Or it might develop a different physical architecture for locating and operating data centers.

These approaches compete with one another—not just with the existing construction pipeline.

A published NVIDIA and Nscale evaluation illustrates the point. Dynamic power allocation allowed more GPUs to operate within the same provisioned power budget, increasing aggregate throughput. Actual power consumption still rose, and the slowest responses became slower. The result demonstrated both the opportunity to use existing capacity more effectively and the trade-offs that must be measured. NVIDIA Developer

The implication is important for any new infrastructure company. Its benchmark is not today’s data center held constant. It is the best available alternative when its own product enters service.

Unconventional locations and energy systems belong in that comparison. An offshore or off-grid design could avoid a constraint that makes a conventional site uneconomic or too slow. But its advantage must survive the full cost of maintenance, backup supply, connectivity and financing.

Novelty is not the investment case. Better delivered economics are.

Flexibility is a product feature, not permission to fail

Workload flexibility creates genuine room for new infrastructure.

In March 2026, Google reported integrating one gigawatt of demand-response capacity into utility agreements. By limiting or shifting a portion of its machine-learning workloads, it can reduce demand during selected periods and help new facilities connect more rapidly. Google also makes clear that flexibility has limits and is available only at certain locations. blog.google

That is evidence for deliberately designed flexibility. It is not evidence that customers will accept unpredictable outages.

The distinction is commercial as much as technical. A customer may agree to defer background work during a defined window. That is different from discovering that a facility cannot operate for an uncertain period. Advance notice, duration, recovery time and the consequences for the workload determine whether flexibility is useful or costly.

Geographic redundancy does not remove those costs. AWS recommends provisioning enough capacity to withstand the loss of an availability zone rather than depending on launching replacement capacity after a failure. Resilience comes from resources and architecture arranged in advance. AWS Documentation

For an emerging provider, the question is therefore not whether another site could theoretically take over. It is whether replacement capacity exists, whether the workload can move, and who pays for the transition.

An inexpensive facility can become expensive once idle hardware, spare capacity and recovery costs are included.

The meaningful comparison is cost per unit of useful compute delivered at the required service level, not electricity price or facility rent in isolation.

Where the company captures value

This does not point to one universally superior position in the value chain. It points to different ways of controlling something customers need.

A compute provider can own the customer relationship, software environment and allocation of hardware. Its opportunity is to make compute a service customers depend on. But owning more GPUs does not establish that dependence. The business must justify its margins after hardware investment, replacement and the cost of keeping capacity productively occupied.

A powered-infrastructure provider can control the delivery of qualified capacity: electricity, cooling, connectivity and operations. It may leave hardware ownership to the customer, but it takes responsibility for bringing the facility online and meeting performance commitments. Its advantage could be a repeatable way to deliver capacity that competitors cannot easily match.

An enabling-technology provider can improve the economics of both. Power conversion, cooling, storage or control systems may become embedded in many facilities without their supplier owning those facilities.

The largest revenue pool is not necessarily the most attractive venture opportunity. Nor does sitting closer to software automatically produce software-like economics.

Our question is what the company controls that a customer cannot readily replace.

A scarce site can be a valuable asset. A standardized design, qualified supply chain and operating system that make many sites viable could support a broader company. The distinction is whether growth multiplies an advantage or merely adds assets.

This also explains why vertical integration should be a choice rather than a default. Owning additional layers may accelerate learning or improve the customer proposition. It may also add capital requirements and risks outside the company’s expertise.

The right boundary is where integration strengthens the advantage—not where it produces the largest reported revenue.

Project finance is earned

A potentially scalable structure separates the technology company from the capital required to own individual facilities.

Corporate equity can support technology, engineering, customer development and operating capability. Project-level investors can fund specific assets against their contracts and expected cash flows. The company may earn technology margins, development and operating fees, and returns on retained ownership.

But separating the entities does not automatically separate the risks.

Core Scientific’s April 2026 financing disclosure described an uncapped parent completion guarantee: the company would provide funds needed to finish the projects if available financing proved insufficient. Its customer agreements also included late-delivery credits and potential termination rights following repeated or prolonged service failures. Core Scientific, Inc.

Those obligations are not evidence of a defective structure. They show what financing and customer commitments can require.

For a new architecture, access to outside capital must be earned through operating evidence, credible construction costs and contracts that lenders can underwrite. A customer’s interest in a demonstration is not equivalent to its willingness to sign a long-term capacity commitment.

The economics must reconcile as well. Rent belongs to the entity providing the contracted capacity. Fees and margins paid to the corporate platform come out of the project’s economics; they are not an additional pool of value independent of what remains for the capital providers.

The question we would ask is:

As deployment grows, does the corporate capital—and contingent liability—required for each additional project decline?

If every installation requires extensive redesign, another parent guarantee and a substantial equity contribution, the company may be expanding its exposure faster than its capabilities.

If a common design, qualified equipment and operating record reduce those requirements, it can begin to scale more than its balance sheet.

Capital structure / 02

Separate the asset. Trace the obligation.

Project financing can fund a facility while some completion risk remains with the parent.

Corporate platform

Technology & delivery capability

Corporate equity supports engineering, customer development and operations.

Project company

Facility & customer contract

Project investors fund an asset against contracted capacity and expected cash flows.

The risk bridge

A parent guarantee can reconnect them

If project financing falls short, the parent may have to supply the funds needed to finish construction.

The advantage must survive the shortage

There is a credible countercase to the broader opportunity.

Existing operators may accommodate more demand than expected. Improvements in hardware and software may change the amount of electricity required. Financing conditions may slow projects before engineering does. The IEA explicitly identifies efficiency, adoption, changing applications and access to capital as important uncertainties in the demand outlook. IEA

A new company can therefore be right about the shortage and still build the wrong business. Its delivery advantage may arrive too late, cost too much, or disappear before it has recovered the investment needed to establish it.

For Aldea, three tests help distinguish an enduring company from a temporary opportunity.

The first is customer economics. Does the solution improve an actual deployment after accounting for reliability, connectivity, recovery and the cost of capital? Interest is useful evidence. Acceptance of binding commercial terms is stronger.

The second is repeatability. Does the next installation reuse meaningful engineering and qualification work? Does operating experience improve the design? Can the team reduce delivery uncertainty without depending on exceptional effort from the same few people?

The third is retained economics. As outside capital funds expansion and competitors improve, why does the company continue to earn attractive margins? The answer needs to rest on proprietary performance, customer dependence or accumulated execution capability—not merely a shortage.

These tests also sharpen the question of global scale. A successful installation in one location proves something important. A second deployment under different local conditions begins to show which advantages belong to the company and which belonged to the original site.

The opportunity is not limited to owning power plants, data centers or GPUs. It includes making those assets more productive and making previously difficult capacity usable.

What matters is whether the capability becomes repeatable—and whether the company retains the value it creates.

Scarcity can make the first project valuable. The company is built by making the next one easier.