For most of the past decade, the working assumption across the data centre industry was straightforward: secure enough GPUs and you could run AI. That assumption, according to John Belizaire, CEO of Soluna, no longer holds.

"The real constraint today is energy access," John says. "Grid interconnection queues in the US now stretch five to seven years in many markets. Utilities are operating at or near capacity in regions with the highest data centre demand. The permitting and buildout timelines for new transmission infrastructure are measured in decades, not years."

The consequence is that a developer can secure the capital, the hardware and the land for a new AI data centre and still face years of delays waiting for power.

John is direct about what this means for the industry: "The conversation has shifted from 'Where can we build?' to 'Where can we get power, and fast?' Speed to power is now the determining factor in where and how fast AI infrastructure gets built."

It is a structural shift, not a temporary squeeze. The energy system was not designed to absorb demand at this pace, and the gap between what AI infrastructure requires and what the grid can deliver has become the defining infrastructure problem of the moment.

Curtailed renewable energy offers untapped opportunity

One response to that gap lies in a resource that is currently being wasted at scale. Curtailment – the process by which renewable energy assets generate more power than the grid can absorb – results in significant volumes of electricity being produced and then discarded. In the US, curtailment levels at wind and solar assets continue to rise annually, a figure that grows as more renewable capacity comes online without corresponding grid upgrades.

John sees this as one of the most underutilised opportunities in energy infrastructure today.

"AI and HPC workloads are well-suited to absorb this energy," he says. "They are computationally intensive, they can operate at scale, and certain workloads can be structured with enough flexibility to ramp up when power is available and modulate when it is not. That flexibility is exactly what curtailed energy needs."

The economic logic reinforces the operational case. Curtailed energy is often available at low cost precisely because it would otherwise go to waste, creating what John describes as "a structural cost advantage for computing infrastructure built to absorb it". At the same time, the renewable project itself benefits, converting what was a sunk cost into revenue without requiring additional generation capacity.

"This is not a theoretical model," he adds. "Soluna has been operating this way for several years, and the fundamentals hold."

Designing data centres around energy assets

What distinguishes Soluna's approach at the operational level is the sequence in which its data centres are conceived. Where a conventional facility begins with the building and then pursues a power connection, Soluna starts with the energy asset and constructs computing infrastructure around it.

"Our facilities are colocated directly with renewable generation, connected behind the meter," continues John. "The data centre and the power source are integrated at the physical and operational level. That integration allows us to respond to the actual generation profile of the asset in real time rather than assuming a flat, constant power supply."

That means designing for variability as a feature rather than a flaw. Renewable generation fluctuates and Soluna's infrastructure is built to treat those fluctuations as a design input. Computing loads are structured to operate efficiently across a range of power states, supported by the company's proprietary energy management software, MaestroOS, which handles dynamic allocation across the facility.

The building design philosophy follows accordingly. Soluna opts for modular, purpose-built facilities rather than traditional hyperscale structures – infrastructure that can be staged and expanded in line with energy capacity rather than built speculatively ahead of it.

"The result is a data centre that is genuinely aligned with its power source rather than simply co-located with it in a geographic sense," John says.

Modular infrastructure accelerates deployment in power-constrained markets

In markets where energy access is the primary constraint, the ability to move quickly on available capacity – rather than waiting for conditions sufficient to justify a large-format build – carries considerable operational and financial weight.

Power access rarely materialises all at once. A renewable site might have 50MW of available capacity today, with expansion potential developing over time as generation is added or curtailment patterns shift. A traditional large-format data centre is poorly suited to that trajectory.

"Modular infrastructure solves that," says John. "You can deploy initial capacity in line with available power, generate revenue, validate the site and expand incrementally as the energy picture develops. Each module is a functional unit of infrastructure, not a fraction of a building waiting to be completed."

The approach also compresses the timeline from commitment to operation and reduces capital risk. Rather than committing to a full buildout before site performance has been demonstrated, operators can move on what is available and scale deliberately from there.

"In markets where energy is the primary constraint, that speed and flexibility is a meaningful advantage," John notes.

Data centres as active grid participants

Beyond absorbing curtailed energy, Soluna's model incorporates a more active relationship with the wider electricity system. Its facilities operate using both behind-the-meter power from renewable generation and grid power under normal conditions. The flexibility comes into play when grid conditions change.

"When a utility faces a demand surge or supply emergency, our data centres can respond to that signal and modulate down, freeing power back to support grid stability," John explains. "That is demand response in practice."

The distinction he draws is important. This is not a matter of flexing because the wind is blowing less, but “because the grid needs support and our infrastructure is designed to provide it”. 

John adds: “That responsiveness has real value to grid operators, and it positions computing infrastructure as an active participant in energy system management rather than just a consumer."

Making this work operationally depends on workload composition. Certain training tasks, batch inference and back-end processing tolerate planned modulation far better than latency-sensitive applications. Designing infrastructure to separate these workload types allows operators to participate in demand response programmes without degrading performance on applications that require consistency.

"The broader point is that flexible load transforms a data centre from a fixed draw on the energy system into a grid asset," says John. "That changes the conversation with utilities, regulators and renewable projects themselves."

Stranded energy creates mutual value

The commercial logic underpinning Soluna's model rests on a straightforward observation about stranded energy: it represents a cost that has already been incurred. The generation infrastructure is built, the fuel is free and the power is being produced. If it cannot reach a buyer, it is lost.

"For a computing operator willing to build at the source, that dynamic becomes an opportunity," asserts John. "You are not competing with grid buyers for power. You are creating a market for energy that would otherwise be nonexistent. That positions you to negotiate power costs that reflect the reality of curtailment rather than the premium of scarcity."

For the renewable project, the benefits extend beyond immediate revenue. Absorbing curtailed energy improves the project's capacity factor and, in some cases, strengthens the underlying financial model. Projects that faced marginal economics with curtailment as a known variable become more viable with a committed computing offtaker.

"It is not simply that Soluna gets cheap power," John says. "It is that the renewable asset and the computing infrastructure are structurally better off together than they are separately."

Defining the industry's next phase

Looking further ahead, John identifies two structural shifts that he believes will reshape the data centre industry. The first is vertical integration between energy and compute. Soluna's acquisition of the Briscoe Wind Farm in West Texas is a direct expression of this thesis.

"Operators that control their power supply will have durable advantages over those dependent on utility allocation," he says. "As energy access becomes more competitive, ownership or long-term control of generation assets will define which infrastructure platforms can scale reliably and which ones are constrained."

The second shift concerns geography. Data centre development has long been concentrated in a small number of established markets, a pattern shaped by where fibre and power infrastructure already existed. As AI drives demand beyond what those markets can absorb, and as renewable generation expands into new regions, John expects that pattern to break.

"Computing will begin to follow energy rather than the other way around,” he says. “That rebalancing will open new geographies and require new infrastructure models."

Taken together, these trends point toward a structural reorganisation of where and how digital infrastructure gets built. 

"AI demand will continue to outpace the grid's ability to deliver power through conventional channels," John says. 

"The data centre industry will have to build its way around that constraint. Renewable computing is one response to that challenge, but the broader shift toward energy-first infrastructure thinking is not a trend. It is a structural reorganisation of where and how digital infrastructure gets built."