.png)
Coordination is now the binding constraint on AI-era infrastructure
Buying behavior in power markets is changing fast. For decades, electricity was a slow-moving commodity. Customers could tolerate long lead times and were cost-conscious above all else. That’s no longer the case. Large scale AI and electrification workloads are now the fastest growing source of power demand, and they care far more about time-to-power than the cost of electricity. We think flexibility and load orchestration may get them to power faster than building their own co-located power plants and that they will increasingly turn to Grid-Enhancing Technologies (GETs). GETs aren’t a new concept, but now that hyperscalers are fixated on time-to-power, we think there will be an explosion in demand for GETs from a new, fast-moving, deep-pocketed customer.
A GPU cluster or a training run is ready to deploy in 18 months, but a traditional grid interconnection takes eight years. That’s an existential threat to a hyperscaler. Most of the conversation about the emerging energy constraint has focused on the supply side, examining new generation technologies (nuclear, geothermal, turbines, etc), power plant construction, transmission buildouts, and ever-larger capex plans.
What’s been underappreciated is how much power capacity is locked up by conservative grid planning assumptions and inflexible infrastructure. Because the grid is relatively rigid today, we have a lot of backup power that sits idly on the grid for those peak electricity demand days that only happen a few times a year (e.g. hot summer days). There is enough capacity on the grid already to meet the projected data center electricity demand over the next five years. This is why we view the energy constraint as a failure of coordination instead of a shortage of steel, copper, turbines and fuel.
The "Worst-Case" capacity trap
To understand why coordination is the unlock, we have to look at how the grid is currently planned. Historically, utilities have used "brute-force" construction to meet worst case scenarios. Think of the power grid like a massive highway system. To ensure that traffic never crawls even on a busy holiday weekend, the utility builds 12 lanes in both directions. For 360 days a year, 8 of those lanes sit effectively reserved for rare peak events. Because the utility’s mandate is to ensure the lights stay on during a record-breaking heatwave in August, they treat the grid as full the moment those holiday-peak lanes are accounted for.
This "worst-case" modeling locks up massive amounts of idle capacity. It’s constrained by the assumption that every load is inflexible and must be served during those rare peak hours.
Unlocking the idle lanes
This is where the opportunity for a massive shortcut lies. If new, large-scale loads (like data centers and factories) can be smarter and more flexible than a typical residential home, they can access those empty lanes without putting people at risk on peak demand days. In grid planning terms, they have to prove flexible “curtailment,” which means they can tolerate consuming less power when the grid needs them to.
If a data center agrees to a minimal curtailment model for just a few hours a year (less than 1% of the time) during those extreme heatwaves, the math changes instantly. According to the Nicholas Institute, this level of flexibility could unlock 76 to 126 GW of additional demand from our existing grid assets. To put that in perspective, most industry estimates suggest we need between 100 and 130 GW of new capacity to meet total data center demand through the end of the decade. If we can unlock 76-126 GW simply by coordinating existing assets more intelligently, we solve the majority of the data center power problem without waiting a decade for new transmission lines. This is why we are so excited about GETs and flexibility enablement.
Grid-enhancing technologies and why they matter now
Grid-enhancing technologies, as traditionally defined, are a set of hardware and software tools designed to increase the capacity, reliability, or utilization of the existing power grid without building new generation or transmission. Common examples include:
- Dynamic Line Rating (DLR): Adjusts transmission capacity based on real-time weather conditions.
- Advanced Sensors: Improve situational awareness and monitoring.
- Power-flow Control: Redirects electricity around congested paths.
- Forecasting & Optimization Software: Allows operators to run the system closer to its physical limits.
Historically, GETs struggled with adoption because they were framed almost exclusively as utility-facing solutions, which never aligned with utilities’ incentives. Utilities have always been incentivized to favor capital buildout (building more assets) over optimization. From the utility perspective, the benefits of GETs accrued to ratepayers, while the costs and integration risks were still on them.
The buyer dynamic around GETs has now flipped. In today’s environment, GETs are capabilities that large load buyers actively want because they believe it helps them get approved for interconnection sooner. It’s more about speed-to-power than marginal optimization. Flexibility, when made legible and enforceable, makes large load additions more palatable to regulators. It reduces perceived system risk and smooths approval pathways.
Under this broader framing, GETs are best understood as coordination technologies. They can’t change the underlying physics of the grid, but they can change how uncertainty is managed.
Where value actually accrues
The opportunity created by this shift in demand can be understood as a stack of solutions running from physical power to orchestration to capital.
At the base is physical acceleration. Modular generation, rapidly deployable substations, transformers, and switchgear exist because the system cannot wait, making buyers price-insensitive to delays. These capital-intensive assets create the operating envelope for coordination technologies.
Above that sits visibility. Much overbuilding is due to ignorance; operators lack real-time, probabilistic insight into system constraints. Software that dynamically maps headroom, forecasts stress, and replaces static reserve margins changes decisions upstream before capex commitment.
The next layer is control and orchestration, offering the most attractive risk–return profile for venture. Large AI loads are volatile. Grid operators care about behavior under stress: ride-through, failover, and predictability. Hardware-enabled software platforms coordinate on-site generation, storage, and grid interaction, turning large loads into manageable system participants. This layer will increasingly be a condition of interconnection or tariff eligibility.
Above control sits planning and interconnection compression. Interconnection queues are coordination failures. Tools that pre-qualify projects, reduce speculative noise, and allow planners to reason about flexible behavior unlock enormous value. Shaving months off a study cycle can unlock billions in downstream investment.
Finally, there is capital and risk infrastructure. Flexibility only scales when trusted. Insurance, guarantees, underwriting, and structured finance that price curtailment risk, uptime, and fuel exposure turn flexibility from an operational promise into a financial asset. As balance sheets determine who moves fastest, this layer becomes decisive.
We believe that visibility, control and orchestration, and planning offer the lowest capital intensity for VC-backed, asset-light models to start, but ultimately, all these layers will be necessary for a resilient, flexible grid.
Startup ideas implied by this thesis
Several concrete company archetypes fall directly out of this stack:
- Flexible Interconnection Platform (“ERIS-for-load”): A product that combines hardware and software to give large energy users a fast, guaranteed way to connect to the grid. It offers speed to the customer and predictable behavior to the utility.
- AI Load Stability Controller (Behind-the-Meter OS): A smart control system for large data centers/AI loads that smooths out power spikes, manages backup power, and allows the grid operator to safely manage the load without impacting the customer's work.
- Grid Headroom Intelligence Platform: Software that forecasts exactly how much spare power capacity (headroom) is available across the grid at different times. This helps planners, regulators, and developers choose the best location for new large energy users.
- Curtailment Guarantees and Risk Infrastructure: An insurance and finance product that prices the financial value of a customer's willingness to temporarily reduce their power use (flexibility). This allows flexible load to be treated and financed like traditional, reliable power infrastructure.
- Interconnection Queue Intelligence & Hygiene: Tools that clean up the grid's waiting list for new projects by filtering out speculative applications and standardizing technical information. This frees up capacity and time without building new physical infrastructure.
The synthesis
The dominant constraint in the next phase of grid scaling is coordination under uncertainty, and we think GETs are how the system adapts to that reality. They allow institutions to bend without breaking by making flexibility visible, enforceable, and ultimately bankable. In doing so, they enable faster growth than infrastructure timelines can alone.
The old mental model was linear. Build more grid to serve more load. The emerging model is nonlinear. Make load smarter, and the grid effectively gets bigger. In a world where time-to-power is strategic, GETs are foundational infrastructure for the AI and electrification economy.