A startup just raised $150 million to make data centers throttle themselves during grid stress. That's a new line item on your next electrical bid
Emerald AI closed a $150 million round on August 25 for software that cuts a data center's power draw on a utility signal instead of waiting years for new transmission. If an owner adopts it to jump the interconnection queue, the switchgear and controls scope on that job just got bigger — and your early estimate probably doesn't have it.
Emerald AI raised $150 million on August 25 for software that makes a data center automatically cut its own power draw when the grid is under stress — a bet that utilities will grant faster grid connections to facilities that can prove they'll throttle themselves instead of demanding full peak capacity around the clock. If that trade catches on with the owners you build for, the electrical and controls scope on your next data center job changes before a shovel ever hits dirt.
What did Emerald AI actually raise, and for what?
The Series A, co-led by Energize Capital and DCVC, values the company at $1.05 billion and brings its total funding past $220 million. Emerald AI's software, Conductor, is already running commercially at multi-megawatt, full-data-center scale for AI companies, data center operators, and electric utilities, according to the company's announcement. Nvidia backed an earlier round and has published case studies showing the software cutting a data center's grid draw by roughly 30 to 40 percent in under a minute and holding that reduction for up to 10 hours before ramping back up.
Why would an owner want their data center to throttle itself?
Because the alternative is waiting years. The U.S. interconnection queue — the line of projects waiting for a utility to study and approve their grid connection — has swollen to roughly 2,600 gigawatts of pending requests. RMI puts the median wait at close to five years to reach commercial operation, with some data center projects in congested regions facing a decade or more. Utilities size that queue to a site's worst-case peak draw. A facility that can prove, automatically and on short notice, that it will cut load during a grid emergency needs a smaller firm-capacity reservation — a much faster study to approve. That's the "fast pass" pitch Emerald AI and Nvidia have made to utilities since an earlier round in March: flexibility traded for speed to power.
What changes on the construction side if an owner takes that trade?
A standard data center electrical design assumes firm power: size the switchgear and generators for peak load, commission for uptime, done. A curtailable interconnection adds a layer that has to be designed, wired, and commissioned on top of that:
| Firm-power design | Curtailable-power design |
|---|---|
| Switchgear sized to worst-case peak draw | Switchgear plus automated load-shedding controls tied to a utility signal |
| Generators/UPS sized for continuous full load | Generator/UPS coordination that can ride through a 30–40% draw cut without tripping protected IT loads |
| Commissioning tests for steady-state uptime | Commissioning tests for a full curtailment event: signal receipt, load-shed, and ramp-back within minutes |
| Interconnection study measured in years | Interconnection study potentially measured in months, but tied to controls acceptance testing |
That's real scope — utility-signal integration, building management system programming, and a commissioning sequence that doesn't exist in a firm-power spec — and it's the kind of addition that shows up as a change order when it should have shown up in the original bid.
What should an estimator or electrical sub actually do with this?
- Ask early which kind of interconnection agreement the owner has, or is pursuing, before pricing switchgear and controls. Firm versus curtailable is now a design input, not a footnote.
- Flag curtailment-readiness as a distinct scope item on RFIs for any data center pursuit, especially on projects where the owner or their energy advisor has mentioned grid flexibility, demand response, or a compressed interconnection timeline.
- Expect the interconnection strategy to change mid-design on some jobs. An owner facing a multi-year queue on firm power has a real incentive to switch to a curtailable agreement partway through design development — budget for that scope to show up late rather than assuming the electrical package is locked.
- Treat the commissioning plan as its own line item. Testing an automated load-shed sequence against a utility signal is a different — and longer — commissioning scope than proving a system holds steady at full load.
The interconnection queue isn't getting shorter on its own, so more owners will look for a way around it. Whether or not Emerald AI specifically becomes the standard, the underlying trade — smaller firm-capacity ask in exchange for automated curtailment — is the kind of interconnection strategy an estimator now has to ask about before the electrical scope gets locked in.
For how the capex behind these projects is moving, see our look at Nvidia's $1.3 trillion 2027 hyperscaler capex number and what it means for data center schedules.
Friday one chart. Every week, one piece of data that should change a decision on your project. Subscribe at constructionaibrief.com.
- What is Emerald AI and why did it raise $150 million?
- Emerald AI makes software, called Conductor, that automatically throttles a data center's power draw when the local electric grid is under stress, then ramps it back up once the peak passes. It raised a $150 million Series A on August 25, 2026, co-led by Energize Capital and DCVC at a $1.05 billion valuation, to expand commercial deployments with data center operators and utilities.
- How much can this software cut a data center's power draw, and how fast?
- According to Nvidia's and National Grid's published case studies on the technology, it can cut a data center's grid draw by roughly 30 to 40 percent within about a minute, sustain that reduction for up to 10 hours, and respond to notices as short as 10 minutes or as long as 2 hours, depending on how the utility signals the event.
- Does grid flexibility actually get a data center connected to the grid faster?
- That is the pitch. Utilities size an interconnection to a site's worst-case peak demand. A data center that can prove it will cut load automatically during system peaks needs a smaller firm-capacity reservation, which can shrink the utility's interconnection study and get power turned on years sooner than a standard firm-power request.
- How long is the typical data center waiting for a grid connection right now?
- The U.S. interconnection queue has grown to roughly 2,600 gigawatts of pending requests, and the median project now waits close to five years to reach commercial operation, with some data center projects facing waits of a decade or more in congested regions, per RMI's 2026 analysis.
- What should an estimator or electrical sub check before pricing a data center's power scope?
- Ask whether the owner's interconnection agreement is firm or curtailable before finalizing switchgear, generator, and controls pricing. A curtailable interconnection typically requires additional automated load-shedding controls, utility-signal integration, and commissioning scope that a firm-power design does not need — and that scope is easy to miss on an early, incomplete set of drawings.