Emerald Conductor
by Emerald AI
Makes AI data centres power-flexible so utilities will connect them faster
Emerald Conductor is a flexibility-management platform that lets AI data centres cut their power draw on demand — by pausing batchable jobs, shifting work between regions and dispatching on-site batteries — without breaching workload SLAs. It is aimed at operators stuck in interconnection queues and at utilities unwilling to energise capacity they cannot control during grid stress.
Emerald AI, founded in 2024 and headquartered in Washington, D.C. under CEO Varun Sivaram, sells Emerald Conductor: a flexibility-management platform that lets AI data centres modulate power draw in response to grid conditions without breaching workload service-level agreements. The problem it addresses is the interconnection queue — utilities will not energise new AI capacity they cannot rely on — and the company argues that treating data centres as controllable rather than fixed load could unlock more than 100GW on the existing US grid. Conductor works across three axes. Temporal flexibility briefly slows or pauses batchable AI workloads during a grid event and resumes them afterwards inside SLA guardrails. Spatial flexibility shifts work to regions where power is abundant, subject to latency constraints. Resource flexibility dispatches on-site batteries and generation alongside the compute schedule. A digital twin of the facility's power behaviour connects the site to the utility in real time and produces the verified telemetry that grid programmes require. In a May 2026 demonstration with Oracle Cloud Infrastructure, NVIDIA, Salt River Project, Arizona Public Service and EPRI, a Phoenix cluster cut power draw 25% for three consecutive hours during a genuine heat-driven grid event with every workload inside SLA; the platform can shed roughly 30% of load within 30 seconds for emergency curtailment. Five commercial demonstrations have run across Arizona, Illinois, Virginia, Oregon and London. On 25 August 2026 Emerald AI closed a $150M Series A at a $1.05B valuation co-led by Energize Capital and DCVC, taking total funding past $220M.
The data-centre or infrastructure executive whose AI build-out is blocked by an interconnection queue, and who can trade a controlled amount of compute flexibility for an earlier energisation date or a cheaper flexible-load tariff.
Grid capacity you can actually get connected — demonstrated as a 25% power reduction sustained over three hours with all AI workloads still inside their SLAs.
At a Glance
- Category
- Infrastructure & Cloud
- Pricing
- Contact for pricing
- Target Market
- CIOs, CTOs, Data Center Operators, Utility and Grid Planners
- Deployment
- Cloud-first
- Founded
- 2024
- Headquarters
- Washington, D.C., United States
- Customers
- Not disclosed; five commercial demonstrations across Arizona, Illinois, Virginia, Oregon and London, plus a named deployment with Silicon Valley Power and the ~100MW Vera Rubin AI Research Factory in Manassas, Virginia
Key Features
- ✓Temporal flexibility
Briefly slows or pauses batchable AI workloads during a grid event and resumes them afterwards, all within SLA guardrails.
- ✓Spatial flexibility
Shifts workloads to regions where power is abundant while respecting the latency constraints of the jobs being moved.
- ✓Resource flexibility
Coordinates on-site batteries and generation alongside compute orchestration so energy assets and workloads are dispatched together.
- ✓AI power digital twin
Simulates the facility's power behaviour so operators and utilities can model a curtailment event before committing to it.
- ✓Real-time utility interconnection
Connects data centres and utilities live, so grid signals reach the workload scheduler within seconds rather than by phone call.
- ✓Verified performance reporting
Produces auditable evidence that a curtailment actually happened, which flexible-load tariffs and grid programmes require for payment.
- ✓Emergency curtailment
Sheds roughly 30% of a data centre's load within 30 seconds to support grid resilience during an acute event.
Capabilities
Use Cases
- •Faster grid interconnection
Offering controllable load lets an operator qualify for flexible-load interconnection programmes and energise capacity years ahead of a firm-load queue position.
- •Peak-event curtailment
During the May 2026 Phoenix heat event a cluster cut draw 25% for three hours with all workloads inside SLA.
- •Carbon-aware compute dispatch
Workloads track five-minute grid CO2 signals so batch training runs land in cleaner hours without operator intervention.
- •Long-duration demand response
Sustains curtailments of up to ten hours against tariff or market signals, converting a fixed cost centre into a grid asset.
- •Coordinating on-site energy with compute
At the ~100MW Vera Rubin AI Research Factory in Manassas, batteries and workload scheduling are managed as one system.
Ideal For
Best For
- ✓AI data-centre operators waiting in utility interconnection queues who can trade flexibility for an earlier connection date
- ✓Colocation and hyperscale sites participating in flexible-load interconnection programmes such as Silicon Valley Power's
- ✓Operators with a meaningful share of batchable training or fine-tuning work that can absorb short pauses
- ✓Campuses with on-site batteries or generation that need compute scheduling coordinated with energy dispatch
- ✓Utilities and grid operators seeking verified, auditable curtailment performance from large new loads
Not Ideal For
- ✗Sites running only latency-critical inference with no batchable workload — there is nothing to defer, so temporal flexibility has no lever to pull
- ✗Small or single-tenant facilities whose load is too small to matter to a grid operator or to earn flexible-tariff treatment
- ✗Operators in territories whose utility offers no flexible-load tariff or demand-response programme, since the economic return depends entirely on that market design
- ✗Teams that want published pricing and a self-serve trial; this is a negotiated deployment involving the utility as a third party
Deployment
Market & Ratings
Not disclosed; five commercial demonstrations across Arizona, Illinois, Virginia, Oregon and London, plus a named deployment with Silicon Valley Power and the ~100MW Vera Rubin AI Research Factory in Manassas, Virginia
Market Analysis
Pros
- ✓Results are from live commercial events, not simulation — 25% reduction sustained three hours in Phoenix with workloads inside SLA, and roughly 30% shed within 30 seconds for emergencies
- ✓Directly addresses the binding constraint on AI capacity today, which is grid interconnection rather than chips or capital
- ✓Strategic investor list doubles as a distribution and validation channel: NVIDIA, Siemens, GE Vernova and RWE are all in the round
- ✓Utility-facing verification and digital-twin modelling are what make flexible-load tariffs bankable, and few software vendors supply that piece
Cons
- ✗The business depends on utilities and grid operators — operationally conservative institutions — changing long-held interconnection and tariff practices, which is slow and outside Emerald AI's control
- ✗Critics argue flexibility distracts from the genuine need to build grid infrastructure faster and could itself pose supply risks
- ✗A widely cited economic objection is that curtailing GPUs never pays: an idle accelerator is expensive, so the case collapses unless utilities offer differentiated service tiers
- ✗Founded in 2024 with results drawn from pilots and demonstrations; there is no long-run production track record and no disclosed customer count
- ✗No published pricing, no self-serve path, and value realisation requires a three-way negotiation with the local utility
Pricing
Enterprise deployment
Contact for pricing
- ✓Emerald Conductor flexibility platform
- ✓Utility and grid-operator integration
- ✓Digital twin modelling and verified curtailment reporting
No pricing is published on the website or in any funding coverage. This is sold as a negotiated enterprise engagement involving the data-centre operator and its utility together, and the commercial case depends on local tariff design — flexible-load interconnection terms, demand-response payments or avoided capacity charges — rather than on a software list price. Expect procurement to run alongside a utility agreement rather than as a standalone SaaS purchase.
Security & Compliance
Sources
This page was written from 5 sources, 4 on domains other than emeraldai.co.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
Nebius AI Cloud
European full-stack AI cloud with published GPU pricing and enterprise compliance
Etched Sohu
Transformer-only inference ASIC shipped as rack-scale frontier inference clusters
fal
Serverless generative-media inference: 1,000+ image, video, audio and 3D models behind one API
Daytona
Sub-90ms stateful sandboxes that give every AI agent its own disposable computer