Emerald AI founder and CEO Varun Sivaram, the energy executive and former U.S. diplomat, has disclosed $90,229,639 sold toward a planned $150,000,000 equity offering, according to a Form D filed August 3, 2026.
The filing says Emerald AI has $59,770,361 left to sell under the offering. It identifies Sivaram as chief executive and director, alongside directors Rob Toews and Markus Specks, but leaves the lead investor, valuation, share price and participating investors unidentified.
For Sivaram, the offering represents a rapid expansion of a thesis he developed over 15 years working across energy companies and government: building more power supply cannot solve the AI industry's near-term electricity constraint by itself. Sivaram previously served as chief strategy and innovation officer at offshore wind developer Orsted and chief technology officer at ReNew Power. He also advised John Kerry on clean energy during the Biden administration. A Rhodes Scholar with a doctorate in condensed-matter physics from Oxford, Sivaram founded Emerald AI in November 2024 after concluding that electricity demand itself needed to become more responsive. In an October 2025 funding announcement, Sivaram wrote that "only building more supply couldn't work on its own." Emerald AI's answer is software that can temporarily slow, move or reschedule selected AI workloads when the power grid is under strain.
A steep funding acceleration
Emerald AI had announced $68 million in total funding before the new offering. That included a $24.5 million seed round led by Radical Ventures in July 2025, an $18 million extension led by Lowercarbon Capital on October 29, and a $25 million strategic expansion round led by Energy Impact Partners on March 31.
If the new Form D is fully incremental to those announced rounds, Emerald AI's disclosed sold and announced capital would reach at least $158.2 million. Emerald AI has not announced that combined figure, and the filing does not establish whether the $90.2 million includes capital from any previously announced round. The step-up is substantial for a company founded less than two years ago. The initial seed brought backing from Radical Ventures, NVentures, Amplo, CRV and Neotribe. The later rounds added Lowercarbon Capital, Salesforce Ventures, National Grid Partners, Eaton, GE Vernova, In-Q-Tel, Samsung Ventures and Siemens, among others.
The new filing provides governance detail without naming the new investors. Rob Toews, a partner at early backer Radical Ventures, remains an Emerald AI director. Markus Specks, founder and chief executive of energy investor Aventurine Partners, is also listed as a director. Their board positions establish governance ties around Emerald AI, though the filing does not identify which investors supplied the capital disclosed in the offering.
Sivaram is selling flexibility as infrastructure
Emerald Conductor connects data centers with utilities and grid operators, models the power demand of AI workloads and adjusts compute or onsite energy resources in response to grid conditions. Emerald AI says the software maintains service-level and quality-of-service requirements while reducing or shifting electricity consumption.
The distinction matters because AI data centers contain workloads with different timing constraints. Real-time inference may need to continue uninterrupted, while model training, fine-tuning and other batch jobs can sometimes slow down or . Conductor is designed to find that room automatically, including by moving work between regions when latency requirements allow.
Emerald AI has produced credible early technical evidence. A field demonstration published in Nature Energy used a commercial cloud cluster with 256 GPUs in Phoenix. The trial reduced cluster power consumption by 25% for three hours during a peak grid event while maintaining the workloads' stated quality-of-service guarantees. The researchers described the approach as software-only, with no additional storage or data-center hardware required.
That result established that flexible AI compute can work under controlled field conditions. It remains a limited proof point: one 256-GPU cluster, one three-hour period and a selected mix of representative workloads. Emerald AI's next test is repeated dispatch at facilities large enough to matter to utilities, while protecting the performance and economics that data-center operators promise their customers.
Sivaram has started moving toward that commercial test. In June, Emerald AI announced a multi-megawatt deployment with NVIDIA and Silicon Valley Power in Santa Clara. The municipal utility is developing a framework that can grant data centers additional grid access in exchange for verifiable power reductions during constrained periods. Emerald AI says Conductor will receive utility signals and modify a participating facility's consumption while NVIDIA workloads continue running.
Emerald AI has also described a planned 96-megawatt deployment at an AI facility in Manassas, Virginia, with NVIDIA, Digital Realty, EPRI and PJM Interconnection. That project was scheduled for later in 2026 in Emerald AI's June announcement. Moving from demonstrations to recurring utility-directed operation at that scale would provide a clearer measure of Conductor's commercial value.
The round finances a race for grid access
The timing follows a shift in the data-center market. Power availability has become a constraint on how quickly new GPU capacity can be installed, pushing operators to seek faster grid connections, onsite generation and sites that already control large power allocations. Sivaram is betting that verifiable demand reductions can become another route to earlier access.
That creates a direct economic incentive for customers. A data center that can prove it will reduce consumption when requested may receive power sooner or avoid waiting for every transmission upgrade required by a fully inflexible load. Utilities gain a controllable resource during peak periods. Emerald AI can position its software inside the operating stack that mediates the agreement.
The $150 million target gives Sivaram capital to pursue that position before utility rules and data-center architectures settle around competing approaches. The hard part will be turning technical flexibility into contracts with measurable value. Data-center operators need guarantees that workload performance will hold, while utilities need reductions they can dispatch, audit and trust.
Emerald AI's Form D shows investors are financing that transition ahead of full commercial proof. The remaining $59.8 million would take the offering to $150 million, giving Sivaram a much larger balance sheet for deployments, utility integrations and product development. The financing also raises the standard for the next set of results: repeatable power reductions across commercial fleets, rather than isolated demonstrations.