Emerald AI Raises $42.5 Million to Power the AI Revolution and Protect the Grid
AI is the defining technology of our time, but it runs on the same electric grid built for the rotary phone.
Training a new AI model requires entire fleets of GPUs (graphics processing units, the specialized processors used in AI). For context, OpenAI’s GPT-5 was trained on over 200,000 GPUs, compared to an estimated 25,000 for GPT-4. Each of those GPUs requires 550W - 1,200W of power. When you add all of that up, training a new model today can consume as much energy as a small town.
New U.S. data centers could require an additional 50 - 100 GW of power by 2030, but it’s not yet clear if they’ll be able to get it. That’s because the infrastructure that powers it all, our electric grid, is failing.
The grid is the vast, interconnected system that delivers electricity across the country, balancing supply and demand every second in real time. It’s arguably the largest machine on the planet, but it was built a century ago and never designed for the AI era. The grid is built to handle peak demand–like when everyone turns on their air conditioner on a hot day–and has lots of spare capacity outside of those hours. This means the grid can transmit way more energy than it does currently if that demand could be evened out.
Interconnection queues, the waiting line to get plugged into the grid, can stretch 7 to 10 years in some places. Meanwhile, grid regulators and operators warn that uncoordinated demand surges from data centers could push utilities to overbuild new infrastructure like transmission lines and substations. New infrastructure is paid for through higher electricity rates, meaning consumers foot the bill.
As a result, some regulators are slamming the brakes and pulling the plug on plans. Left unchecked, these tensions threaten to slow down AI deployment, not because of technological limits, but because of regulatory inertia and poorly utilized infrastructure.
We’re watching a trillion dollar technology revolution hit a century old wall of regulation and infrastructure. That’s where Emerald AI comes in.
Emerald Helps AI Help the Grid
Emerald is reimagining how data centers draw power. Instead of treating data centers as a liability, Emerald’s platform turns them into grid assets that are flexible, intelligent, and responsive. Using real-time intelligence, Emerald can orchestrate workloads, throttling, shifting, or pausing compute when the grid is stressed. All without compromising performance. In their first pilot, the team cut data center power use by 25% during peak hours at a facility in Phoenix, helping prevent grid overload without sacrificing performance.
Critically, over 100GW of new load could be added to the grid by enabling intelligent flexibility in new data centers, all without additional spend on new transmission or generation. By providing that intelligent flexibility, Emerald can help avoid lengthy interconnection queues and grid upgrades, enabling faster connection to the grid and ensuring the U.S. can maintain its place in the AI arms race without piling on consumers’ electricity bills.
The Team: Builders at the Intersection of Energy, AI, and Policy
Founder Varun Sivaram and his team understand that fixing this problem is a systems challenge as well as a software one.
Varun is a physicist turned energy technologist turned policy expert. He has led strategy and innovation at Ørsted, was CTO of ReNew Power (one of India’s largest power development firms), and served as a senior diplomat for Clean Energy at the State Department. When we met Varun at an electricity conference a few months ago, we knew we had to learn more about what he was building.
His co-founders Ayşe Coskun, Shayan Sengupta, and Aroon Vijaykar, bring world-class experience across energy systems, hyperscale compute, and market design.
According to the Emerald team, “AI data centers can deliver the economic development and grid-friendly support that communities and power utilities compete to attract. AI factories can serve as grid stabilizers and unlock vast quantities of power capacity that already exists by more effectively using today’s grid infrastructure. As a result, the power system becomes more affordable and more reliable, not less.”
New Capital to Solve New Challenges
Today, Emerald announced an additional $18 million seed extension, bringing total seed funding to $42.5 million. The seed extension was led by Lowercarbon Capital, with additional participation from Trust Ventures, NVIDIA, Kleiner Perkins Chairman John Doerr, and others.
The new capital will help develop Emerald’s Conductor™ software to commercial scale
and expand demonstrations and deployments across North America and the United Kingdom.
The team also announced a collaboration with NVIDIA to launch the world’s first
commercial‑scale, power‑flexible, 96MW AI factory. This factory is a massive commercial and technical milestone that will serve as a reference for future AI factories so that one day we can build a network of these power-flexible factories across the globe.
Navigating a Century-old Regulatory Regime
Like many of the industries where we invest, the greatest barriers aren’t technical, they’re regulatory.
Emerald’s approach means tackling outdated utility rules that don’t align with dynamic energy use, or decongesting the grid. Further, scaling nationally means navigating a patchwork of rules across different states and regions (in addition to state and federal regulators, much of our transmission grid is managed by 7 market operators, called regional transmission organizations (RTOs) or independent systems operators (ISOs)). Interconnection standards assume data centers are inflexible, but we’ve seen major breakthroughs in recent months. Last week, DOE directed FERC to initiate rulemaking on proposals including having large, flexible loads be studied on an expedited track–potentially within 60 days–if they agree to be curtailable or dispatchable. RTO Southwest Power Pool (SPP) is rolling out a clear path to interconnection agreements within 90 days for large loads meeting certain conditions and willing to curtail energy use during demand peaks.
But these are exactly the kinds of challenges we must help innovators overcome. Places where the inertia of the status quo keeps American’s energy reliability, electricity bills, and national security in jeopardy. Overcoming that inertia means translating groundbreaking technology into language that policymakers understand, and partnering with the key stakeholders that control our grid.
Turning Constraint Into Opportunity
At Trust Ventures, we believe that when regulation becomes the bottleneck to innovation, it’s a signal. A challenge that demands entrepreneurial creativity. That’s exactly what Varun and his team at Emerald AI are doing by turning the biggest constraint on AI’s growth into its next great opportunity.
If Emerald AI succeeds, they will undoubtedly make AI deployment faster, cheaper, and more reliable, but they will also expand the effective capacity of the U.S. grid without a single new power plant. The solution is obvious. We just need to get out of our own way.
Congratulations to the Emerald team!