As energy becomes the foundation of the AI era, the energy industry is also being reshaped by AI. This article is the 10th issue of the "Energy Singularity" series, focusing on how AI-native energy companies are seizing a new entry point in AI infrastructure amidst the AI computing power explosion.
Since entering 2026, the anxiety in the tech circle has been shifting from models, code, and chips to further down the AI industry chain. The industry has begun to discuss a more fundamental question: with so much AI computing power, is there enough electricity to support it?
At the recent NVIDIA GTC Taipei conference on June 1, Huang Renxun unveiled NVIDIA DSX, the third-generation MGX cabinet architecture, and an 800VDC power supply solution. These redesigns integrate computing, networking, storage, power, cooling, and control systems as a whole, aiming to redefine the "power systems engineering" within AI factories.
What NVIDIA is trying to do is optimize the internal system efficiency of AI factories, allowing computation, networking, power, and cooling to work together to increase the Token output per watt of electricity.
Meanwhile, the external power supply aspect of AI factories is also becoming a new bottleneck: where to locate data centers, where there is available capacity for connection, how quickly projects can be grid-connected, and how to adjust loads based on grid pressure after construction.
Against this backdrop and demand, a group of "AI-native energy companies" has gradually emerged.
A particularly interesting point is that, although they are energy companies, these companies do not build power plants or lay power lines. Instead, they rely solely on code and algorithms, attempting to redefine the flow, price, and pace of electricity.
The capital market has also begun to reprice these types of companies.
In May 2026, Sutter Hill Ventures, a Silicon Valley venture capital firm that was an early investor in NVIDIA, jointly led a $64 million Series A financing round for a company called "GridCARE" with renowned investor John Doerr.
GridCARE uses AI in the electricity access and energy scheduling processes to help AI factories quickly identify available electricity resources, complete connection planning, and participate in subsequent load scheduling.
In the past, the imagination of energy tech companies mostly came from new energy, energy storage, batteries, and grid equipment; but after the explosion in AI computing power demand, whoever can help data centers quickly find, connect, and efficiently use electricity may become a key link in the AI infrastructure chain.
Companies like GridCARE have also begun to emerge in Silicon Valley and other areas of the United States.
One such company is Emerald AI, based in Washington, USA, which has raised a total of approximately $68 million in 16 months. The company is backed by NVIDIA NVentures, Energy Impact Partners, as well as power industry giants such as Eaton, Siemens, and GE Vernova. Individual investors in this round include Jeff Dean and Fei-Fei Li.
In May 2026, another company founded by a quantitative trading entrepreneur, Amann Shariff, called Shatterdome Energy, completed a $3.5 million Pre-Seed funding round.
These companies are mostly targeting the most constrained aspects of AI infrastructure: "finding electricity" in the grid, determining available capacity, facilitating faster connections to shorten the grid integration queue; adjusting computing tasks during grid stress, peak shaving; and engaging in real-time power trading and scheduling for new energy sources, energy storage, and large industrial users using AI.
The rise of these companies holds significant reference and learning value for the industry.
In the AI era of energy competition, it is not just about building more power plants and expanding the power grid but also about organizing new energy sources, energy storage, the power grid, and computing power loads more efficiently. In the future, those who can quickly find, connect, and schedule electricity may gain a more advantageous position in the competition for AI infrastructure.
Aside from chips and processing power, electricity is becoming the new speed limit for AI systems. The power industry itself is also being rewritten by AI.
01 Power Anxiety in the AI Era: It's not a lack of electricity, but "available electricity"
In the AI era, the apparent power anxiety is "power shortage," but fundamentally, it is "lack of available electricity." Many power resources are not non-existent but rather have not been adequately identified, scheduled, and delivered.
In May 2026, Silicon Valley venture capital firm Bessemer Venture Partners released the "Roadmap: AI Data Center Stack" report, which presented a set of figures: by early 2026, 190 gigawatts of hyperscale data center projects had been announced globally, but only 12 gigawatts were actually operational, 21 gigawatts were under construction, and the remaining 148 gigawatts existed only on paper. Of the projects planned to go live in 2025, over a quarter were stuck in the electricity and permitting stages.
A research report released by Stanford University in December 2025 also pointed out that the utilization rate of the U.S. power grid is less than one-third most of the time. The grid intelligence operation company GridCARE provided more specific numbers: even in areas with the most strained power usage, the actual grid utilization is less than 32%. The issue is not a lack of electricity, but a lack of the ability to distribute it.
GridCARE's co-founder and CEO, Amit Narayan, coined a term for this phenomenon, calling it the "Time-to-Energize Crisis," referring to the several-year gap between electricity demand and actual power supply. A large amount of existing grid capacity remains unused due to constraints in traditional scheduling and grid integration processes.
Describing the current situation, he once said, "The current AI frenzy has reached such an uncontrollable level that people think sending chips into space may be faster than finding electricity on Earth."
Behind this bottleneck lies a significant business opportunity. According to GridCARE's calculations, unlocking 1 gigawatt of power capacity for grid connection ahead of schedule can unleash $250 billion in value.
Lead investor Sutter Hill Ventures is one of NVIDIA's early investment firms and can be said to have been involved throughout the rise of the "computing era." The firm's Managing Director, Vic Miller, publicly stated, "A year ago, few were talking about electricity as the bottleneck for AI. Today, it has become an insurmountable challenge for the entire industry."
Co-investor John Doerr, who was also an early investor in Amazon and Google, explained the investment logic in just one sentence: "GridCARE provides affordable, sustainable energy by unlocking the idle power in our existing grids."
GridCARE has launched a "Power Acceleration" software. Its core technology involves using AI to real-time simulate and analyze billions of operating states of the grid, including line congestion, blackout risks, weather changes, demand fluctuations, etc., and then identifying idle power and redirecting it to where it is needed.
Currently, this model has been successfully applied in the first case study. GridCARE is partnering with Portland General Electric to release over 400 megawatts of grid capacity in Hillsboro, Oregon, enough to support six data center connections. The initial 80 megawatts are expected to be operational by 2026.
02 From Power Finding to Power Adjustment: Teaching AI Factories to Implement "Off-Peak Electricity Use"
GridCARE focuses on the grid side, trying to extract more capacity from the existing power distribution system.
There are also energy startups that focus on the software layer, but with a completely different approach.
A company called "Emerald AI" is exploring turning AI data centers into dispatchable grid assets, allowing data centers to adjust their power consumption based on grid conditions. For example, when the grid is under stress, some AI tasks can temporarily slow down, be postponed, or be migrated to operate in other regions; once the grid stress is relieved, they can resume higher load.
The underlying logic here is that AI factories do not need to run at full capacity all the time. Model training tasks can be paused and then resumed, batch inference tasks can be moved to another region. As long as the data center can proactively lower power based on grid commands, the grid stress will be much less, and there is no need to spend money on new lines just for that peak load.
The product launched by Emerald AI is called the "Conductor" platform, which is like installing a "flexible and adaptive" brain for data centers.
Its function is similar to an intelligent valve installed between the grid and the data center. When the grid is tight, the platform receives a signal to instantly reduce the facility's power consumption while ensuring that critical AI tasks running on NVIDIA GPUs are not affected.
At COMPUTEX Taipei, Emerald AI announced a partnership with NVIDIA and Silicon Valley Power to launch the first commercial multi-megawatt project in Silicon Valley.
The starting point of this project is Silicon Valley Power's "Flexible Load-Interconnect Program." The core of this program is actually to solve the problem of long queues for data centers to be connected to the power grid.
Siva Ramamurthy commented on this, stating: "Silicon Valley Power's 'Flexible Load-Interconnect Program' has proven that this path is feasible from a regulatory perspective. NVIDIA's DSX OS, DSX Flex, combined with our Conductor platform, has enabled this technical solution to be implemented at a commercial scale."
03 From Single Point Scheduling to Platformization: Upgrading AI to a "Virtual Power Plant" Version
Compared to GridCARE and Emerald AI, AI energy company Grid AI seems to have a larger ambition.
Grid AI aims to use a unified AI platform to connect all distributed energy resources, ranging from a single household's air conditioner to a large AI data center's backup power, all into one scheduling system.
They have divided this idea into three levels to implement.
The first category is ordinary households and small businesses, where AI automatically manages devices such as air conditioners, electric vehicles, batteries in the background, helping users consume more electricity when the price is low and less electricity when the price is high or the grid is strained;
The second category is commercial and utility scenarios, unifying assets such as energy storage, electric vehicle fleets, and distributed energy resources for participation in electricity market transactions;
The third category is AI data centers and large industrial parks, coordinating power generation, energy storage, and load to provide these high-energy-consuming facilities with more stable and cheaper electricity.
In a sense, Grid AI is essentially like building an "AI-powered virtual power plant." A traditional virtual power plant aggregates many "small power sources, small batteries, small loads" to help the grid alleviate pressure; Grid AI expands the boundary of the virtual power plant to AI data centers and large industrial parks, creating an AI energy scheduling platform that covers household, commercial, utility, and large-scale industrial electricity usage scenarios.
In addition to optimizing the grid and loads, AI is also starting to enter the electricity market trading phase.
U.S.-based AI energy trading service provider Shatterdome Energy positions itself as the "financial infrastructure layer" of the energy world.
A rooftop solar panel, a wind turbine, a set of energy storage batteries, which were once just scattered power generation equipment; but in Shatterdome Energy's system, they can be packaged into a tradable energy asset. The platform decides when to sell electricity, when to store electricity, and when to use trading tools to hedge price risks based on electricity price fluctuations, weather changes, power generation forecasts, and market demand.
Shatterdome Energy's AI tools focus on subtle signals in the electricity market that human traders find difficult to detect in a timely manner. For example, sudden congestion on a certain transmission line, an area where power generation speed cannot keep up with demand, or an upcoming abnormal price fluctuation at a specific node. The algorithm can make decisions on these changes as soon as they occur and execute trades faster than human traders.
As the share of new energy sources increases, the electricity market becomes increasingly unpredictable: weather affects wind and solar power output, data centers suddenly increase load, and local grid congestion causes rapid price differentiation in different regions. For electricity companies, a wrong prediction or slow scheduling could result in fines and trading losses.
With the advent of AI, energy trading is becoming more like a high-frequency game, not only helping businesses "save on electricity bills" but also assisting power companies in more accurately predicting supply and demand, responding to price changes faster, and reducing losses due to judgment errors.
The technology services company Digiqt's survey in September 2025 revealed: AI traders are rapidly penetrating the energy market. They have brought about tangible changes. For example, a medium-sized power company used to incur monthly losses of €50,000 to €150,000 solely due to forecast deviations, but after implementing AI, these losses decreased by 15% to 30%.
04 "Flexible Load": A New Solution to the AI Factory's Electricity Connection Challenge
Many startup stories have been told, but what are the actual results? Can AI data centers really "listen to the grid"?
In March 2026, an experiment provided the answer.
A test was conducted jointly by the UK's National Grid, NVIDIA, Emerald AI, and the Electric Power Research Institute (EPRI): after a signal was issued by the grid, the London data center reduced its power consumption by one-third in approximately one minute. Most importantly, the AI tasks running on NVIDIA GPUs did not experience any interruptions.
Another longer test lasting ten hours maintained the data center's power at around 10% capacity for an extended period, with the workload unaffected.
These two results show that an AI data center is not just a "power-hungry" machine running at full capacity all the time. Instead, it can act like an adjustable load, proactively stepping back during grid stress.
If operators can demonstrate their ability to proactively shed load during grid stress, the grid does not need to expand solely based on theoretical peak demand. This, in turn, reduces the pressure on grid development, and data centers may have shorter integration waiting times.
The significance of this London experiment lies here: although preliminary, it proved that at least on the AI data center side, the capability of "flexible response" can be empirically validated.
05 Conclusion: Software is Redefining the Power Layer
Whether it's GridCARE maneuvering in a congested grid, Emerald AI teaching data centers to shift electricity usage, or Shatterdome Energy using algorithms in power trading, they all point to the same trend: in the AI era, electricity is not only about quantity but also about efficient utilization and management.
These AI-native energy companies have not built new power plants or erected high-voltage lines. However, the software layer they have created is becoming an essential part of the power grid system.
This also echoes Huang Renxun's previously proposed "AI Five-Layer Cake" framework: Energy is located at the bottom layer, with chips, infrastructure, models, and applications layered on top. Without continuous, stable, and schedulable power supply, even the most powerful chips and models cannot truly run.
This may be a profound transformation in the AI era: the power grid, a behemoth born in the industrial age, is being reassembled line by line of code.
Ultimately, whoever possesses smarter algorithms holds the key to driving the AI civilization.
Welcome to join the official BlockBeats community:
Telegram Subscription Group: https://t.me/theblockbeats
Telegram Discussion Group: https://t.me/BlockBeats_App
Official Twitter Account: https://twitter.com/BlockBeatsAsia
