The surge in AI computing power is creating new challenges for power infrastructure, as AI-native energy companies such as GridCARE, Emerald AI, and Shatterdome Energy are emerging. These companies do not build power plants; instead, they use AI algorithms to redefine electricity flow, pricing, and dispatch. GridCARE has secured $64 million in Series A funding to help AI facilities identify available power resources; Emerald AI has launched its Conductor platform, enabling data centers to adjust their power consumption based on grid conditions; and Shatterdome Energy uses AI to participate in electricity market trading. Major chipmakers like NVIDIA and early investors are also making significant investments. Industry analysts note that competition in the AI era is shifting from “building more power plants” to “organizing power resources more efficiently,” with electricity becoming the new speed bottleneck for AI systems after chips.Article author and source: Tencent Technology
Under the surge of AI computing power, how can AI-native energy companies seize the new entry point into AI infrastructure.
Since the beginning of 2026, anxiety in the tech industry has continued to spread deeper into the AI supply chain, moving beyond models, code, and chips. The industry is now grappling with a more fundamental question: With all this AI computing power, is there enough electricity to support it?
At the just-concluded NVIDIA GTC Taipei conference on June 1, Jensen Huang introduced the NVIDIA DSX, the third-generation MGX rack architecture, and an 800VDC power supply solution, redesigning compute, networking, storage, power, cooling, and control systems as an integrated whole to redefine "power systems engineering" within AI factories.
NVIDIA aims to optimize the internal system efficiency of AI factories, aligning computation, networking, power, and cooling to maximize token output per watt.
Meanwhile, the power-related aspects outside the AI factory are also becoming new bottlenecks: where data centers are built, what available capacity exists, whether projects can be connected to the grid as soon as possible, and how load adjustments can be made based on grid pressure after construction.
Under these circumstances and demands, a growing number of "AI-native energy companies" have emerged.
An interesting point is that, despite being energy companies, these firms do not build power plants or lay power lines—they rely solely on code and algorithms to redefine the flow, pricing, and timing of electricity.
Capital markets are also revaluing these companies.
In May 2026, Sutter Hill Ventures, a Silicon Valley venture capital firm that early invested in NVIDIA, co-led a $64 million Series A round in GridCARE alongside renowned investor John Doerr.
GridCARE applies AI to power connection and energy dispatch processes, helping AI factories quickly identify available power resources, complete connection planning, and participate in subsequent load dispatch.
In the past, the growth potential of energy technology companies was primarily driven by new energy, energy storage, batteries, and grid equipment; however, following the surge in AI computing demand, those who can help data centers quickly locate, connect to, and efficiently utilize power may become critical links in the AI infrastructure chain.
Companies similar to GridCARE are also beginning to emerge in areas such as Silicon Valley in the United States.
Emerald AI, headquartered in Washington, USA, has raised approximately $68 million in funding over 16 months, backed by NVIDIA’s NVentures, Energy Impact Partners, and major power industry players including Eaton, Siemens, and GE Vernova. Jeff Dean and Fei-Fei Li have also invested personally.
In May 2026, Shatterdome Energy, founded by quant trading background entrepreneur Amann Shariff, also completed a $3.5 million Pre-Seed funding round.
These companies are primarily targeting the most constrained aspects of current AI infrastructure: locating power within the grid—identifying where available capacity exists and where connection can be expedited to reduce grid connection wait times; adjusting computing workloads and shifting electricity usage to off-peak hours during periods of grid strain; and enabling real-time electricity trading and dispatch using AI for renewable energy sources, energy storage systems, and large industrial users.
The rise of these companies holds significant reference and inspiration value for the industry.
In the AI era, the competition for energy is not just about building more power plants or laying more transmission lines—it also involves more efficiently integrating renewable energy, energy storage, power grids, and computing loads. In the future, those who can quickly locate, connect to, and manage electricity will hold a more advantageous position in the race for AI infrastructure.
Beyond chips and computing power, electricity is becoming the new bottleneck for AI systems. Meanwhile, the electricity industry itself is being reshaped by AI.
Image generated by AI
01 Power Anxiety in the AI Era: It’s Not a Lack of Electricity, But a Lack of “Usable Electricity”
In the AI era, power anxiety appears on the surface to be a shortage of electricity, but at its core, it’s a shortage of usable electricity. Much of the power resources aren’t absent—they simply haven’t been fully identified, coordinated, or delivered.
In its May 2026 report, "Roadmap: The AI Data Center Stack," Silicon Valley venture capital firm Bessemer Venture Partners presented the following figures: As of early 2026, 190 gigawatts of hyperscale data center projects had been announced globally, but only 12 gigawatts were operational, 21 gigawatts were under construction, and the remaining 148 gigawatts remained on paper. Of the projects planned for launch in 2025, more than a quarter were stalled due to power and permitting issues.
A Stanford University research report released in December 2025 also found that the U.S. power grid operates at less than one-third of its capacity for most of the time. GridCARE, a smart grid operations company, provided even more specific figures: even in the regions with the highest electricity demand, the grid’s actual utilization rate remains below 32%. There is no shortage of electricity—only a shortage of capacity to deliver it.
Amit Narayan, co-founder and CEO of GridCARE, has named this phenomenon the "Time-to-Energize Crisis," referring to the multi-year gap between electricity demand and actual power supply. A significant portion of existing grid capacity remains unused due to limitations in traditional scheduling and interconnection processes.
When describing the current situation, he said: "The current AI frenzy has gotten so out of hand that people believe sending chips into space might be faster than finding electricity on Earth."
Behind this bottleneck lies a tremendous business opportunity. According to GridCARE’s calculations, helping just 1 gigawatt of power connect to the grid ahead of schedule can unlock $25 billion in value.
Lead investor Sutter Hill Ventures was one of NVIDIA’s early investors and has essentially been involved throughout the rise of the "computing era." The firm’s managing director, Vic Miller, publicly stated: "A year ago, few people talked about power as a bottleneck for AI. Today, it has become an unavoidable challenge for the entire industry."
John Doerr, an early investor in Amazon and Google, also participated in the follow-on investment. When explaining his rationale, he simply said: "GridCARE provides affordable, sustainable energy by unlocking idle power from our existing grid."
GridCARE has launched software called "Power Acceleration." Its core technology uses AI to simulate and analyze billions of real-time grid operating conditions—including line congestion, outage risks, weather changes, and demand fluctuations—and identifies underutilized electricity, directing it to where it is needed most.
This model has already successfully delivered its first case. GridCARE is collaborating with Portland General Electric to unlock over 400 megawatts of grid capacity in Hillsboro, Oregon—enough to support six data centers. The first 80 megawatts are expected to be operational by 2026.
02 From Finding Power to Adjusting Power: Teaching AI Factories to Shift Electricity Usage Off-Peak
GridCARE focuses on the grid side, aiming to unlock additional available capacity from existing transmission and distribution systems.
Energy startups also focus on the software layer, but with a completely different entry point.
A company called Emerald AI is exploring the transformation of AI data centers into dispatchable grid assets, enabling data centers to adjust their power consumption based on grid conditions. For example, when grid stress is high, certain AI tasks can be temporarily slowed down, delayed, or migrated to other regions; once grid pressure eases, operations can resume at higher loads.
The underlying logic here is that an AI factory doesn't need to operate at full capacity all the time. Model training tasks can be paused and resumed later, and batch inference tasks can be moved to other regions. As long as the data center can proactively reduce power consumption based on grid instructions, the strain on the grid will be significantly reduced, eliminating the need to invest heavily in new infrastructure to handle peak loads.
Emerald AI's product, the Conductor platform, acts like a flexible, adaptive brain for data centers.
Its function is similar to a smart valve installed between the power grid and the data center. When the grid is under strain, the platform receives a signal and instantly reduces power consumption while ensuring that critical AI tasks running on NVIDIA GPUs remain unaffected.
At COMPUTEX Taipei, Emerald AI, NVIDIA, and Silicon Valley Power announced a partnership to launch the first commercial multi-megawatt project in Silicon Valley.
The project originated from Silicon Valley Power’s “Flexible Load Interconnection Program,” which aims to address the issue of lengthy wait times for data centers to connect to the grid.
Sivaram commented: "Silicon Valley Power’s Flexible Load Interconnection Program has demonstrated that this path is viable at the regulatory level. NVIDIA’s DSX OS and DSX Flex, combined with our Conductor platform, have brought this technical solution to commercial scale."
03 From Single-Point Scheduling to Platformization: An AI-Enhanced "Virtual Power Plant"
Compared to GridCARE and Emerald AI, AI energy company Grid AI seems to have a larger appetite.
Grid AI aims to connect all dispersed power resources through a unified AI platform, integrating everything from a single household's air conditioner to the backup power of an AI data center into centralized management.
They broke this idea down into three levels for implementation.
The first category consists of ordinary households and small businesses, where AI automatically manages devices such as air conditioners, electric vehicles, and batteries behind the scenes, helping users consume more electricity when prices are low and reduce usage when prices are high or the grid is under strain.
The second category involves commercial and utility scenarios, where assets such as energy storage, electric vehicle fleets, and distributed generation are centrally coordinated to participate in electricity market transactions.
The third category consists of AI data centers and large industrial parks, which optimize power generation, energy storage, and load demand to provide these high-energy-consuming facilities with more stable and cost-effective electricity.
To some extent, Grid AI is essentially building an "AI version of a virtual power plant." Traditional virtual power plants aggregate numerous small power sources, small batteries, and small loads to alleviate pressure on the grid; Grid AI expands this boundary to include AI data centers and large industrial parks, creating an AI-powered energy scheduling platform that covers residential, commercial, utility, and hyperscale energy consumption scenarios.
In addition to optimizing grids and loads, AI is now entering the trading segment of electricity markets.
The 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, and a set of energy storage batteries were once isolated power generation devices; but within Shatterdome Energy’s system, they can be bundled into a tradable energy asset. The platform determines when to sell electricity, when to store it, and when to use trading instruments to hedge against price risks, based on fluctuations in electricity prices, weather conditions, generation forecasts, and market demand.
Shatterdome Energy’s AI tool focuses on subtle market signals that human traders struggle to detect in real time—such as sudden congestion on a transmission line, a region’s generation falling behind demand, or an impending abnormal price fluctuation at a specific node. The algorithm can identify these changes as they occur and execute trades faster than human traders.
As the share of new energy sources increases, the electricity market is becoming increasingly difficult to predict: weather affects wind and solar generation, data centers can suddenly spike demand, and local grid congestion can cause rapid price divergences across regions. For power companies, inaccurate forecasts or slow dispatching can lead directly to fines and trading losses.
After the introduction of AI, energy trading has become more like a high-frequency game—requiring not only help for businesses to "reduce electricity bills," but also assistance for power companies to more accurately forecast supply and demand, respond faster to price changes, and minimize losses due to misjudgments.
A survey by technology services company Digiqt in September 2025 found that AI traders are rapidly penetrating the energy market. They are bringing tangible changes: a mid-sized power company previously lost between €50,000 and €150,000 per month in imbalance penalties due solely to forecasting errors; after integrating AI, these losses decreased by 15% to 30%.
04 "Flexible Load": A New Solution to the Power Connection Challenges of AI Factories
Startups have told many stories, but what are the real results? Can AI data centers actually "listen to the power grid"?
In March 2026, a trial provided the answer.
The UK’s National Grid, NVIDIA, Emerald AI, and the Electric Power Research Institute (EPRI) conducted a joint test: after receiving a signal from the grid, the data center in London reduced its power consumption by one-third within about a minute. Crucially, the AI tasks running on NVIDIA GPUs were not interrupted.
Another test lasted longer, continuing for ten hours. The data center maintained power at approximately 10% for an extended period, with no impact on workload.
These results show that AI data centers are not just rigid, always-on power hogs—they can also act like adjustable loads, voluntarily reducing demand during periods of grid stress.
If the operator can demonstrate the ability to actively reduce load during periods of grid stress, the grid will not need to be expanded entirely based on theoretical maximum capacity. This can alleviate pressure on grid infrastructure and potentially shorten the waiting time for data centers to connect to the grid.
The significance of this trial in London lies here: although it is a preliminary experiment, it demonstrates that "flexible responsiveness" is a verifiable capability—at least on the side of AI data centers.
05 Conclusion: Software is redefining the power layer
Whether it’s GridCARE intelligently optimizing power flow in congested grids, Emerald AI teaching data centers to shift electricity usage off-peak, or Shatterdome Energy using algorithms to participate in power markets, all point to the same trend: in the age of AI, electricity isn’t just about producing more—it’s about using and managing it smarter.
These AI-native energy companies have not built a single power plant or installed a single high-voltage line. Yet the software layers they have created are becoming a vital component of the grid system.
This also echoes Huang Renxun’s earlier “AI Five-Layer Cake” framework: energy forms the bottom layer, followed by chips, infrastructure, models, and applications. Without consistent, stable, and dispatchable power, even the most powerful chips and models cannot function effectively.
This may be one of the profound transformations of the AI era: the massive power grid, born in the industrial age, is being reassembled line by line of code.
In the end, whoever has the smarter algorithm holds the key to driving AI civilization.