EV fast charging breaks less at the charger than at the grid. In this interview, ElectricFish founder Anurag Kamal explains how battery-buffered systems route around interconnection queues and reshape deployment economics.
1. At CES 2026, ElectricFish is launching Turbo Charge and the 400squared platform into a moment when transformer lead times stretch multiple years and interconnection queues are effectively stalled. Was there a specific early design decision where you realized this problem couldn’t be solved by better charging alone, but only by bypassing grid bottlenecks altogether?
Anurag Kamal: The realization came early during my time at Michigan Tech, while working on lithium-ion battery health at both grid scale and automotive scale. On one end, grid-scale storage systems were massive, centralized assets—slow to deploy, complex to permit, and increasingly difficult to manage safely as energy density increased. While, automotive batteries were modular, tightly controlled, and designed for fast power delivery—but locked inside individual vehicles.
What stood out was the “missing middle”. There were no widely deployed community-scale energy storage systems—large enough to matter for the grid, yet small enough to be deployed quickly, replicated safely, and managed with the rigor of automotive systems. That gap stayed with me.
The broader insight followed naturally: the electrical grid is moving back toward decentralization—closer to how it originally started—and electric mobility is undergoing the same shift. When we ran the math, it became obvious that DC fast charging was the right beachhead for this transition. A conventional DC fast charger typically triggers $150,000 or more per port in utility upgrades and takes 12–18 months to deploy. At the same time, interconnection queues were already averaging five years, with some regions seeing seven-year waits driven largely by data-center demand.
The decision to build around a ~400 kWh battery reservoir wasn’t just a technical choice—it was an acknowledgment that grid constraints weren’t temporary. We weren’t going to out-wait the problem. We had to route around it. That decision became the foundation of ElectricFish’s architecture: local energy reservoirs, orchestrated by software, bypassing the grid bottleneck entirely.
2. Battery-buffered fast charging has been tried before. Companies like FreeWire validated the technical concept but struggled commercially. When you studied those outcomes, what was the most dangerous assumption you deliberately chose not to inherit at ElectricFish?
Anurag Kamal: The most dangerous assumption we rejected was that this was primarily a hardware optimization problem. Companies like FreeWire tried to solve fast charging by packing as much innovation as possible into a constrained pedestal—custom power electronics, aggressive battery chemistries, and bespoke hardware—before proving field reliability or unit economics at scale.
We made the opposite choice: start with proven, safe battery systems, give ourselves physical and thermal margin, and design the system around grid services and software orchestration—not just charging—because DC fast charging is the beachhead, not the business. We’re not trying to deploy chargers faster than the grid; we’re building infrastructure that routes around the grid constraint entirely.
3. From a utility’s perspective, a 400squared site appears as a modest low-power commercial load rather than a high-power charging hub. How intentional was this architectural invisibility, and how much of your deployment speed advantage comes from regulatory perception versus pure hardware capability?
Anurag Kamal: Completely intentional. Our 400squared unit draws just 30 kW from the grid—roughly the load of a small commercial kitchen or a few residential heat pumps. From the utility’s perspective, we’re adding a modest, predictable load that fits within existing infrastructure capacity. There’s no capacity study required, no transformer upgrade, no demand charge spike.
The deployment speed advantage is roughly 50% regulatory and 50% hardware. The hardware is containerized and requires no trenching—an electrician can have it wired and running in hours. But the real time savings come from avoiding the 12–18 month utility approval and construction process entirely. We’re not invisible to the grid; we’re just polite neighbors rather than demanding tenants.
4. You’ve described ElectricFish as enhancing the grid rather than depending on it. Internally, does your planning assume meaningful improvement in U.S. distribution infrastructure over the next decade, or is the business built on the premise that these failures persist?
Anurag Kamal: Our business model is robust to both scenarios. If grid infrastructure improves dramatically, we still offer faster deployment, lower upfront costs, and grid services revenue that traditional chargers can’t match. If grid constraints persist—which we believe is the more likely scenario given that over 70% of U.S. transmission infrastructure is past the midpoint of its 50-year life expectancy—then our value proposition only strengthens.
We’re not betting against the grid; we’re betting that distributed energy storage is valuable regardless of what happens at the transmission level. Even a perfectly upgraded grid benefits from local energy reservoirs that smooth demand peaks and provide resilience during outages.
5. Stargazer coordinates multiple AI agents with inherently conflicting objectives. When the Grid Agent wants to discharge for price arbitrage but the EV Agent predicts an imminent fleet arrival, which objective wins, and what principle governs that arbitration?
Anurag Kamal: The governing principle is simple: EV charging commitments always take priority. We’re an EV charging company first—a driver who pulls up to our unit must be able to charge. Grid services are valuable, but they’re interruptible; customer experience is not.
In practice, the Coordinator Agent maintains a reserve threshold—a minimum state of charge that’s protected for EV demand. The EV Agent’s predictions inform that threshold dynamically. If fleet arrival patterns suggest high utilization in the next hour, the reserve increases and the Grid Agent’s arbitrage window shrinks. The Cost Agent weighs the financial trade-off, but it never overrides the customer-first principle.
6. Demand charges can account for the majority of a fast-charging site’s electricity bill. How does ElectricFish treat the battery not just as an energy asset, but as a financial hedge against grid-imposed volatility?
Anurag Kamal: This is one of our most important but least visible advantages. Traditional DC fast chargers can see demand charges account for 50–70% of their electricity costs because they create massive, unpredictable spikes in grid draw. Our 30 kW connection is essentially flat—we’re drawing the same modest load whether we’re charging one car or twenty.
The battery isn’t just storing energy; it’s storing optionality. We buy power during off-peak hours when it’s cheapest, we avoid demand charge spikes entirely, and we can sell back during peak periods when prices are highest. The Stargazer Cost Agent is constantly optimizing across time-of-use rates, demand charge tiers, and real-time wholesale prices. For site hosts, this translates to predictable economics rather than bill shock.
7. Your background is rooted in lean software development rather than heavy infrastructure. What is one lean principle that proved hardest to translate into hardware, but ultimately became central to avoiding the capital traps that affected earlier players?
Anurag Kamal: The lean principle that proved hardest—but most critical—to translate was MVP thinking. In software, you can ship early and iterate; in hardware, premature optimization is a capital trap. We defined our “minimum viable system” not as the smallest charger we could build, but the smallest grid-aware energy node that could be safely deployed, generate revenue immediately, and improve with software—because hardware without grid intelligence isn’t a solution, it’s just steel.
That choice forced us to treat software and hardware as a single system, not independent layers, and helped us avoid the mistake of building expensive infrastructure ahead of proven demand or grid value.
8. ElectricFish brings together power electronics engineers, AI researchers, and infrastructure operators. How do you resolve tension between software’s bias toward rapid iteration and electrical engineering cultures where failure has physical consequences?
Anurag Kamal: We’re still learning, and we’re very deliberate about managing that tension—because failure in hardware has real, physical consequences. The way we resolve it is by separating where we move fast from where we lock things down. Software, controls, and optimization live in a sandbox where rapid iteration is encouraged; power electronics, safety systems, and fault boundaries are treated as immutable once validated.
Critically, software isn’t just moving fast on top of hardware—it’s what allows the hardware to behave well. Control software, diagnostics, and AI-driven optimization reduce stress on components, manage thermal and electrical limits, and extend asset life. We rely heavily on simulation and staged validation before anything touches hardware, and we make it explicit internally: speed is never rewarded if it compromises safety. If we fail, it would genuinely suck—so we design both the culture and the system to ensure learning happens early, virtually, and collaboratively, not in the field.
9. The philosophy of behaving like a pump rather than a parking space implies extreme utilization and thermal stress. What did high-temperature desert testing reveal about where theoretical performance assumptions break under real operating conditions?
Anurag Kamal: The Hyundai Proving Ground testing and MotorTrend’s SUV of the Year evaluations were designed specifically to stress-test our assumptions. Hyundai had our unit at the Mojave Desert for almost 5 months, continuously doing their own stress-tests, and found our system to be completely reliable. Then we took the same unit to Burning Man, and back to Mojave for the SUV of the Year review. In the MotorTrend event alone we ran 37 charging sessions in just over a week, delivering 1,119 kWh with peak power of 313 kW through continuous triple-digit temperatures. At the same time, we have a unit running in Detroit, operating equally well at below freezing temperatures.
The most important validation was thermal management. Our liquid-cooled LFP battery chemistry performed exactly as designed—no thermal throttling, no degradation in charge rates. Hyundai’s engineers specifically noted that ‘the built-in safety redundancies gave us confidence this is viable for customer-facing sites.’
10. Your revenue-share model shifts upfront risk away from gas station owners and onto ElectricFish. If utilization ramps more slowly than expected, where does this model fail first: hardware economics, battery degradation, or capital availability?
Anurag Kamal: We’ve stress-tested the model against slow-ramp scenarios. The answer is that grid services revenue provides a floor that traditional charging-only models don’t have. Even at low EV utilization, the battery is generating value through demand charge avoidance, peak shaving, and demand response programs. We’ve run sensitivity tests and break-even happens well before average utilization rates for each prioritized geography – and we will expand into others only after eliminating those uncertainties. As a last resort, the 400squared is not fixed on the ground: it can be lifted and moved to another location.
11. As virtual power plant participation matures, do you expect revenue from grid services to eventually rival or exceed revenue from EV charging, and if that happens, what business does ElectricFish fundamentally become?
Anurag Kamal: Yes, we expect grid services revenue to grow substantially as VPP markets mature and FERC Order 2222 implementation expands. Whether it rivals or exceeds charging revenue will depend on market-by-market grid conditions and regulatory frameworks.
If that happens, ElectricFish becomes what we’ve always intended to be: an energy platform company, not a hardware company. EV charging is our beachhead—immediate revenue and market validation. But every unit deployed is a node in a distributed energy network. At scale, thousands of ElectricFish units become a virtual power plant that can provide services no single asset could offer.
The thesis we’ve described internally is that we could be the architects of the new, modern electrical grid. That’s not hyperbole—it’s the logical endpoint of distributed, intelligent energy storage deployed at community scale.
12. Looking toward 2030, when thousands of ElectricFish units may be deployed nationwide, how do you define the company’s end state: a hardware manufacturer, an energy-as-a-service platform, or an intelligence layer coordinating physical energy assets at the grid’s edge?
Anurag Kamal: The third option—an intelligence layer coordinating physical energy assets at the grid’s edge—is closest to our vision, but I’d frame it slightly differently.
By 2030, we see ElectricFish as the operating system for distributed energy at the community scale. We’ll have hardware in the field, but our competitive moat will be the network effects of thousands of intelligent nodes learning from each other, coordinating across regions, and providing grid services that no centralized asset could match.
The analogy I use is this: In the old world, the cost of electricity was the cost of fuel needed to fire the power plants. In the new world, the cost of electricity is the cost of storage. Whoever can store renewable energy and deploy it when needed wins. We’re building the infrastructure for that future—400 kWh batteries in every block, creating a distributed, resilient network that makes the grid stronger, not weaker.
Editor’s Note
This interview surfaces a structural constraint at the grid layer rather than at charging hardware.
