Lots of words have been written lately (many by us!) about our failing grid, AI’s insatiable need for power, and the distressing rise in the cost of electricity delivery relative to the incredible progress that’s been made to make its generation more affordable.
Sit in on a middle America town hall and the prevailing public narrative seems to be, AI companies are going to build gleaming behind-the-meter microgrids while everyday people make do with the aging, existing power grid. If you press the question, “We’re a rich country, why can’t we fix it?” you’ll get answers back like, “There’s a multi-year backlog for transformers, you know.” And we don’t accept that.
America is a wealthy country; we should not have to choose between winning the AI race or burdening citizens with electric bills. We shouldn’t have to choose whether our energy is clean or plentiful. I am not satisfied with these tradeoffs, and neither is Drew Baglino and his team at Heron Power.
Heron Power is on a mission to transform power generation and delivery, and we believe this is the team to do it. Drew and his team revolutionized the automotive industry at Tesla. And in the process they invented the power electronics technology that paved the way for the electric grid of the future. One that’s built on software, faster to build, more resilient to maintain, and makes electricity more affordable for all. Heron can change the course of our energy industry: we are honored to partner with Drew and the team for their Series B, and motivated by the work ahead.
The challenge of the century
For most of its history, the electric grid was designed around a simple assumption: supply and demand were both relatively stable and predictable. Large power plants, whether coal, gas, or nuclear, generated electricity at a steady rate, and utilities adjusted output gradually to match demand as it rose through the day and fell at night (with peaker plants here and there to fill in the gaps). Customers drew power when they needed it. The relationship between generation and consumption was slow-moving and well-understood.
The problem is that the grid’s core hardware, particularly transformers, has barely changed in half a century. The transformers stepping voltage up and down across every link in the grid are the same devices that were deployed in the 1970s: oil-filled steel cores that transform electricity through magnetic induction. They cannot monitor conditions, adapt to changes, or be coordinated through software. They are also very hard to manufacture and even harder to repair, and if you are an energy developer trying to add new capacity to the grid, the wait time for a new transformer is years.
The grid was engineered for a predictable world that no longer exists. That assumption has broken down on both sides of the equation simultaneously, and quickly. On the demand side, the load profile is increasing quickly, and changing in ways the grid was never built to handle. AI data centers draw enormous amounts of power. Electric vehicles charge in patterns that are hard to forecast. Industrial facilities are switching from gas to electric processes. Demand is no longer smooth. It is spiky, distributed, and increasingly difficult to predict.
On the supply side, the need to get power projects online quickly has changed the calculus of how projects get built and introduced a different kind of variability. For instance, solar is now the cheapest and most efficient source of electricity to ever exist on this planet, and can be deployed much more quickly than a massive natural gas plant. But it’s distributed; it fluctuates with weather and time of day, producing surpluses at some moments and shortfalls at others; and it varies by geography. The grid now must absorb intermittent generation from thousands of distributed sources rather than dispatching from a small number of controllable plants.
The result is a system that must coordinate across far more variable nodes than it was ever built to manage. The old model assumed the grid could operate with slow, centralized control. The new reality requires speed to deploy and real-time coordination at scale.
The solid-state solution
The only practical way to manage complexity at this scale is software. This is not a new insight. It is the same principle that drove the shift from hardware routers to software-defined networking in telecommunications. When the coordination problem becomes too fast-moving and too distributed for centralized human operators to manage, the answer is to embed intelligence in the network itself.
The electric grid is going through the same transition. Managing thousands of variable nodes across generation, storage, and load requires software that can monitor conditions, make decisions in milliseconds, and coordinate responses. This is what people mean when they talk about the “smart grid” or distributed energy resources: not just different equipment, but a fundamentally different operating model, driven by software.
The obstacle to a software-defined grid has been that the core hardware, namely transformers and the many pieces of related equipment, cannot participate in software-driven coordination. Conventional transformers are passive magnetic devices. They cannot be monitored in real time, cannot respond to changing conditions, and cannot be updated or managed remotely. To build a grid that operates through software, the hardware itself has to change.
This is where solid-state transformers come in. Unlike conventional transformers, which move electricity through magnetic cores, solid-state transformers use power electronics — semiconductor switches controlled by software — to regulate voltage and power flow in real time. This makes them programmable, responsive, and capable of being coordinated as nodes in a managed network.
The practical benefits are immediate: projects can come online faster because solid-state transformers are compact and modular, the grid can operate more reliably because equipment can respond to faults and disturbances automatically, and the system can scale more affordably because software coordination reduces the need for costly hardware redundancy.
Building a generational company
While building towards Tesla’s Master Plan, Drew realized he had the technology solution to the grid’s critical bottleneck. The need for software-defined grid infrastructure has been building for years, and the window between “technically feasible” and “widely deployed” is where generational companies are built. He and his team set out to take their power electronics expertise and deploy it against grid-scale challenges.
Just a year later they are preparing to break ground on a 40-gigawatt factory here in the United States to manufacture their first product, HeronLink, at scale, and we are proud to support them on this journey. There aren’t many American founders who have built large-scale factories in the United States this century, not just once but several times over. But Drew Baglino has.
Heron Power shares our belief at Andreessen Horowitz that America is not a zero-sum nation. We deserve access to cheap and abundant electricity and we should win the AI race. We must have the capacity to manufacture critical technology here at home. When faced with existential challenges that threaten our society and way of life, we deploy technology against them. And we win.
