.svg){kind=icon}
If you are building a large data center, a semiconductor factory, or a new manufacturing facility today, you may already have the chips, the capital, and customers waiting, yet the entire project still depends on one basic question: when can you actually turn it on?
Over the past few weeks, I’ve been spending a lot more time thinking and talking about what the accelerating AI adoption means for the energy industry, and one phrase has come up again and again in my conversations: time-to-power. It is becoming a critical constraint on the intelligence economy, and an important opportunity for BGV.
Energy itself is not a new area for BGV. Over the years, we’ve built an ecosystem of relationships across utilities, energy companies, infrastructure players, and technology partners, and we’ve had meaningful outcomes in the sector, including Schneider Electric’s recent acquisition of AiDash, which uses AI and satellite data to help utilities monitor and manage critical infrastructure. That end-to-end investment-exit cycle deepened our conviction that software and AI can create real value around physical infrastructure.
What feels different today is the pace of change. The energy system is being pulled in several directions at once, and many of the assumptions that held even a few years ago are being tested by a new wave of demand. So for us, the question is less about whether energy belongs in our investment strategy and more about what comes next: where should we expand the BGV ecosystem, and which new categories of companies are best positioned for the next phase of the market?
To work through that question, I’ve been speaking with utilities, energy operators, infrastructure investors, AI experts, founders, and people who have spent decades building and maintaining physical infrastructure. Those conversations have helped sharpen the view of where the biggest bottlenecks are emerging and, more importantly, where technology can realistically help.
We are entering a period where access to power is becoming a real constraint on growth. AI gets most of the attention today, and for good reason, but since large data centers consume extraordinary amounts of electricity, AI is only one component of a much bigger shift. Manufacturing is coming back to the U.S., transportation and industrial processes are becoming more electric, energy security has moved higher up on the strategic agenda. What’s more, many parts of the world still need significantly more power simply to support economic development.
The new bottleneck, it turns out, is time.
Getting power to a site involves much more than producing electricity. There are transformers, transmission lines, grid studies, interconnection approvals, permits, engineering work, equipment orders, construction, and the people with the skills to make all of those pieces come together.
What I find especially interesting is that these steps are increasingly interdependent, which is to say that a delay in completion of one part of the process can slow down the entire project. For a company investing billions into a new facility, getting power six or twelve months earlier can easily matter more than saving a few dollars per megawatt-hour, which is why speed at which the power projects can be completed is becoming economically valuable.
A simple, illustrative example shows why. Take an AI data center with 100 MW of IT capacity and roughly $1 billion invested in land, building, and power infrastructure before the first GPU is switched on, and assume an 8% cost of capital and a lease rate of about $150 per kW per month, which is in line with current hyperscale pricing. Every month that facility sits waiting for power costs roughly $7 million in carrying costs and leaves about $15 million of monthly revenue on the table.
Bring that facility online six months earlier and roughly $130 million is at stake: about $40 million in carrying costs on idle capital plus $90 million of revenue pulled forward, before counting the GPUs that would otherwise sit idle. By comparison, shaving $5 per MWh off the power price saves that same facility about $4 million a year. Months on the critical path are worth far more than cents on the kilowatt-hour.
Grasping this was like a light switch going on inside myself. The critical nature of time-to-power (TtP) has since become much more tangible for me.
What software cannot fix
Before going further, I want to be clear about where the hard limits are, because the people who have spent decades in this industry were quick to point them out to me.
The deepest constraints on time-to-power are physical and political. A gas turbine or a large transformer has a multi-year lead time, and no amount of capital or software shortens it. Rare earths, specialty steel, and skilled trades are all in short supply. On the political side, ratepayers in Texas and elsewhere are starting to push back against footing the bill for data center buildout, and that backlash will shape what gets permitted and where.
None of this is going to be solved from a laptop. Software will not close the gap between how fast compute demand is growing and how fast the physical grid can grow. What it can do is make sure the equipment, capacity, and capital we already have are not sitting idle while the physical buildout catches up.
That is a narrower claim than “AI fixes the grid,” and it is the claim this article makes. For a seed-stage investor it is also the more interesting one, because the value of a single month saved on a single site is large enough to build a company on, even if the national picture stays constrained for years.
These needs, driven by the demand that AI has set in motion, are creating several key areas of opportunity for value creation.
1. Find more power before building more power
The obvious response to rising energy demand is to build more power generation and transmission capacity, and more supporting infrastructure - capex we all clearly need.
At the same time, this physical infrastructure takes years to build. So the more refined, practical question becomes:
What can help us get more out of what we’ve already got?
The answer is that there is a meaningful opportunity to make the existing grid more flexible, visible, productive, and durable. That could include everything from helping projects move through interconnection faster, identifying where power capacity actually exists (and when), coordinating batteries and distributed generation more intelligently, and allowing flexible loads to respond autonomously to what is happening on the grid in real time.
One way of thinking about some of this is re-visualizing capacity as “virtual megawatts.” The idea here is that you do not always need another physical power plant to create additional usable capacity. New technologies including software, optimized data, automation, and more intelligent power grid coordination can unlock capacity that is already available, but poorly utilized.
That concept also came through in conversations with other infrastructure investors, including corporates, where grid-enabling technologies, interconnection, siting, permitting, operations software, and tools that help utilities understand and move capacity more intelligently are becoming increasingly important areas of focus.
2. Build energy infrastructure with much better software
One of the things that surprised me as I started going deeper into the energy space was how much highly sophisticated infrastructure is still developed through a mix of PDFs, spreadsheets, email, consultants, and pre-electricity era manual workflows.
If you think about everything required to bring a large energy project online, the list is miles long: endless feasibility studies, environmental work, engineering, grid studies, procurement, contracting, permitting, construction, commissioning, and dozens upon dozens of compliance checks, often involving multiple organizations vast seas of technical information, and rewind-repeat cycles of doing it over until it's done right. It’s exhausting to even think about it.
Seen through the lens of what AI can now do, however, this increasingly looks exactly like the kind of environment where AI can create real value.
To be clear, the opportunity is far larger than a tool that summarizes a document or instantly answers a question. As investors, we are much more interested in emerging next generation AI software that can help complete the work itself, whether that means preparing filings, reviewing technical documents, identifying conflicts, tracking dependencies, surfacing delays before they become expensive, or automating components of engineering and project management.
This fits very naturally with BGV’s broader view of Enterprise 5.0, where AI moves from being an assistant toward becoming part of the execution layer.
The energy segment could become one of the most economically valuable industries for this shift because the value of accelerated time and process optimization is so concrete. When a project costs hundreds of millions or even billions of dollars, reducing months of delay through automation, performance enhancement, or coordinated, accelerated or synchronous deployment can create a huge economic impact in both the near and long terms. The real measure of these tools is not how many documents they process, but how many months they take off the critical path.
3. Data centers are becoming energy systems
As technology, energy, and financial systems become deeply intertwined, modern AI data centers now sit at the intersection of compute, power, cooling, land, grid infrastructure, and capital - playing an increasingly pivotal role in driving cross-industry growth.
This creates room for, or, more bluntly demands, a new software layer that is only starting to form.
Where should a facility be built based on real power availability? How much should come from the grid versus on-site generation or storage? When should batteries charge and discharge? Can certain computing workloads move to another hour or location when electricity is constrained? How should power contracts be structured, and how should cooling and water use be optimized alongside everything else?
Today, the answers to many of these questions still live in separate systems and separate teams, often disconnected from each other.
Over time, I think there may even be an opportunity for something that looks increasingly like an operating system for AI infrastructure, connecting power procurement, on-site generation, storage, compute workloads, cooling, and eventually the grid itself.
What makes this opportunity even more exciting is that compute is a very different kind of electricity consumer from many traditional industrial load. Some workloads can potentially move across time or geography, which means a data center does not always have to behave like a completely rigid source of demand. With enough intelligence and coordination, compute itself could become part of how the energy system manages flexibility.
4. The Physical Incarnation of AI
Here is where buildout within the energy space runs into a constraint that traditional software solutions cannot fully address: people.
We need electricians, engineers, technicians, inspectors, construction workers, and many other skilled professionals to build, operate, and maintain this infrastructure. And many of these jobs take place across enormous geographical footprints in environments where safety is of vital importance.
This is where our excitement about Physical AI becomes especially relevant.
As investors, we are interested in spatial AI, embodied intelligence, closed-loop agentic maintenance, and human-robot shared autonomy that can help field workers do more with less dependence on physical human capabilities and occupational hazards.
What stood out to me in one recent conversation with a major utility was how practical the demand really is. The interest in robotics was about using them to inspect infrastructure, work at heights, reduce electrocution risk, and support maintenance in dangerous environments.
That is exactly the kind of use case I find compelling because the value is easy to understand and measure. Fewer truck rolls, fewer failures, faster repairs, safer workers, and more productive assets all translate directly into economic value.
5. Making resilience economically viable
As we build more infrastructure, we also create more assets exposed to extreme weather, equipment failures, cyber risks, and other forms of disruption.
Resilience is increasingly becoming an economic quantity in its own right.
If new enterprise 5.0 software can alert an operator that a particular asset is at risk, automatically change how equipment is operating before a storm, reduce downtime, or make an asset easier to insure, then resilience starts to affect very real financial outcomes.
This opportunity becomes exponentially large when risk intelligence connects directly to action, because the value can ultimately show up in lower insurance premiums, better financing terms, higher uptime, lower maintenance costs, and potentially even a lower cost of capital. And all this can offer ongoing competitive advantage well after the initial grace period is over and industry standards themselves have changed accordingly.
Where BGV fits
It is clear there’s going to be massive investment in the near term across new power generation, natural gas, renewables, nuclear, geothermal, and other technologies, and we are watching all of it closely.
At BGV, though, our natural position is the intelligence layer that sits just above the physical asset.
Rather than trying to finance the power plant itself, build a new turbine factory, or pick the winning reactor design, we are much more interested in the technologies that help those assets get built faster, operate better, and become more productive. One nuance worth naming: because critical equipment is so constrained, we are open to businesses that extend the life of what already exists, whether that is refurbishment, repurposing, secondary markets, or supply chain intelligence for long-lead-time components. That is not the same as manufacturing, and the economics can be very good.
In practice, that means backing companies that deliver one of four outcomes: bringing power online sooner, unlocking usable capacity, improving reliable operation, or reducing the amount of power needed in the first place. The first is where compressing TtP starts. The next two are where the value keeps compounding, through better utilization, maintenance, and resilience, long after the power is switched on.
The fourth is the demand side, and I think it may be the most underrated of the group. Smaller models, more efficient inference, non-GPU architectures, and intelligence that runs on the device rather than in a hyperscale facility all reduce the megawatts a unit of useful AI requires. Those are software and chip problems, not turbine problems, which means they can move far faster than anything on the physical side. They also sit squarely in BGV’s core enterprise AI practice, so this is not a detour from what we already do.
The wedge can come from almost anywhere across energy subverticals. For nuclear, this could mean next gen software for licensing, engineering, inspection, construction, or supply chains. For natural gas, it could mean asset optimization, or tools that coordinate generation with electricity demand. For the grid, it could mean new tools that unlock capacity, accelerate interconnection, or give operators significantly better visibility into how systems are performing.
The energy system of the near future will require a huge amount of physical infrastructure, including more wires, turbines, batteries, transformers, substations, and generation assets. Alongside this physical buildout, however, the system will also require much more intelligence to coordinate those assets, make better use of existing capacity, automate complex work, and reduce the time it takes to bring new infrastructure online.
That intelligence layer is where we see a very large opportunity, and where we want to keep expanding both our investment activity and the ecosystem of partners around BGV.
If you are building in this space, I’d love to hear from you.