At Innovation Endeavors, we have invested in the intersection of emerging technology and the physical economy for over a decade. In the construction industry specifically — the largest portion of the architecture, engineering, and construction (AEC) lifecycle — the macro market conditions continue to increase the urgency for innovation. Construction represents about 8% of global gross annual output and 8% of the global workforce, and demand is projected to grow by 70% globally by 2040 — climbing north of $22 trillion — expanding to almost the size of the U.S. annual GDP today. With that backdrop, the past decade has already proven that this trajectory will be impossible to actualize with traditional approaches. In the U.S., the average number of vacancies in construction labor doubled from 2017 to 2023, and it’s expected that 40% of the pre-2020 workforce will retire by 2030. Amid shrinking workforces in advanced economies, labor alone cannot solve flat productivity, much less industry growth; we will inevitably need technology to meet the moment. 

AI has the potential to bring unprecedented productivity gains to the construction industry and AEC more broadly. The opportunity is compelling not only because the industry has historically been slow to adopt digital technology — and therefore, the potential for improvement presents distinctly compelling white space for startups — but also because AEC runs on a uniquely vast amount of structured and unstructured data. The amount of documentation is staggering if we include the full end-to-end labor of a commercial construction project, from architecture and design to project completion. The full lifecycle includes architectural drawings, concept designs, blueprints, engineering plans, BIM models, CAD models, site plans, shop drawings, materials specifications, cost estimates, project schedules, and more. Construction spends <1% of revenues on digital technology, less than a third of what’s common in industries like automotive and aerospace, yet the documentation for a large project typically spans tens of millions of pages, which are often physically printed out. This treasure trove of data, which documents the details of multi-year projects employing thousands of workers, is a valuable training set for LLMs, agents, and automation. However, because most of this data is not public domain and generally isn’t labeled, more generalized model providers are unlikely to win the day or be the most performant and reliable in producing geometric, volumetric, and spatial representations.

So, what value levers, AI applications, and startups are most exciting, and how can we buck the historical trend in AEC to adopt new technology and increase productivity? Some of our recent investments in AI x AEC — like Trunk Tools and Parspec — are pointing the way. Trunk Tools is building the brain for the construction job site, ingesting millions of project documents to create a knowledge graph that will power agents for search, task-based incentives, and beyond. Parspec is automating large portions of the pre-construction workflow by creating an advanced configure-price-quote and product search platform for distributors, digitizing procurement across the industry. Both are already selling to many of the largest builders and distributors in the U.S.

Earlier this month, we hosted a salon-style dinner with leaders from North America's leading general contractors, architecture firms, wholesale distributors, and startups to share what’s top of mind. 

Three broad industry themes emerged from the conversation:

1. Requirements for the data needed to train models are quickly changing, creating excitement and uncertainty: Recent progress in AI research around fine-tuning and few-shot learning has been remarkable. Every month, we see significant improvements in the ability to produce accurate results based on small amounts of labeled examples. As noted earlier, these developments are particularly important in areas like AEC, where training data is scarce and typically under lock and key within architecture and construction firms. This specialized data need is a major reason why the largest model providers won’t naturally be the ones to solve the unique challenges of geometric, spatial, and volumetric representations, and therefore, the opportunity for startups, along with legacy players like Autodesk, is compelling. Leveraging relatively small numbers of high-quality designs, blueprints, and engineering documents — combined with the power of ever-improving pre-trained models — will be a major unlock for the industry. 
2. Predictability is the core value driver for digital technology in construction: The construction industry is notoriously bad at predicting costs, timelines, and requirements. This presents a massive challenge for builders, given that their entire margin is the delta between their bid on the project and the final cost to complete construction. The ability to win in the bidding room and execute on that budget over a multi-year project lifecycle is top of mind for key decision makers. Machine learning approaches, along with generative AI, can provide dramatic improvements here, particularly given that, for the most part, the industry has not seen the uplift promised from the last generation of building information modeling (BIM) and cost estimation software. Stitching together materials, labor, equipment, overhead, contingency, and regulatory data will require fundamental improvements in how builders log, label, store, and access vendor and historical data. 
3. More repeatability, modularity, and “kit-of-parts” approaches are key to bringing down costs: Automate the 80% and then tune the 20% — this is the path forward, according to many innovators in AEC. Historically, each project has been treated like a special flower, with minimal learning and work transferred from project to project. Companies like Katerra failed in large part due to being stretched too thin across too many non-uniform and custom projects. There will always be the Burj Khalifas of the world, but most buildings look like countless buildings before. Customization creates challenges not only because it’s inherently more complex but also because it contributes to the huge amount of tribal knowledge living in the heads of experienced AEC knowledge workers, which is not easily accessible in any central repository. Soon, many of these wise men and women will retire. The historic challenges of costly one-off designs, repeating mistakes, and limited automation will worsen. The good news is some companies are already efficiently reducing complexity and deploying infrastructure for repeatability. Aro Homes, a verticalized home builder incubated at Innovation Endeavors, is taking a product-first approach and allowing buyers to experience their future home before they buy. Home-buyers can customize to their desires, much like car-buyers can customize their Mercedes, such that no two vehicles are the same but still benefit from significant cost reduction and high-end design. In Aro’s case, customers get an Olsen Kundig-designed home that normally might be out of their price range.

With these trends in mind, here are some of the innovations we are excited to see in AEC:

1. Models that can produce geometric and spatial representations while understanding design intent: Imagine if you could record a conversation with a client describing their vision and immediately produce initial concept designs. Imagine if architects could extrude from 2D to 3D effortlessly (i.e., as described by Autodesk research) and auto-generate parametric logic, radically collapsing design cycles. Imagine if you could pull forward initial structural engineering reviews and bake them into the design stage, validating constructability and materiality much earlier in the process. These tremendous improvements are quickly becoming possible and will dramatically reduce costs and increase speed. 
2. New data representations, like entity component systems (ECS): Beyond the challenges of understanding geometric relationships and how to tokenize and normalize them, there is exciting potential for fundamentally new data and systems approaches in AEC. Historically, some of these techniques have been mostly used in video game development. ECS is a software architecture that follows composition over inheritance, potentially unlocking more modular, scalable, interoperable, collaborative, and simulated AEC workflows. This could solve a lot of today’s challenges that follow from siloed and heterogeneous systems and data types. 
3. Automating the shop drawing process: If construction has one master governing class of documents, it’s the shop drawings. Producing them is one of the most labor and cost-intensive parts of the lifecycle – when engineers, fabricators, contractors and suppliers translate vast amounts of documentation into detailed, technical instructions for the build process. With the right models and workflows, agents and automation could radically slash costs. 
4. Permitting automation: Permitting is one of the least glamorous but most painful steps in getting anything built. Local building codes, zoning laws, environmental regulations and safety standards create delays and friction, costing builders precious time and resources. Companies like PermitFlow and GreenLite are innovating here, and we expect to see new players going deeper into the value chain over time.
5. Tools to structure and classify the vast amounts of historical project data: Given the lack of centralized and accessible data noted above, the vision of a unified data model across materials, products, designs, fabrication, costs, and scheduling is tantalizing. If AEC is going to improve toward making original mistakes, builders and architects need to learn from everything they have ever done across the project lifecycle. Technology to make sense of the vast, multi-modal troves of historical data is rapidly improving and will help solve the lack of transferability across projects, as well as associated rework.
6. Collaboration tools that don’t unduly increase costs and complexity: Given the dozens of stakeholders in the lifecycle of a project across owners, architects, engineers, general contractors, subcontractors, fabricators, suppliers, and more, collaboration is a massive challenge in AEC. Tools like BIM have improved things, but builders note high implementation costs, steep learning curves, integration issues, and sometimes further delays. Agents can offer a new era of seamless collaboration as more of the workflow is automated and abstracted away. 

Historically, construction has been governed more by fear of loss than hope for gain. Founders building in the space must have a deep appreciation for the industry's relationship-based, risk-averse nature and recognize a fundamental asymmetry of risk in decision-making. Economic buyers across the value chain bear massive risks, from legal liability to hefty penalties for delays and, most critically, errors that cause rework. In the U.S., it’s estimated that rework alone costs the industry $50-100 billion per year. This wariness of new tech that might cause errors is not totally irrational, but it is precisely what’s holding back the solutions that could actually reduce rework in the first place.

Innovation will require high trust and easy-to-use tools. The industry has been pitched shiny objects, from drones to sensing and monitoring to project management apps, for decades. Few companies have been able to deliver scalable value, with incumbents like Autodesk and Procore continuing to dominate the market. Building excitement and adoption for new AI capabilities and aligning incentives within massive, slow-moving orgs will require technologists and industry veterans to join forces. Perhaps most critically, if AI can be understood as an augmenting force multiplier rather than a replacement risk, the AEC industry can become a leading beneficiary of the remarkable platform shift underway. If you’re working on bringing AI to AEC, we’d love to hear from you.

This will be a 2-part series, beginning with the problem statement, followed by the solutions we are excited to see.

‍

Compute will be the currency of the future. If that’s true, then it will be pegged to energy. This reality — that scaling AI hinges on scaling our electrical infrastructure — has become starkly clear in the last year, leading AI leaders like Sam Altman to acknowledge that a platform shift in computing will require massive amounts of power. In March, Altman said that “energy is the hardest part” of the journey to scale AI, and this week, he announced his investment in Exowatt, a solar power data center company. However, few AI leaders will explicitly acknowledge the more concerning truth, namely that deploying massive compute resources and interconnecting them to the grid will directly displace electrical loads that are actually decarbonizing.

AI is simply more energy demand, no different than more air conditioning (granted, AI can likely provide broader solutions to climate change over time). All of the electrical load needed for EVs, green industrials, and electrifying buildings will be in competition for capacity on the grid — capacity that was scarce even before the advent of AI. In other words, independent of AI, we will need at least three times more electricity to decarbonize the global economy, and as data center demand explodes, our finite grid capacity will be strained by both emissions-reducing electrical loads and computing demand. Therefore, the bottleneck on the grid is one of the most important challenges facing the energy transition, but it’s also quickly becoming one of the most immediate risks to scaling AI. Solutions that allow AI to scale — without displacing critical decarbonization infrastructure — are essential to avoid derailing our collective emissions reduction goals.

The relationship between emerging digital technology and energy has historically been overlooked and is only now coming to the forefront of the public discourse. The venn-diagram of experts who understand the esoteric nature of power markets and who understand the future electricity demand of AI is small. This lack of overlap — combined with the utilities being highly regulated, poorly incentivized, and slow to adopt novel machine learning and data science approaches — is, in large part, why the multi-trillion dollar energy sector and the grid itself have remained slow to innovate and anticipate impending complexities.

Electricity consumption in North America and Europe has remained relatively flat for the previous two decades, but since 2020, as the energy transition to clean electric power for mobility and buildings has taken off, we’ve arrived at an inflection point. For some forward thinking folks (kudos to Evan Caron, Shayle Kahn, and Chase Lochmiller), it’s been clear for years that the megawatt hour (MWh) was going to become the fundamental currency of innovation. As AI joins the party, on top of already climbing electricity demand for the energy transition, that future is here. Some estimates indicate the amount of electricity required to power the world’s data centers could jump by 50% by 2027, and this month, nine of the top ten U.S. electric utilities indicated that data centers are a main source of customer growth.

So why exactly is this global experiment so challenging?

Part 1: The problem — the electricity gauntlet

Before the AI inflection point

The escalating supply <> demand crunch on the grid has been a slow-motion car crash for decades, coined by Energy Impact Partners as the electricity gauntlet. To understand how we got here, we need to dig into how the planning and expansion of electrical load on the grid is managed.

In order to plug any generation or electrical asset into the grid, it must go into the utility or Independent System Operator (ISO) interconnection queue. The ISO is an NGO that is generally responsible for ensuring the reliable operation of the electrical grid within a specific geographic area, overseeing the transmission of electricity, managing grid reliability, coordinating power generation and distribution, and facilitating wholesale electricity markets. ISOs also often develop and implement standards and protocols for grid operation, including management of grid congestion, maintenance of frequency and voltage, and compliance with regulatory requirements.

Electrical utilities play the most important role: they own and operate the actual grid infrastructure for generating, transmitting, and distributing electricity to consumers within a specific service territory. Utilities are responsible for building and maintaining power plants, substations, transmission lines, and distribution networks to deliver electricity reliably and safely to homes, businesses, and factories. They typically provide services like billing, customer support, outage response, and energy efficiency programs to end customers.

In unregulated markets such as Texas and parts of the northeast, the ISO handles the lineup of generation and new supply, like power plants, solar, and wind, while the utility manages the demand side, encompassing all electrical loads on the grid, such as buildings, charging infrastructure, and increasingly, data centers. In regulated markets, the ISO and the utility are more tightly coupled, and the utility is responsible for both, such as PG&E in California.

The Federal Energy Regulatory Commission (FERC) regulates the power sector and strictly manages the communication between ISOs and utilities. Even within the same organization, like PG&E, planning and interconnection management is often not well coordinated across generation and demand. Siloed data and fundamental mismatches across large geographies lead to huge delays, bottlenecks, and backlogs in interconnection. The other way to slice this problem is transmission scale vs distribution scale. If you’re small and don’t need system upgrades to deploy your project (distribution scale), developers can work with the utility, whereas if you’re larger and stand to impact the transmission network (transmission scale), the ISO gets involved.

The complexity of this planning and interconnection process is exploding as the grid becomes more decentralized and intermittent with increasing renewables penetration. The grid was designed in the early 20th century for large-scale, dispatchable power plants (i.e., coal and natural gas). Decades later, accelerating penetration of smaller, less predictable assets (e.g., rooftop solar) is making scenario modeling and optimization significantly more difficult for these legacy organizations. Today, roughly 80% of power is dispatchable in the US, and declining at an accelerating rate.

Perhaps the simplest way to describe the current struggle of the grid is the interconnection queue backlog. In the last two years, the interconnection backlog has grown 40% a year, exploding to 2600 GW (which is more than all existing U.S. generation today), 80% of which is currently solar and storage projects. The time to interconnect has ballooned from two years per project in 2010 to five years today. In many ways, this is a tremendously encouraging signal of progress to electrifying our economy. The global deployment of renewable energy grew 50% year over year in 2023, and at least a 20% CAGR is expected for the coming decade in the U.S.

This growth rate is both astounding and unexpected. For example, The PJM Interconnection, the largest ISO in the U.S, this year tripled its growth expectations for electricity use over the next decade in its service footprint. This rate of change has surprised even the experts, and speaks to accelerating electrification and the fragmented coordination and planning process across key constituents.

Satisfying the accelerating demand for EV charging and building electrification will be a massive undertaking. California recently published a study indicating that the state will require $50B in grid, transmission, and distribution upgrades by 2035, after a decade of lower-than-historically-average infrastructure deployment. This estimate could be off as a result of regulatory decisions, grid program design decisions, or market decisions based on customer or auto OEM behavior, but are likely directionally right. To electrify 13M vehicles — the 2035 target for California — will become a more than ten-year interconnection problem.

Utilities, through no fault of their own, are historically quite poorly incentivized to find the most optimal and low-cost path to grid upgrades given the rate payer model, which mandates they only get paid for building more CAPEX like transmission lines, substations, and transformers. For this reason, utilities pay less attention to optimizing distribution (the grid downstream from a substation that connects transmission to buildings and infrastructure), and instead focus on building more stuff, even if it’s sometimes redundant or overbuilt. Utilities are increasingly experiencing access to capital challenges and declining creditworthiness due to the recent disasters around fire and mismanagement.

Despite all these challenges, there is some good news in terms of utilities’ role in addressing the escalating electricity crisis. Utilities certainly want more load on the grid so they can charge consumers for it. Therefore, if it becomes unaffordable to buy an electric vehicle due to grid constraints, they won’t see gains in demand. Utilities are eager to scale load — EVs, building electrification, and data centers — but they are less excited about the challenges of intermittent generation. Grid operators have historically been slow and unoptimized in how they go about addressing this challenge. Jigar Shah, the director of the Loan Programs Office in the US Department of Energy, is addressing this by pushing for change.

After the AI inflection point and the coming crisis

Everything described above was true long before the explosion of AI and compute demand, and that demand has arrived with remarkable pace. The CEO of Arm, the $100B chip company, recently highlighted that AI models “are just insatiable in terms of their thirst” for electricity. “By the end of the decade, AI data centers could consume as much as 20% to 25% of U.S. power requirements. Today, that’s probably 4% or less.” Some states like New Mexico expect 5x more power usage in the coming decade than the last 100 years combined. Texas power company Vistra has seen its stock surge 85% year to date, largely based on AI hype.

All the major tech companies have woken up to these challenges, and many have been planning how to strategically manage electricity needs for over a decade. Google has been a leader in carbon-free energy for years and has made strong progress in hitting its 2030 goals of 100%, 24/7 clean power for all operations, including data centers. But, now, as their appetite for compute, and thereby more electricity, has reached a fever pitch, the path forward is becoming fraught. Bill Gates and 20 tech exes gathered at a summit with utilities in March to discuss this exact issue, which is surprising for an energy industry that tends to take years to respond to new market conditions. Similarly, Texas’ ISO, ERCOT, is hosting its inaugural innovation summit this year.

So, where do we go from here? How can we scale compute and hit our decarbonization goals? The solution will require mass coordination and capital across grid operators, utilities, energy developers, tech companies, and startups. In our next piece, we will share some of the solutions we hope to see.

Thank you to Evan Caron, James McWalter, and Aram Shumavon for reviewing this.

By Josh Rapperport and Sam Smith-Eppsteiner

‍

In Part 1, we laid out the history and inflection point behind how we arrived at the emerging electricity bottleneck. Now, we will discuss some of the potential solutions. These proposals will focus on areas where technology and startups can play a key role.

Planning and pre-construction

If we are going to make it through the “electricity gauntlet,” we need to have sophisticated processes and workflows before anything gets built. This planning extends from the management of the interconnection pipeline, discussed previously, through to the energy developer that actually builds projects. Given some of the utility workflow is difficult to intersect with technology — due to slower adoption cycles and regulatory limitations — much of the heavy lifting, qualification, and planning for projects fall to the energy developers.

Today, energy developers are woefully underserved with modern technology and software. For the average utility-scale project, soft-costs (cost beyond the actual materials and construction) are around $400k per project, around 12% of total system costs, and climbing. This represents $10B of soft costs annually for U.S. energy developers, and that figure is expected to grow multiple times over as the market expands and we continue to accelerate the deployment of clean power, energy storage, and data centers.

The challenge for energy developers is largely one of messy, decentralized data across a multi-step workflow. The complexity lies in the combination of land parcel information, grid data, local permitting challenges, and the boots-on-the-ground human behavior in any given municipality. There are more than 20,000 townships in the U.S., all with unique codes, siting constraints, and permitting requirements. This massive amount of data is, at best, hidden away in hundred-page PDF documents on government-maintained websites and, at worst, not digitally recorded at all. The grid is an even more vexing data problem. Not only are public records of grid transmission lines unreliable and outdated, but the load capacity maps that indicate where there is free capacity are often not public and require expensive pre-applications to unlock.

At a fundamental level, the rate of project delays and cancellations is not improving, but is instead climbing. Developers see 80%+ failure rates today — potential projects that die somewhere along the multi-year journey prior to construction. If you ask a big developer about their pipeline, they will gladly share the gigawatts of projects they have underway. But if you ask about their risk-adjusted pipeline — accounting for permitting, siting, feasibility, interconnection, and financing risk — they likely don’t have a precise answer. Ideally, developers would spend their time on the most valuable 20% of projects that will actually get built, but having this type of continuous pipeline analysis is difficult.

Given the rapidly expanding ecosystem of small energy developers (aka wildcatters), the development process is becoming even more cumbersome as some developers resort to buying up as many easements and land options as possible and then sitting on them and waiting for buyers or future projects. Given big tech companies have excellent credit, strong cash flows, and are distinctly price-insensitive and time-sensitive in getting their systems up and running, we can expect these dynamics to escalate as they try to get an edge in the interconnection queue for their data centers.
Ultimately, this macro challenge all comes down to understanding the best use of any given parcel of land, based on real-time constraints of electricity supply and demand. Where should we build data centers vs charging vs battery storage vs solar? It’s untenable to brute force the problem with more transmission alone, given the $50B price tag in California.

Today, developers use a wide swath of point-solution tools to try and answer these questions and manage the multi-year project pipeline.

1. In GIS (a decades-old mapping tool), they maintain buildable acres maps including data like flood plains, endangered species, slope, and indigenous land.
   ‍
2. Then, they search for the interconnect data across public grid capacity maps, and they often need to manually draw 3-phase transmission lines and map the substation capacity.
   ‍
3. Next, the developer collects land parcel data across county, municipality, township etc, looking at ownership data, permitting, zoning, setbacks, and siting.
   ‍
4. Finally the human element comes in, which is distinctly difficult to capture with data. Developers work with land agents, send out mailers, visit city council members, and meet with local politicians and lawyers to find opportunities where individuals are open to solar or data centers.
   ‍
5. Finally, geotechnical and environmental consultants are needed to sign off and green-light the project.
   ‍
6. Many of these steps are run in parallel to keep projects moving forward, making it difficult to keep the numerous workstreams aligned given the complex project management process required.

‍

Startups and what we hope to see

Numerous exciting new companies have emerged to address this painful and inefficient workflow. Paces aims to build a much easier-to-use platform that captures all these layers of data in a single source. Furthermore, Paces hopes to capture the complex interdependencies across projects and geographics, some of which are counterintuitive. For example, most energy developers are all looking at the same hundreds of counties to build projects, depending on where there is favorable permitting, regulatory frameworks, and available grid capacity. One might think assessing neighboring counties would provide the most greenfield opportunity, but these geographies actually become less receptive to development as local sentiment changes, and NIMBYism kicks in. Defining and surfacing ever-evolving project risk also requires highly up-to-date data on the grid and quickly changing regulations, such as the new ruling last week announcing the Energy Department will take over as the lead agency in charge of federal environmental reviews for certain interstate power lines. This intersection of land information and grid constraints is a compelling opportunity for startups, with folks like Kevala, Transect, Neara, and Spark all building software solutions. For permitting specifically, tools like PermitFlow and Pulley are gaining traction within energy development and more broadly. To win, these tools must be 10x better than what developers use today, given a strong build vs. buy bias within the largest players.

We expect to see new kinds of solutions both in business model and technology:

• Grid modeling: Given that utilities remain the least innovative stakeholder, there may be an opportunity for third parties to provide advanced modeling and scenario planning for the grid. Building a data lake for utilities and running the multi-dimensional analysis of planning and best-use is very compelling, though it’s worth noting utilities can be challenging customers for a swift go to market.
• Energy developer soft-cost automation: Another interesting potential innovation is to build a more full-stack vendor to offload much of the manual and expensive soft costs from developers. We imagine a third-party standalone company that can provide the extensive suite of required paperwork and analysis such as FEMA studies, habitat and wetlands studies, wildlife studies, geotechnical analysis, surveys, title search, title insurance, easements, lien search, historical designation studies and more. This kind of work is well-suited for automation and LLMs.

• Government data API: Given the amount of government-related data — from regulation to permitting and siting to subsidies — required not only in grid-scale energy development, but also in small residential projects (e.g., heat pump install), we are envisioning a data infra layer that surfaces all relevant information to developers, contractors, and financing stakeholders. There may be an opportunity to build horizontally here to support the myriad end use cases that require an understanding of government context.

Crypto was one of the first technologies to get down to this level of detail, building bitcoin mining operations near the cheapest 24/7 power or using curtailed power like solar. This brings us to the next component, the orchestration of electricity with compute and other time-dependent loads.

Orchestration

To avoid hundreds of billions being spent on grid upgrades, we need to capture the maximum value of flexible assets — assets that can time-shift electrical supply and demand — like batteries, charging, and compute. Time-shifting can be thought of as the ability to temporally coordinate electrical demand and supply down to very granular levels and smooth out the sharp peaks throughout the day. Historically, electrons on the grid are consumed in real-time the moment they are generated, meaning that the entire grid is designed to accommodate the peak load in a given year, which is typically a hot summer day as cooling systems kick in. In summer 2019, electricity prices in Texas spiked to a record $9000 per MwH (200x the median price), and similar peak events also occurred in the winter of 2021 with the massive Uri storm freezing the state, causing hundreds of billions in damages. With warming summers, more extreme weather, and increasing grid unreliability (both with increasing demand for electricity and increasing supply intermittency), these events will only continue to escalate in frequency and severity.

The good news is flexible loads have huge potential to solve this problem. For example, in mid-August last year, a 1.8 GW discharge from batteries onto the grid in Texas quickly reduced wholesale electricity prices from $5,000 to $2,700 per megawatt hour. Energy storage, largely in the form of batteries today, is the ultimate flexible asset as it enables optimization of charge and discharge. However, vehicle charging and some kinds of data centers also have very compelling time-shifting potential on the demand side. This is because compute doesn’t always need to be running 24/7 at constant intensity (i.e., inference demand — which constitutes 85% of all data center AI workloads — coming mostly during the busiest hours of the day) unlike crypto mining operations, for example. Certain computing workloads can be timed to when prices are cheapest, cutting operational costs by up to 50%, as can a significant part of vehicle charging.

It’s difficult to say exactly how much electricity is time-shifted today, but it’s likely in the single digit percent. Many experts believe that number can be closer to fifty percent or higher. To get there will require sophisticated technology to orchestrate millions of vehicles, batteries, high-performance computing systems, and more.

Startups and what we hope to see
The critical components to make advanced orchestration work are data and controls, and financial infrastructure. We are seeing exciting innovation in both dimensions, and expect many generational companies to be built here. Orchestration, as discussed here, is inclusive of programs like demand response, virtual power plants, and synchronization, and is paired with the increasingly decentralized nature of electricity.

On the data and controls side, the key is to have real-time awareness of as many electrical loads as possible, accurate forecasts of both generation and load, and the ability to optimize the dispatch across as much of the grid as possible. On the more commercial side, companies like Equilibrium Energy are building advanced models of the grid to unlock situational awareness and then using large battery portfolios — enabled by a tolling fund or agreement — to optimize price and performance across consumers. Relatedly, Verse is building tools for corporates to unlock the PPA-for-storage model, which will allow customers to orchestrate and optimize their own energy needs. On the consumer side, companies like Branch Energy and David Energy are reimagining the energy retail model, which has historically been infamous for its low NPS, by orchestrating batteries and home devices to provide cheaper, cleaner power.
On the financial side, for energy markets to become more highly orchestrated and efficient, we need the type of advanced financial market infrastructure that many other asset classes have had for decades. Today in the U.S., tens of billions of dollars of power are traded daily using antiquated, over-the-counter approaches that only allow for large power blocks and 4–8 hour time windows in day-ahead markets. We see an opportunity for real-time, granular futures products in energy that would enable sophisticated hedging and risk management for asset owners and operators (more exciting news from Innovation Endeavors to come here soon!).
‍
We are keen to meet teams rethinking this massive orchestration challenge across data, storage, time-shifting, and financial instruments:

• Grid enhancing technology: This includes innovations across a variety of systems, such as dynamic line rating, which uses better data to understand the real-time circumstances of the grid and allow for more power to flow, optimizing beyond today’s conservative line ratings. Transformers as a grid-enhancing technology have also not been updated in decades. They are both supply-chain constrained and not designed for more advanced orchestration, contributing to conservative line ratings.

• Bi-directional EVs: We hope that EVs can be fully utilized and orchestrated as part of the solution to time-shift power, both on the demand response side and offering capacity and power to the grid, with vehicle-to-home and vehicle-to-grid approaches.

• Beyond bi-lateral energy: More generally, as far as financial instruments, the current Power Purchase Agreement (PPA) model is a fairly brittle and crude system, and we welcome conversations with teams innovating beyond bi-lateral agreements in power.

Building more behind-the-meter generation

Perhaps the most fundamental solution to the backlog on the electricity grid is to not rely on it at all. This is what behind-the-meter energy assets represent: instead of consuming power from a far away generation system and patiently waiting in the interconnection queue to gain access to the grid, energy developers and consumers can build the generation and consumption in one location. This means they can use their own power, on-site and “behind” their meter, and scale their infrastructure accordingly. There are important cost considerations given that behind-the-meter can be more expensive, but to put it simply, as much energy as possible should be behind-the-meter.

Behind-the-meter is a return to the earliest use of electricity, long before the complex network of transmission and distribution was built, where all generation and transmission was more locally linked. Modern innovations like microgrids have seen meaningful uptake, allowing customers to build their own infrastructure on their side of the meter. This approach can be difficult to scale, but it provides a compelling alternative to large-scale grid upgrades.

Startups and what we hope to see

More recently, many core innovations on both local generation and behind-the-meter compute have taken off. Most fundamentally, the self-supply of energy is beginning to realize its huge potential as modular reactors become cheaper and much more efficient. Small modular reactors (SMRs) and microreactors for nuclear fission, long only used in maritime applications, are being pushed forward by companies like Radiant. Microsoft has hired a director of nuclear, and plans to make major investments in scaling nuclear-powered data centers. Thermal and battery energy storage can typically reduce carbon emissions most effectively when deployed behind-the-meter, with companies like Antora bringing more scalable and cost-effective thermal storage solutions to market and Form Energy developing long-duration batteries. Advanced geothermal energy is getting traction, with companions like Fervo and Quaise offering the ability to build behind-the-meter baseload.

AI brings a unique opportunity to optimize across the entire hardware stack, from energy through to the chips themselves. Companies like Cruso have pioneered utilizing what would be wasted behind-the-meter energy, in the form of flared gas, to power data centers. Verrus spun out of SIP last month — they plan on using microgrids, sophisticated energy management, battery storage, and advanced chip sets to build cheaper and more optimized data centers.

•Behind-the-meter full-stack: We welcome conversations with teams building next-gen behind-the-meter systems that combine on-site generation with energy use, storage, or compute.

• Behind-the-meter analytics and software: Many energy and data center players won’t want to build software layers for integration, controls, and orchestration, and may outsource, creating opportunities for startups. Generally speaking, AI training clusters can be deployed essentially anywhere in the world that makes economic sense—as opposed to locally constrained assets like charging — so we are interested in analytics and infrastructure-intelligence software to help find untapped value. Oftentimes, sitting behind-the-meter is about understanding where there are already sunk infrastructure costs and innovating in how we can repurpose them for an expanding an electric and AI future.

Closing thoughts

It’s worth noting that much of our ability to navigate the electricity gauntlet will depend on policy, which is not discussed here, given it remains generally outside the scope of technology innovation. Structural changes like “first-ready, first-served” for the interconnection queue or policy that mandates that data centers participate in demand response programs would have a major impact.
Additionally, huge amounts of development effort are going into designing more energy-efficient chips and compute-efficient models, which will hopefully help mitigate the ballooning electricity demand for AI.

Independent of better policy, chips, and models, there is an enormous opportunity to reimagine the power sector so that we can continue accelerating the decarbonization of our global economy and scale AI infrastructure. At the most basic level, energy needs to be at the center of any techno-optimistic vision of the future, and we look forward to supporting founders working toward this ambitious goal.
Thank you to Evan Caron, Aram Shumavon, and James McWalter for reviewing this.