To understand how we arrived at this moment—and where the smart capital is moving to solve it—we have to look at how the workhorse of corporate procurement is evolving, and why physical constraints are forcing the market to innovate.

If you're interested in the continued conversation I have been having on this topic on X, you can find that here: https://x.com/ArushiSF/status/2055134572812062753?s=20

Read more

• Real Estate Investment Trusts: they provide the “Powered Land”—actual sites with grid connectivity and physical space for infrastructure like solar, battery storage (BESS), and modular data centers. Companies like Prologis have already secured gigawatts of power capacity across their portfolios and Prologis is specifically discussed below. Prologis, Inc. (PLD) is the world's largest industrial real estate investment trust (REIT).

• Energy Enterprise Resource Planning software: This is the digital brain inside the business model of commercializing powered land. The article discusses Arcadia/ENGIE Impact example, explaining that the ERP software layer provides highly granular telemetry needed to “see” into the grid. It allows operators to programmatically identify distribution-level soft spots and match compute loads to available headroom in real-time, and on a planning horizong, enable REIT-style deals to unlock that headroom for lucrative turnkey load energization or optimization solutions.

Boring Ol’ REITs - A Business Model Ready for the Edge

Edge compute is dominating the squawk circuit right now, and for good reason. Building on our recent piece covering Distributed Data Resources (DDRs)—a term coined by —we are witnessing a massive shift in the capital cycle. The market is moving aggressively from the gigascale, grid-connected projects of 2022–2023 toward a new era of distributed, megascale investments. The idea here, is that smaller, distributed nodes can be bundled and sold to an offtake customer to power computation seamlessly across a decentralized campus. The business model is beautifully pragmatic: move compute infrastructure directly to commercial distribution nodes where grid headroom already exists, and where well-placed power generation solutions are available—either through legacy load-serving utilities or Independent Power Producers (IPPs).

While it all feels so new and snazzy, the business has been shaping for a while, much of spurred by the boring old “REIT” business model. A REIT - or Real Estate Investment Trust - is a vehicle typically focused on acquiring and leasing physical buildings. However, the smart developers realized early on that controlling the real estate footprint means controlling the local distribution grid: between the relationship from the customer meter to the utility substation, and the customer’s load to the load-serving energy sources, there is a full stack of utility-like services waiting to be optimized. This piece focuses on how real estate giants have spent the better half of the decade quietly (and loudly) tooling their massive property portfolios into physical power banks. And, we explore what sorts of synergies having advanced customer portfolios can bring to the space, even where the real estate owners/operators are diversified but the load meters of those entities are under common management by a third party energy ERP firm.

Relatedly, I am seeing that large load developers are fundamentally driven by a “firm power or go home” mentality if they indexed to some version of this thesis somewhere around 2021-2022. My own anecdotal experience confirms this: the loads already scheduled for grid upgrades are largely indifferent to experimental energy setups. Instead, they subscribe tightly to their version of a distributed data center thesis, locking into energy supply agreements and converted interconnection points that benefit from a legacy strategy of securing the real estate and the firm power first. Because established REITs already control these global and local footprints, they are perfectly situated to execute this strategy at scale.

To be clear, the appetite for bespoke, ground-up gigacampuses requiring 2N grid-connected power has not dried up: it is just that the capital available to allocate to new projects of this dimension, is dwindling. So, we are seeing a diversification play from sobered capital providers. They are excited to back projects that already have critical boxes checked: controlled real estate, a customer thesis capable of operating across an aggregation of smaller facilities, and guaranteed power-to-load structures that build on commercial energy service already procured at those locations.

Prologis Case Study: From Landlord to Decenralized Utility

Industrial real estate behemoth Prologis (NYSE: PLD) is executing exactly on this distributed thesis. They have evolved from a traditional development REIT into a global digital infrastructure provider, managing 1.3 billion square feet across 20 countries. In Q1 2026, they formalized a massive operational leap, committing $8 billion to $10 billion in data center investments over the next five years.

This pivot is anchored by what Prologis terms a “5.6 GW power bank.” This represents an aggregated portfolio of utility power capacity attached to their global logistics footprint.

The evolution required core competencies and executive-level FTE strategists who executed on capacities well beyond the traditional REIT model and well before we got excited about edge compute. In 2023, here is what a former Prologis power executive was up to:

1. Overseeing origination, sales, development, engineering, execution, asset management, and operations across the U.S., Canada, Europe (12 countries), Latin America, and Asia (Japan, China, Singapore, India).

2. Growing Prologis’ operating rooftop solar and storage portfolio to hit a clear path for 1 GW.

3. Expanding the energy portfolio, delivering behind-the-meter and front-of-the-meter solar, storage, energy supply contracts, PPAs, vPPAs, and customer meter management solutions.

4. Develop and cement C-suite relationships with electric utilities in 22 U.S. states and 12+ countries, showing up to position Prologis Energy to the top 50 electric utility CEOs at EEI CEO and Board of Directors meetings. (Prologis ambition for distributed digital infrastructure sites is 85% U.S.-heavy today).

5. Partner with real estate teams to enable power access for high-energy-intensity customers, data center interconnections, and onsite/offsite energy solutions.

6. Build, own, operate a front-of-meter energy storage business, including grid-scale battery storage solutions and C&I behind-the-meter offerings across multiple markets. A key component of this is developing a utility meter management program to accelerate the deployment of onsite energy solutions (like rooftop solar and storage) for Prologis customers.

7. Assemble pipeline with several GWs of strategic grid interconnections; this means mapping high-value Prologis real estate directly to utility substations and transmission lines that actually have capacity, essentially turning dormant land banks into highly lucrative, ready-to-deploy power distribution networks across major U.S. and European cities. This is power studies and power flow modeling work that aligns high-value Prologis real estate with energy infrastructure in major cities across the U.S. and Europe.

By leveraging partial or total deregulation in certain markets, companies in this space can diversify how they generate energy, pair their “powered land” with their own power sources, and structurally dictate better economics.

Additional Sections below include:

1. How to Build a Power-First REIT

2. Enterprise Resource Planning is a Grid Utilization Unlock for Distributed Capacity (Of Any Kind, Not Just AI Load)

3. The Shift from Commercial Meter Optimization to Distributed Compute

4. Managing AI Load Volatility at the Edge

Read more

"Sharma Frank’s framing is probably right. Storage and inference together are both ways to move kilowatt hours to the time and place of maximum value, regardless of congestion across grids. The future for data centers is likely to look a lot like the grid’s shift from a few big plants to a distributed ecosystem. The transition is already underway."

Article: From DERs to DDRs

Author: Peter Hirschboek - ImpactECI

This article covers:

• Peter’s exciting piece released on March 12, 2026: From DERs to DDRs (What the Evolution of Electric Generation Can Teach the Data Center Industry).

• Arushi’s BloombergNEF Summit remarks on behalf of EmeraldAI, released on January 27, 2026: Energy and Technology in the 21st Century- From Innovation to Prosperity.

About DDRs: Distributed Data Resources

My colleague and a fellow advisor to Emerald AI, Peter Hirschboek of ImpactECI, recently coined a term to more wholly describe a phenomenon I narrated (by accident during Q&A) from the BloombergNEF stage in San Francisco (Jan 26-27, 2026).

Peter writes on March 12, 2026 about the “Distributed Data Resource” - finally coining a DER acronym non-energy people are already familiar with from another world (and even better IMHO, from the world of dance). This feat alone is worth applause in our dullard, self-aggrandizing energy wonk space where every acronym we release with flourish and pomp actually sounds to others like distant, piercing garble from poor-quality headphones. (We all need to try to be cooler like Peter).

Peter’s piece effortlessly moves through the winding ways of power grid and resource distribution evolution our grids have seen since right before the turn of the last century. It then moves right into the more compressed narrative of the history of compute and computing. He writes of the parallels thusly:

“Computing followed a compressed version of the same story. Local development first, then centralized, then distributed. The major difference from electricity is that computing had a distributed layer running in the background even before the hyperscale era fully matured. Both layers are now expanding simultaneously.” Article Link.

With this expansion, Peter reasons that both systems are experiencing parallel trends of costs and penalties when they cannot actuate on the investment narrative of either or both together. I’d wager that if they could, collective costs would certainly go down, and the two would have a happier marriage: that is, a procylical relationship would reduce penalty effects for one sector (chips), via necessary actuation of cost savings to the other sector (power grids).

He notes:

“Power systems bear the brunt of these types of penalties with line losses, congestion, and the capital costs. New high-voltage lines are slow to permit and sit across multiple jurisdictions, and can cost millions of dollars per mile. Large-scale transmission expansion, due to this, can be both hard and expensive. End-users also pay for transmission and distribution infrastructure, capacity payments, ancillary services, regulatory fees, and taxes. These stack-up charges can cause delivered electricity to cost as much as the underlying generation costs or more. The DERs that site generation close to load can reduce parts of those costs, even when their raw production per kilowatt-hour is higher than that of a large central plant.

For data centers, the parallel is structurally similar. The friction that these large facilities face isn’t a T&D rate schedule, but rather the cost and delay of grid upgrades and power interconnection. To counter this, a 10 MW substation-scale site that taps into existing distribution headroom can often move faster and at a lower interconnection cost versus a 100+ MW campus. These types of sites can bring latency and data sovereignty benefits that hyperscalers can’t. Even with per-unit facility costs being higher on a smaller scale, the system-level friction cost can be lower.” Article Link.

It’s the right frame of thinking, and clearly explains why someone like myself who dabbles in BESS and DER policy, mechanics and market design is sudd=denly yanking on 12 threads simultaneously in the realms of “AI Factories” and “Grid Integration” - and not to say the least, advising companies like EmeraldAI.

So today we are connecting even more dots on DDR strategy; below I am writing out my own speaking, instead of what I usually do, which is speaking my own writing.

Things I Said at BloombergNEF

On New Solutions and Designing for Dispatchability

“So Albert [BloombergNEF CEO], you and I talked about this and I use Substack a lot, and so does Pete Buttigieg. He said something yesterday or a couple days ago about how those who will own the future are the ones who can see what really comes next. And the same mindset for technologists is to be in that exact space, right?

So EmeraldAI is an example of a company that is in that state where we see what comes next. In one simple phrase, we have to design our power and tech solutions for dispatchability. Which means it is part of the economic demand curve of how we call on every asset, every element in the system, whether it’s the transmission system or the generation system, without thinking about tech solutions as the last resort, the catastrophic result, the thing that we have to pick up the phone to call—like MISO picking up the phone this week to call on load-modifying resources. It’s literally ‘hello, I need you now, but there’s 10 minutes of latency.’ We’re not going to live in that world. And we can’t afford it.

And all manner of communication—look at some of the issues that Sandhya (EDP Renewables) brought up. All of these processes, whether it’s how we build the grid, or how we move to concurrent, streamlined permitting, it all benefits from this approach of being part of a least-cost model. And our tech solutions, all of them, they all fit inside it. And they get us to a point where you can pick up the phone and say to the AI load, say to the generator, say to the battery, say to the coal plant, it really doesn’t matter: ‘come off, or help.’ And it’s instantaneous, and we can rely on it as a planning solution, and then people can rely on it as a security solution. And we get away from this world where we are so reactionary, as just citizens, about what AI could and couldn’t be for the country. It’s just going to be in every single part of what we do and what we build.”

On ERCOT, PJM and Transmission Infrastructure

“[I’m happy about policy movement where] FERC ask PJM to look at firm and non-firm contract demand for the first time. That’s a lot of higgledy-piggledy most people don’t understand. The simple version of this is that in most of the FERC or the federally regulated markets in the US, PJM included, our loads conventionally don’t pay for the transmission [system upgrade to the specific load project…. transmission cost allocation is technically two-sided in the sense that both generators and loads contribute to system upgrade costs and are studied in light of firm upgrades for moving power to and frot - though they do so through different regulatory “buckets”]. They accept service by asking their utility for power, and then the utility telling the RTO - ‘okay this is what we’re going to need’. Then PJM turns around and starts procuring all the supply and the reliability and the buffer, and that’s all a generation question.

And just for direct contrast, the exact opposite happens in ERCOT. The one non-federal RTO market. In ERCOT, load only pays for the transmission system. Load ratio share is spread pro-rata across ERCOT. So generation actually takes the system as it’s built. It’s the complete opposite.

And so if you’re working in both [markets], your head kind of explodes from day to day because the mechanisms are so different.

But the thing that excites me about this contract demand solution [in PJM] is that it brings loads that are having material impact on this system into a world where they will co-invest, not just in PPAs, but in the physical and the technological solutions to put the demand of load into socialized infrastructure cost, and build transmission at the same time as a load is coming.

That’s the most exciting thing to me. Right, because we all know this, but social transmission infrastructure as it grows, if load kilowatt hours, megawatt hours grow in proportion to it, the socialized cost remains relatively flat. And that’s what we really need. And we need to reverse that particularly in PJM where that is, because of the essential structure of how we pay for the thing, for regional transmission, we have this like wide gap between the fixed cost, that we’re spreading auction by auction by auction, and then the actual volumetric consumption where the load that’s coming isn’t really consuming yet, so they’re not paying into the system. The structural challenge and solving it very quickly I think is the core urgency. Our last speaker was talking about Governors coming to complain about PJM… “[W]ell, you’re complaining about an issue that has been a structural fundamental of this market for a long time in terms of how we plan and pay for infrastructure.”

On Reframing the Flex Narrative for Hyperscalers - Invest in Dispatchability Now instead of Unplanned Surprises Later

“So one of the things it looks like that resonates a lot [with me about dispatachabilty solutions] is that it looks like the opposite of Texas Senate Bill 6. It looks like this [in a future state]: we will ask loads to do a little bit as part of least cost dispatch, and a lot less because the system is about to conk. And once I explain that to… whether it’s hyperscalers or colo developers…I think the narrative resets completely [through my languaging]. That, um, if you’re building any wiggle room at all in the design of these AI load campuses regardless of the size…whether that wiggle room comes from workload, from managing like the thermal buffer at the site correctly, from locating batteries, from shifting the workload if it’s inferencing to a different actual physical site. All of those solutions, someone has to invest in them. And your mindset needs to be in that space where you’re putting yourself into an economically viable solution so that you’re not on the wrong end of the next grid emergency. [it means policy is a price signal to build for dispatachability now instead of this industry waiting for a grid operator surprise in the future which causes far more harm to the economic value of the load.]”

On Batteries, AI Inferencing, Both Moving Kilowatt-Hours without Wires

Dana (Moderator): “And so the questions I was actually getting through the chat were starting to be kind of technology specific. So you’ve given your answer to this next one. So turning to the rest of the panel, with data centers being this really important source of energy demand, what’s the role you see of batteries in that specific area?”

Arushi Sharma Frank: “…So a few things…one is that right now there’s a lot of very interesting argument about whether the batteries belong right next to the rack, they belong at the chip, they belong next to the server, they belong next to the diesel generator that’s stuck to the data center building. Do they belong in, uh, two-hour or four-hour durations just to stabilize the whole load so it doesn’t crack the next thing, which is the powertrain to the connection to the grid?

And then of course we’ve got many companies that are invested in this kind of like hybrid microgrid thesis, right, around using longer duration batteries to do different things.

Fundamentally, energy storage is a lot like what I think AI inferencing itself will be like within the next six years. And by that I mean that we are able to move the value of a kilowatt-hour across space and time without a wire, right. And the reason that inferencing is like energy storage in this way—and you all know this—is that a lot of us who are appropriately grid nerds, and we’re correct, like, battery is neither load nor generation. It’s really this, like, independent thing that we’re trying to stuff into several different market frameworks or rules or whatever.

… inference, much like energy storage as an arbitrage and a consume-less/consume-more solution, moves. It can move.

So one of Emerald’s demos that I’m now publicly allowed to talk about was, moving inference workload running Nvidia GPUs from a data center hotspot outside Chicago to another data center hotspot where there was room currently on the grid in Virginia. We just executed this demo and there’s going to be peer-reviewed results… coming out from EPRI DC Flex.

Um, but the smallness of that scale, that it’s a pilot—that’s also how energy storage started too as an industry, where we’re trying to figure out how you optimize across wires and without wires, and continue to optimize even when the wires get built. Because I think one of the things that’s lost right now is one of the reasons establishing surplus interconnection or load flexibility is so hard for our wires planners is, like, they intentionally built the system for a lot of redundancy. And just because there’s headroom, it doesn’t mean they want you to use it.

So we have two very different mindsets that have to come together. And I think if we get that mindset correct, it’s likely we evolve risk-based engineering.

…and then we’re also putting energy storage into the same space as inferencing itself. If you really believe that we’re going to shift from 90% LLM [training] to90% inferencing in five years, 10 years, pick your number—if you really believe that, then storage and inferencing together are both ways to move kilowatt-hours around to the maximum use place irrespective of congestion across grids.”

Designing for “Extra Headroom” Does Not Mean that Operators Want to Use It All - DDR Solves Inside this Inefficiency for Compute-Power Maxxing

As Peter flags in his DDR piece, I said at the BloombergNEF event that “One of the reasons that establishing surplus interconnection or load flexibility is so hard for our wires planners is that they intentionally built the system for a lot of redundancy. And just because there’s headroom, it doesn’t mean they want you to use it. So we have two very different mindsets that have to come together, and I think if we get that mindset correct, it’s likely we can evolve risk-based engineering.” Peter reasons that “[f]or the time being, this means DDRs should be able to scale more quickly than large facilities because finding the soft spots in that headroom is easier at 5 MW than 500 MW.”

Ultimately, the transition from DERs to DDRs isn't just a neat historical parallel—it is an operational imperative. We can no longer afford to keep forcing 1 GW hyperscale pegs into congested transmission holes.

I’ve really embraced the four-category framework in my own design-focused writing and narrative, if the surprise blast at Bloomberg wasn’t evidence enough of the same. I think we are moving into a more conscious period for data centers and DERs because we understand the layers of compute and where they fit better than the 2022 narrative of “big big in one location.” It’s a more conscious trade-off, informing our collective work to improve both sides of the system - compute actuation and grids. For actual, system-level speed to market, our work is to mitigate the very real, physical risks of hyper-concentration across either sector, and arrest the endless inertia with our wires planners over massive new interconnections that take a decade to build.

By co-locating DDRs with microgrids and decentralized power solutions, we will soak up curtailed and stranded electrons exactly where they already sit and reduce the complexity of the whole.

Bonus:

This piece started with the working title:

In addition to the amusement bonus, it is a subtle tip: keep on muddling to achieve your best work. Whilst AI tools sneak into our mindspace to make our muddles for us, leaving us to do nothing but robotically evaluate choices which we neither originate nor contend with ourselves, our cognitive autonomy hinges on our capacity to muddle. For more on muddling, an article you might read or listen to, from at the .

Cosmos Institute

The Last Temptation of Claude

In 1972, researchers offered children from the Bing Nursery School at Stanford a simple choice. Left alone with a single marshmallow, they were told they could eat it now or wait fifteen minutes and receive two instead. Years later, the scientists tracked down the original participan…

Read more

3 months ago · 247 likes · Harry Law

-A.S.F.

Bonus Bonus:

Wow, if you made it to the end, you might have read in Peter’s article that we often develop the edges of concrete thesis arcs on Whatsapp. In such an exercise, when I first met him, I explained in pictures sent over Whatsapp, how I got to where I am, designing systems and policy simultaneously to fit a DDR-maxxing electrostate.

• New debates around SPP policy for CHILL/HILL have cropped up, with all sorts of fascinating consequences for tariff uptake in 2026.

• PJM has… evolved?! FERC has been very busy. I have writing and signed onto comments, published articles, and spent serious time explaining complexities to staffers at agencies and lawmaking bodies on top of that, in my teaching capacity.

• ERCOT is in a hard spot, trying to catch up to where the market and the politicians want them to be on large load growth, resolving serialized studies’ delays for projects, deposits and commitments to show readiness, and controllable load resources.

• While ERCOT goes through this process, I have led the advocacy to retain a speed-to-power solution in operations for loads that pay their full freight in planning, but agree to curtailment via SCED in operations. This is the Controllable Load Resources solution.

• The United Kingdom has become a proving ground for comments I wrote last year on the situation with PJM’s IMM questioning the viability and cost-effectiveness of emergency-only style DR programs.

• California’s utilities begin the transition between distribution to transmission-level flex programs.

• EmeraldAI demonstrates computation (inferencing) workload geo-shifting from one state in PJM to another.

• The City of New Orleans, Entergy New Orleans, Energy Hub, Together New Orleans, Alliance for Affordable Energy, AND MORE… if you haven’t seen the Canary and Latitude Media pieces, and you are a solar-BESS champion, please check those out.

There is more. If I continue to write, however, we’ll loose the thread on what I’m here to tell you about today!

Announcing, the phenomenal work of EmeraldAI in the United Kingdom. Throughout the day, you’ll see press releases.

Bloomberg story: https://www.bloomberg.com/news/articles/2026-03-03/ai-data-centers-may-not-need-constant-peak-power-study-finds?embedded-checkout=true

Press Release: https://www.nationalgrid.com/uk-first-trial-ai-grid-technology-successfully-demonstrates-ability-data-centres-adjust-power-needs

Whitepaper: https://ngpartners.cdn.prismic.io/ngpartners/aabLtVxvIZEnjRNr_v11_emerald-ai_march2026.pdf

For all of my subscribers: The doc linked below is the first of what I hope will be many technical team reviews Emerald AI leads to help be a part of the answer on how we create forward-looking tech solutions to marry planning and operations, agnostically in capacity and vertically integrated markets in the U.S.

Here is what that actually means in practice:

• Replacing Overbuild with Software: Instead of planning costly grid upgrades for ‘worst-case’ peak loads, we can use smart software to dynamically flex AI demand when the grid is stressed.

• Market Agnostic: This isn’t just a niche PJM solution. This capability works everywhere, from deregulated capacity markets to vertically integrated utility territories.

• Location Doesn’t Matter- Trustworthy, Layered Dynamic Response Does: Whether an AI factory is co-located directly with a power plant or connected to the wider transmission system, computation itself becomes a lynchpin to enroll willing loads into voluntary, dispatchable grid service. It also becomes a backstop for non-firm/temporary flexible consumption approaches, like supplemental interconnection service that could give a load more power over its initial firm allocation by drawing on a secondary or emergency rated line. The word “layered” is important - recall how the PJM FERC Co-Location narrative includes guarding against “drifting.” Simply having the capacity to orchestrate compute to stablize load draw, is enough to give a second layer of backstop reliability to a transmission utility that wants to enable batteries or other dispatch solutions for a load, but wants a true mechnically and electrically viable backstop to ensure that if solution A fails, solution B solves for the NERC obligation behind it to protect the grid.

Emerald AI 2025 Holiday Filing: How Compute Supports Colocated and non-Colocated Load Dispatch

Read more