Developing Power Into The Data Center Hype

Background

Setting aside superfluous mainstream use cases, entirely worthwhile applications of artificial intelligence appear poised to accelerate global load growth at alarming rates, resulting in delayed thermal retirements, and probably a non-trivial amount of new thermal plants. In the world of power[1] development, hardly an hour passes without this theme threatening to influence project development strategies and choices. But how should data center load growth projections influence power project development strategies??

So much has been written on the panacea of data center load growth and “hyperscaling[2]” – first in the context of the cryptocurrency boom, and more recently artificial intelligence – that I hesitate to contribute to the hype-cycle. The most vexing variant of this chatter suggests that – barring any dramatic transmission improvements or changes in electricity consumption patterns – extreme versions of the near future would have us regress toward a dirtier (on balance) energy sector, vanquishing hard fought momentum the other way[3]. However, this is hyperbole stemming from an underappreciation for i) how the hype-cycle works, ii) short memories, and iii) how hard it is to develop any sort of infrastructure, data centers included. ?The graph below shows global i) projected renewable energy generation additions compared to ii) projected data center load growth, which reveals enormous room for fossil retirements amidst the load growth.

But, while I don’t believe we’ll see actual regression, the fact that this scenario is taken seriously by serious people – and AI hyperscalers’ plans are not, then, immediately cancelled – conveys the severity of the situation.?

It sounds to me like us humans – viewed as a global society – are not prepared to voluntarily slow down the implementation of power-intensive technology, even at the climate’s peril. Reasonable people may disagree on whether that’s “right”, or not, but that debate is purely academic: there will be no self (or government) imposed demand curtailment. If you watch DC owners’ and operators’ actions and behavior, you will discover that their growth plans – specifically rollout of AI products – are not negotiable.? You may be thinking “but I hear about data centers talking about clean power all the time, what gives?” Their actions tell a different story. Indeed there was a time when most data center developers prioritized sustainably powered facilities over most everything, but that time was before ChatGPT.?

The AI stakes have changed, and so has the willingness of DC operators to sacrifice speed and scale to run their facilities on renewable energy. This is more than nerds running rampant with a new toy; enterprise demand, commercial demand, national security concerns, and patriotism have converged to produce a tidal wave of demand. Futurists and intelligence operatives alike fear the ramifications of China getting the upper hand, and to them the race is on.? DC operators feel both the opportunity, and the pressure, to deliver fast. Like a gold miner who prefers nickel-plated pickaxes in normal times, but will take any pickaxe she can get when the cry of “Eureka!” echoes down from the mountaintop… sustainability is expendable if it gets in the way of “compute” power.?

"Tech companies really do want clean power. It isn't cheap talk. But if the clean power industry and transmission grid can't deliver at the necessary speed and scale, we aren't going to just stop building data centers. The compute is going to get built one way or another, with one source of power or another",” offered David Berry, CEO of Cloverleaf Infrastructure, a company that aims to bridge between DCs and clean power by teeing up compute-ready sites with a clean power strategy in motion.?

We shouldn’t be too hard on the data center sector, though. Taken as a whole, I think this community strongly prefers to do the right thing.? Many of the “hell-or-high-water” developers would prioritize green DCs if they saw a path to securing enough clean power. Power developers need to seize the opportunity to help show that path.

That begs the question: What should the power sourcing goals be for a new DC?? I believe some of the de-prioritization we’re seeing is based on the misguided view that it's all or nothing:?you either i) get your gold star tree-hugger stamp-of-approval for being powered by 100% renewable energy, or ii) you don’t, and may as well torch some coal because no one will give you “credit” for a “mostly” or “partially” clean-powered facility.? We need to correct this perception and find a way for society to reward data center operators for being mostly clean powered, even if not entirely so.?

Things are happening very fast right now, and many of the actions/steps taken today have very long-term ramifications.? Load “signals” enable certain behavior for utilities, ISOs, and public utility commissions, and many of the corresponding actions can’t be easily reversed. You can’t unbuild a natural gas plant – once it’s built, you better believe the powers that be will ensure it’s used enough to get a good return (for the owner, rate payers, etc.).? Those load signals are flying around the market right now, broadcast by huge firms who believe they are in the race of their lives.?

We’re not, of course, in the race of anyone’s life; we’re in the middle innings of an AI bubble –valuations, venture capital funding, businesses prematurely shoehorning in AI use cases, media headlines… even articles, like this one, covering the intersection of data centers and the power markets. The situation seems a textbook case of “Amara’s Law”: We tend?to overestimate the effect of a technology in the short run and underestimate the effect in the long run. What we’re seeing now, in the immediate-term, is irrational exuberance. To illustrate this, it’s useful to think about where we are in the “hype-cycle”. A team from Energy + Environmental Economics (E3) wrote a compelling paper on this[4]. Their cornerstone exhibit:

Yet there is money to be made while the bubble is still expanding, and some extremely powerful and competent people who want to cash in understand that better than any of us.? Humans in such circumstances are selfish, rash, sloppy, and over-confident.? They do things like convince an RTO in ERCOT, or Georgia Power[5], to build a new natural gas plant to serve insatiable theoretical DC load.? Sober minds can sense that doing so is irrational and not in the public good. Alas, sober minds do not drive behavior during a bubble.?

Not everyone will agree we’re in a bubble, but I bet we can all agree that expectations for DC hyper-growth have huge ramifications for developers of clean power. The main ones:

1. Overall demand expansion and upward pricing pressure form load growth, generally
2. The “shape” of that demand and what that means for technology mix
3. The geographical footprint of new load
4. The energy offtaker universe and financeability thereof

The rest of this piece is focused on how those ramifications should influence energy and storage development strategies. And how it should not. ?If you care to dive deeper into the foundational phenomenon of DC load growth and what it may mean, the “Further Reading” section at the very bottom offers suggestions.

The Timing Mis-match Problem

The problem isn’t load growth, it’s how we rise to meet it.? Most load growth is correlated to progress – electrification, technological advancements attached to social goods, and onshoring of manufacturing.? The problem comes if we address the load increase by delaying carbon-emitter retirements, or overbuilding new thermal facilities.?

As covered in the introduction, the DC sector is not going to sacrifice their expansion plans while waiting for clean power solutions.? And, as many of us know all too well, the timelines to develop operating power plants is ever-lengthening.

Acknowledging that there are plenty of exceptions, it seems the general “Pyramid of Needs” for the DC community as a whole:

As with all things climate change related, the problem with the “I’ll do the right thing later” mindset is the Time-Value-of-Carbon . Emissions avoided now have a disproportionately positive impact as compared to emissions avoided/captured later. This effect is especially amplified when one considers DC power demand signals through the lens of the evolving energy mix. The medium-term trajectory of renewable energy and energy storage capacity additions can definitely support the projected load growth over the same period - globally, domestically and at the ISO level (at the locational level there are exceptions, but those locations should be avoided by DC developers if we’re doing this right).

But if you zoom in to the next couple of years and simulate particularly aggressive DC build out, that demand shock is harder to absorb with mostly clean power. Many believe the resulting gap would be largely filled by carbon-emitting thermal capacity. ?

The first illustrative graph sketches that scenario. Though load flattens a little as the hype-cycle calms, the “damage” (more green-lit emitting power facilities) is done. The second graph depicts a more deliberate pace of DC expansion; one more easily powered by clean energy. ?

There’s no doubt every scenario requires thermal resources, at least in the near term, because i) supply and demand must meet locally and ii) managing the intermittency of renewable generation takes time; but we’re trying to minimize the climate impacts of DC load growth, not achieve perfection.

But I’m not so sure the popular narrative – that it’s only possible to meet near term load growth with new natural gas plants – is quite right. At times it seems people forget that gas projects have to go through the interconnection queue too. Moreover, it’s not as though clean energy (and storage) plants meant to meet DC load growth are starting at the back of the queue.? Ric McConnell of GridLab: “Yes, there are record numbers of GW in queues, and the time to get through the queues have increased. But utilities have a headstart; there are 300 GW of projects with executed interconnection agreements ready to construct today.”[6]? That compares rather favorably to the 13 GW of natural gas plants at the same state. Intermittency and the hourly production profiles of solar/wind are the challenge, not capacity.

Intermittency is a big challenge, indeed; hence the recommendation we be realistic and accept that we do need thermal in the near term. But the more hastily we lurch forward, the bigger the challenge and the greater the lasting damage to the domestic energy mix.

Load Signaling and the Utility Response

To appreciate that challenge, let’s unpack the development processes for new load, and new power facilities, with a focus on grid integration and power markets. These processes differ materially depending on location and facility type/size (I’ll use ERCOT as the guinea pig), but the meat of things is the same no matter where and what is being developed. ??

The process by which new DC load, and the corresponding transmission and generation upgrades, come to exist is an iterative dance – a fugue, of sorts - between i) the DC developer, ii) the ISO, iii) the transmission owner/operator (“TSP”), and iv) power facilities’ owners/operators.

As with all infrastructure development, the development path for i) data centers and ii) related power facilities is a minefield of circular dependency.? This manifests as DC developers’ reticence to fully commit to a specific facility/location until the power plan is clear and executable, and power developers’ reticence to commit to a facility/location until their offtaker and price are established.? Each stakeholder wants the other to “go first.” The infographic above illustrates how these highly interdependent development cycles progress. ?Walking through from left to right:

1. Picking the spots

Data Center and Power Developers: Criteria for selecting facility locations are largely the same. Both consider real estate availability/costs, permitting ease, proximity to end users, prevailing power costs (DC wants low, power wants high), labor costs, disaster risk, interconnection complexity/costs.? Unique considerations include network connectivity for DCs, and irradiance/wind resource and REC values for power. ?

Parties are often “siting blind,” so to speak, in that in the early stages potential load/power additions are mostly speculative.? The most substantive action either stakeholder can take is to perform power-flow and load-flow analyses which stress test potential scenarios to see how the DC or power plant would fare if certain variables on the opposite side (supply/demand) changed.

Moreover, both stakeholders’ siting and development is excruciatingly bespoke from one spot on the map to the next. Dillon Stambler, a Partner at CRS[7]- a top-shelf consultant to DC developers, power developers, TSPs, and utilities – described it this way: “Nodal characteristics might all be different only 20 miles apart, and little things like using the wrong terminology for a specific region can result in critical parties discounting projects from the outset. Technical folks on the grid and utility side want to help, but they can't digest everything. Data center developers are doing their best to find local knowledge and networks that can help spur collaboration with folks on the grid/utility side by demonstrating sophistication."

2.?Power requirements, IX queue entry

Data Center Developer: Once a site is selected and the power requirements determined, the developer submits a formal load interconnection request to the local TSP or utility. The DC is now “on the map” and “in the system” for the supply side to see, if they look hard enough.

Power Developer: Once a site and point of interconnection (“POI”) is selected, the developer applies for entry into the power resource interconnection queue. The project has a “queue position” and is “in the system” on the power side.

TSP: This is, arguably, the trickiest phase to navigate for the TSP/utility/RTO. They must sift through all these supply and demand signals to somehow arrive at a set of actionable grid upgrades (or hypotheses thereof). The variables are perpetually shifting, which might understandably paralyze the TSP.? As the TSP sends cost, timeline, and complexity signals back to stakeholders on both supply and demand side, the “answers” they provide might result in DC or power developer actions that materially change the inputs underpinning the very answers just provided. For example, if a large BESS project receives a feasibility with an estimated 5-year upgrade timeline and estimated costs of $0.50/w, the BESS project may withdraw from the queue. That, in turn may trigger a significant increase to the line upgrade costs borne by a would-be DC customer, causing the DC developer to scrap their site…and so on.

Importantly, being “in the system” does not mean “definitely going to happen”.? Much of the noise behind dramatic load growth projections arises from this fallacy.? There’s considerable daylight between i) entering the interconnection queue, and ii) an operational data center. Developing infrastructure is very hard.? This same fallacy applies to power developers (see here for more on that ).

3.?Interconnection Process and Studies

Data Center Developer and Power Developer: The path through “studies” begins. Most ISOs have 3 studies: “Feasability,” “System Impact Study,” and “Facilities Study.” They all cover similar ground, but with increasing granularity and accuracy. Unfavorable feasibility study results might cause developers on either side to abandon a site. ?Back to the drawing board.

TSP: The “whack-a-mole” effort described above continues in a more formal context. As the studies progress the probability of project attrition decreases, but their task remains so complex that it warranted ~100 pages in the FERC Interconnection Reform Order.[8]

4. Early power negotiations

Data Center Developer and Power Developer: As the studies begin, the DC side and the power side enter the match-making and price discovery process.? Both sides are testing their hypotheses about power prices, offtaker creditworthiness, contract terms (length). Here we have another circular dependency problem.? The assets being developed still carry fatal flaw (attrition) risk, and no one wants to incur significant liabilities for an asset that may yet die.? As further elucidated below, I believe there’s an opportunity to “right-size” parties’ commitments to each other (appropriately sized security deposits) in partially binding term sheets, such that i) each party can stomach putting $X at risk, yet ii) $X is also sufficient to engender the confidence each party needs to take the next big step. Failure to gain that confidence or unfavorable price discovery findings might lead to project/site abandonment (either side).

5.?Final IX studies and financial security deposits

The final studies’ codify the necessary upgrades (e.g., new substations, transmission lines, etc.) which are transmuted into an “interconnection agreement” on both the load and the power sides. Depending on the ISO, the upgrades might be paid for by the DC developer, the power developer, the TSP, or some combination thereof.? Sometimes this is prescribed by ISO tariffs (or public utility commission regulations), in other instances it is bespoke and even negotiable.? In Texas, for example, the RTO/TSP eventually pays for and owns the upgrades. However, the load or power customer must take the risk of the upgrade costs by posting financial security equal to the estimated total upgrade cost, and stands to lose whatever portion is incurred by the TSP, should the developer cancel the project. ERCOT decides how that responsibility/risk is allocated between load and power stakeholders, but ERCOT and/or the TSP will usually engage developers (and other stakeholders) in a (hopefully) constructive dialogue.

Data Center Developers: The financial security deposits are due whenever the customer wants the upgrade work to start, usually concurrent with execution of the interconnection agreement.? At this point, the DC developer is very “pregnant”, so they will not take this step until their DC has entered “when, not if” territory.? Having a very high degree of confidence in power price/mix/shape at this stage is paramount. Moving beyond this threshold without concrete agreements representing a substantial percentage of expected power needs is not recommended. To be clear, promising conversations and non-binding LOIs – though better than nothing – lack the appropriate “teeth” for this leap forward in risk/liability.?

Power Developer: The power developer’s posture is directionally similar, although in many markets the depth of market demand, and/or the ability to sell “merchant” electricity, offers enough optionality and liquidity to get power developers comfortable with posting financial security deposits without executed offtake agreements. Such markets (e.g. ERCOT, PJM) offer muted circular dependency, putting power developers in a better position to move through the development stages independent of specific term sheets or binding contracts with DCs.?

A brief note on utilities, ISOs, and “the Egg that Can’t Be Uncooked”: ?this “moment” in the process – where a new DC signs a new large load interconnection agreement - serves to lock in load expectations which are, in turn, used by: ISOs, to assess grid reliability, reserve planning, and market mechanisms associated with ensuring electricity demand is served; The utilities or other owners of power facilities, to determine if it is prudent to construct new facilities or delay the retirement of existing facilities; and The regulating authorities (utility commissions, primarily), to instruct approval decisions related to new facilities or changes to retirement timelines for existing facilities.

So it may be moot if a portion of added load is later cancelled, downsized, or materially delayed. From GridLab’s great article on this topic: “GridLab is concerned that the noise around load growth is driving a false sense of urgency, and utilities are using this to push for new gas to meet this rising load and in the worst case, delay coal retirements .”? Overeagerness in the near-term, therefore, can’t be harmlessly corrected for later. That IX agreement will often serve as, in effect, a “license to build” for parties who may prefer to be adding thermal resources to their fleet of power facilities, but are held at bay by regulatory hurdles. If you’re serious about expanding your DC footprint sustainably, wield your load signals with extreme caution.

6.??Regulatory approvals, permitting, contract finalization

TSP, ERCOT, and Local Authorities: The necessary regulatory approvals and permits are obtained for the upgrades. This may involve state and local regulators, environmental assessments, and public hearings.

Data Center Developer and Power Developer: The parties will usually be finalizing the power contracts during this stage, while also finishing up their site development work (zoning, title curative, environmental, etc). ?As construction approaches, each party will enter an EPC contract for the construction of the facility, and most developers will raise project financing.

7.??Construction

The TSP is performing the upgrade work, and the DC and any new power facilities are under construction. These parties would usually look to calibrate their construction schedules such that they reach completion around the same time.?

8. Synchronization and Energization

TSP: Commissioning all new transmission and distribution facilities.

Power Developer:? Commissioning and energization of the facility, in coordination with the TSP.

Data Center Developer: Preparing to receive power and begin operations.

In reality, the process constantly lurches forward and backward, and opportunities to risk-gate capital outlays with risk mitigation are few. Any experience remotely as linear as what I described is but a developer’s dream.? And, perhaps most importantly, supply never meets demand on a 1:1 ratio, but rather a portfolio of power resources and contracts are stitched together to meet demand, or vice versa.? In any case, the key milestones and decision points highlighted will hold, even if the sequence crumbles underneath entropy.

What Data Center Developers Should and Shouldn’t Do:

Should (Expanding On Select Points Above):

• First, do no harm.? Commit to something like this line in the sand: “We will not let our efforts to expand DC capacity i) cause the construction, or ii) materially delay the retirement, of carbon-emitting baseload power facilities.” (Natural gas peakers and thermal backup generation are currently unavoidable; perhaps counterintuitively, I think accepting this is the path to minimizing CO2 levels). To accomplish this, DC developers need to be more reserved in their load signaling, and the power sector needs to get better at filtering i) substantive DC development actions, from ii) shareholder call fluff, vacuous press releases, etc.. ?
• Treat the electrical grid as your resourceful ally.? It’s our nation’s most impressive infrastructure, and it will get you the clean power you want if you engage with it properly and respect the challenges utilities face. David Berry of Cloverleaf: "Utilities are not the enemy. They are between a rock in a hard place. They have so many stakeholders. They have a signal of growing load, but lots of noise. We have to make them partners, not just service providers."
• Come up with a power plan before engaging with the TSP to codify that plan, using expert support as needed.? Chisel away at your DC designs, procurement, and site development work so that when you do signal load additions, you’re signaling a vetted project, not an aspirational hyperscale flex. ???
• Be willing to enter term sheets with would-be power facilities before all your ducks are in a row. Investing in this extra step bears fruit (more on this below).? This supports the development efforts on the supply side, resulting in effective partnerships founded on trust and risk-sharing.
• Structure term sheets to have mutual at-risk security deposit postings. The deposits don’t need to be massive, but large enough to sting a bit if lost.? It often makes sense to start with mirrored, small deposits that ratchet up as time passes and commitment thresholds are breached.

Should NOT (Expanding on Select Points Above):

• Don’t worship the false prophet of technological shortcuts. America’s cauldron of innovation will continue searching for helpful and profitable technological advances, and some will be found and implemented.? But don’t confuse a worthy idea for a deployable solution and hitch your wagon to an unproven company’s unproven product rollout.? The technology you require already exists in the form of cheap, proven technologies, even beyond wind and solar. David Berry again: "We have existing, proven technologies that can improve the situation- battery storage, grid enhancing technologies, flexible backup generation with lower-carbon fuels. All these technologies are underutilized and under-implemented. We develop projects that maximize use of existing technologies to unlock more power for large loads." ?That’s exciting - embrace it.? Large infrastructure moves slow, so the temptation to try and “hack” the problem with miracles-in-a-container or elaborate microgrids is understandable. But emergent technology is not on your schedule either. Most are a Webvan-esque[9] amount of invested capital (and probably a decade) away from being cost competitive (if ever). Microgrids are alluring as they make a great photo and press release, but are extremely difficult to execute upon. The only large scale microgrids “in the field” today are military applications, and you don’t want to know how much those cost. Have a quick look, if you must, but then get on with the work of procuring power from grid-connected facilities using proven, cheap technology.
• Don’t expect to be the “belle of the ball” with everyone courting you and your theoretical future power demand…no matter how fancy the name attached to it. The leverage is on the supply side of the market. That balance shifts when new DC load becomes “bankable” – i.e. there’s a believable schedule, a creditworthy counterparty, and willingness to make real commitments.? Negotiating leverage comes with execution and the conviction demonstrated by taking risk (spending money, posting deposits, incurring liability in contracts, doing the hard work of development, knowing your corner of the grid).
• Don’t rely too heavily on grand, programmatic partnerships wherein you anticipate one independent power producer to solve all your clean power needs across many states, many ISOs, many sets of circumstances.? Even the more ubiquitous IPPs – Nextera, Constellation, etc. – would fall short of satisfying your demand after their available power is soaked up, because the reality is that every node in the country is a unique and special snowflake, as are a DC facility's hourly power needs. So the odds that an IPP has all the requisite "puzzle pieces" at the right spot are very low. Efforts to get more scale from fewer relationships are best directed toward groups like Cloverleaf who play an integrator role by developing DC-ready “powered shells”.? This is as close to plug-and-play as you will find.? Alternatively, regional experts can play a similar role. "CRS can help stitch together hourly-matched procurement from a mix of technologies, and get the TSP or utility engaged in that process constructively so ships don't pass in the night," says Dillon Stambler.
• Get caught up in the “noise” of what “everyone else” is doing.? While they are out producing the press releases that are giving you FOMO, you’ll have your head down doing the hard work to get actual DCs – clean ones – in the ground. Never has the “Tortoise and the Hare” fable been more apropos.

What Power (RE, BESS) Developers Should and Shouldn’t Do:

Should (Expanding On Select Points Above):

• Follow electricity demand patterns as you always have - there’s nothing especially new about this electricity end user trend.? Be cognizant of where it makes sense to build or expand DCs (see section below) and factor this positive attribute into market strategy together with other considerations.
• Partner with sincere DC developers that are ready to commit to your project with capital-at-risk, substantive development work, and exclusivity. A good developer seizes this rising opportunity but is cautious of looky-loos or relying too heavily on one specific execution before that execution is de-risked.
• When in doubt, include storage in the energy mix if you’re targeting a DC offtaker. DCs value firm capacity and particular hourly profiles, both of which are more easily achieved with batteries in the mix. The main reason to not include batteries is, rather simply, if you don’t expect the added revenue the batteries enable to outweigh the costs of including them. As noted, DCs are not especially price sensitive, so odds are you will find yourself compensated for the added EPC cost.
• Be careful about creditworthiness of the entity in the power contract(s). Developers often mistake i) a conversation with, say, Google/Microsoft/Facebook/OpenAI for ii) a PPA with their primary “topco” as the legal counterparty. Establish the entity and/or the guarantors on the power buyer side early in the conversation. This is a commercial matter that should be negotiated, not a legal matter to leave until contract finalization – a shockingly common error.? Any large company is certainly developing their DCs in subsidiary vehicles, and – if they’re smart – will prefer and attempt to contain the liability of long-term contracts lower in the organization. Don’t get played here.
• Be thoughtful about the settlement point and get this on the table early.?
• Develop wind projects in solar-heavy markets. As a solar guy, it pains me to say it, but in many markets we’ve overcorrected toward solar during the last 15 years.? The hourly profile of high-capacity wind drastically reduces how much BESS is required to deliver the power shape that DCs need. Experienced wind developers who have moved on from certain markets might consider taking another look at some of their old stomping grounds (and new ones too).
• Explore BTM capacity if they have space for it and it seems important to them, but do the work to ensure this is realistic.? Can they sublease the acres and/or license the roof? Can you get clean title with subsurface endorsements in the title policy where applicable? Would on-site installations require a different zoning designation? Are there environmental considerations? Is it buildable? ?Are you prepared to enter into an agreement that is contingent on the on-site portion? If not, figure out if this kills the deal early in the discussions.?

Should NOT (Expanding On Select Points Above):

• Don’t become a data center developer. In the renewable energy and storage space, we tend (as a group) to wander outside our sweet spots. It comes from a good place – a sense of urgency combined with courage (usually not arrogance) – but doesn’t always serve us well. If a DC developer that had never developed power told you they were developing a PV + storage project, you would chuckle, right? Well…
• Don’t pursue any project/location that is 100% contingent upon an unrelated party successfully developing a DC that you have nothing to do with.
• Don’t obsess over design and sizing decisions prematurely, just because a would-be DC is signaling interest and would love your free consulting services.
• Don’t give away your SRECs for free because you waited too long to get this on the table, and find yourself very pregnant with the contract when a PPA redline comes back with a “…electricity and all environmental attributes” slipped in.? You missed the window!
• Don't give away PR-related attributes, such as project naming rights, for free. Many of the companies developing DCs care deeply about public relations. Use this as a negotiation chip, not an afterthought.?
• Don't allow yourself to be blind to the load-side interconnection process. You have every right to have visibility into the IX process – costs, work being performed, timeline, failure points. Don’t make all these things your responsibility, but recognize that they might be your risks.

?

Term Sheets and Deposit Agreements

I don’t purport to be breaking any new ground here, but am surprised at how infrequently parties enter into pre-PPA binding agreements that establish bi-lateral commitment.? Development is all about eliminating variables and mitigating risks.? This is true on both the demand and supply side of this DC power market.? Commonly a PPA won’t be executed until deep in the development process – partly because key variables (e.g., the COD date, etc.) are unsettled, and partly because negotiating a PPA is hard and expensive.? We can partially mitigate this risk, however. We have the technology. ??

My recommendation is to pursue a term sheet and “Deposit Agreement” in the middle stages of development. In this coupling, the term sheet is just the standard summary of key terms the parties expect in the final PPA. Most of it is non-binding, but it’s detailed, and some of its details are linked to binding terms in the Deposit Agreement, which addresses the way the buyer and seller of electricity/capacity/SRECs make commitments and take risk together.? It compels each party to post financial security that is forfeitable under certain conditions, such as the would-be buyer getting a better offer for power and abandoning the agreement contemplated in the term sheet, or the seller failing to place the project in service. This interim step does a lot of work for both parties:

• Financial wherewithal – each party can see that the counterparty is financially sound, and institutionally functional enough to cause a meaningful amount of money to go out the door.
• Putting Money Where the Mouths Are – early discussions for buying and selling power are rife with bluster.? At some point it becomes important to find out what the other party will actually stand behind.? A right-sized deposit forces out true expectations.
• “Vows” – quite simply, there’s something very powerful in two parties agreeing to count on, and deliver for, the other. Motivation and clarity of purpose produce a steady tailwind.
• Liquidated damages – a DA establishes a payout to a party that gets left at the altar, which helps each side manage their risk, even if the amount isn’t a full “make-whole.”

The signing deposit can start small – maybe $10k-20k/MW or thereabouts – and grow as certain milestones are reached.? The milestones should be thresholds where the risk on one or both sides ratchets up.? Posting of IX upgrade deposits, GIA signing, major equipment purchases, and construction starts are all logical milestones.? A deposit schedule might look something like this:

The deposits should mirror each other.? Importantly, forms of financial security should include cash, surety bond, or letter of credit (from an investment grade party in the case of bond/LC). At COD this security deposit falls away, possibly replaced by operating phase securities, if/as applicable. The circumstances under which the deposit is forfeitable is a negotiated point, and situation dependent, but definitely includes “termination for convenience”, and probably any event of project death (on either side) excluding force majeure. ?That may sound daunting, but at some point each party has to bet on themselves and commit to the project. With the sliding amount, the “size” of that bet can be right sized for the circumstances.? Finally, the Deposit Agreement might include liquidated damages in connection with non-binary timing/date delays.? Not only is each party betting on the other side coming through, but they are betting on them coming through by a specific date, and bear costs if the counterparty is late.?

If all that sounds like a hassle, consider the hassle we all deal with when our project development plans are squashed by another party flaking. Those dead deals can destroy entire companies. Doing this agreement will almost certainly be worth it.?

?

Where Is the Play?

Dozens of organizations have produced data-based analysis on the topic. If you were to read just one, GridStrategies is the best deeper read (the full report not the summary slides). Since you’ve trusted my own take this far, I’ll share my main takeaways after a quick tour through my favorite graphics.

GridStrategies' ISO Map

This shop seems to have done the most real work on the sub-markets question. The infographic below is not very granular, but it’s a good place to start – existing and new DC MW by ISO.?

Wall Street Journal, Dept. of Homeland Security

This is a “second derivative graph”, meaning it’s not showing expected growth, but rather the growth of the expected growth.? It aims to show which ISOs have become hot DC targets recently as measured by changes to forecasted growth.

S&P Submarkets Map

This one gets more granular on the geographical submarkets and differentiates between i) existing, and ii) planned DCs.

JLL Report

This recently updated real estate-oriented piece gets into sub-geography. The second image zooms in on activity in the last 4.5 years.

My takeaways:

• DC load is emerging nearly everywhere. ?Given long power development cycles, by the time de novo development projects reach the power marketing stage there will probably be a DC development in the same electrical neighborhood. It’s ok to let this development theme “come to you.”
• Diversification is important, but focus is usually more so - even if you have high conviction in a dozen different DC-rich markets doesn’t mean you should pursue them all.? Just because you can, doesn’t mean you should. Develop in markets you know.
• Historical DC hubs such as Northern Virginia and Silicon Valley are a bit “picked over” and expected to be less popular locations on a go forward and relative basis. Yet it seems every quarter Northern VA is expected to fade, and every quarter it tops the charts of new capacity added. ?Probably not markets to chase anew, but if you’re already there, the game is not up.
• Texas will continue to attract DCs – especially in the Dallas, Austin, and San Antonio metro areas – although reliability issues, stringent AI privacy laws, and unfavorable supply/demand characteristics put it in the “somewhat overrated” category for me.? Many developers (on both sides) confuse ERCOTs transparency and “approachability” for ease.? Nevertheless, it is the malleable “wet clay” of power markets and, rightly or wrongly, will be a major DC host. Power developers simply must land busbar PPAs to make this DC market worth the risks from the power perspective, as the marginal value (busbar vs. hub settled) in ERCOT is unmatched, and basis risk is an ever-irksome albatross. DCs can stomach some basis volatility in their OPEX.
• With its cheaper land, low labor costs, and central location, the upper Midwest has long been a desirable DC host. That seems poised to continue, with the Columbus area, Iowa, Wisconsin, and the Twin Cities experiencing the most new demand. The MISO queue is comically long, so if there were any market in which to seriously explore BTM, it might be MISO, especially given the prospects for getting DC-adjacent land are better than most geographies.
• The Chicago area has proven a very tricky DC development sandbox historically, so lucrative rewards await those who navigate the challenges.? Northern Illinois is crowded with power developers due to the successful community solar regime, which may understandably cause one’s eye to wander. Southern Wisconsin has emerged as an interesting pocket, for example.?
• Salt Lake City is all the rage right now, and one of the poster children for the new DC load wave. This strikes me as a very actionable market from a power development perspective.
• Atlanta is probably the second poster child. The jump in construction activity in the Atlanta region is only matched by SLC.? Of course, here – and in Alabama, another spot with considerable activity – one must contend with Southern Power and the particularly imposing monopoly they enjoy.
• Phoenix has a lot of DC capacity and stands to keep growing, despite the unfathomably bad ?cooling costs (in every imaginable sense of the word). The growth here is centered on Phoenix, but radiating increasingly outward. Importantly, few DCs are well-positioned to receive power from the north – in Navajo land or otherwise. Despite the creativity and boldness the Navajo Tribal Unity Authority has shown, developments in/near first nation’s land will have a high attrition rate. Google, Amazon, and Microsoft are all here, which offers useful competitive pressure (not always the case).
• The pacific Northwest is a major market due to high demand around Hillsboro, Seattle, and Central WA, but geographical/topographical constraints severely limit the universe of deliverable locations.? Much like Chicago, I consider it a high degree-of-difficulty market to pursue, but one with a good return if creativity, or just-plain-good development prevails.

You might be wondering where DC development prospects are unfavorable:

• New England – expensive land, constrained grids, expensive power.
• Florida – natural disaster risk, FPL doing FPL things.
• Most of California – obviously the demand is extreme in parts of California, but electricity and land are outrageously expensive, and earthquake risk renders much of California challenging.
• New York - strict environmental regulations and zoning process. High compliance costs. New Jersey is emerging as a better alternative, though it has its own problems.

Conclusion

Data center load is obviously a huge opportunity for the power sector, though somewhat “booby-trapped” with fallacies. Renewable energy and storage developers are poised to benefit massively from the load growth expected from DCs.? They should, however, be wary of the hype-cycle and the possible trap of overreacting to a “shiny object”.? DC load is load much like any other. While it’s unwise to select development markets solely to take advantage of the data center wave, it seems reasonable to prioritize certain markets like WECC (the Northwest, Arizona, Utah), parts of Texas, MISO, and Northern SPP because of the opportunities afforded by data center load growth.? Developers should take great care to leg-into risk and insist that risk be taken by the DC developers too. If the risks are respected, understood, mitigated where possible, and navigated with an appreciation for the circular dependency challenges, many profitable projects - and avoided carbon emissions from thermal plants – await.

Further Reading:

EPRI - Powering Intelligence: Analyzing Artificial Intelligence and Data Center Energy Consumption. The most cited and data-rich article.

Energy and Environmental Economics - Load Growth Is Here to Stay, but Are Data Centers?? The aforementioned masterpiece. It also contains useful analysis on marginal economics/timing/difficulty that gets at diminishing marginal returns: it’s about a third as “hard” to develop a DS that is 80% powered by clean energy compared to 100% powered by clean energy.?

Brattle Group – Electricity Demand Growth and Forecasting In a Time of Change. They did some real work here, with the most substantive bits being an analysis of how new load signals are being incorporated into forecasting on a utility-by-utility basis.?

Ric O’Connell of Gridlab - What’s the Rush With Load Growth? I’d planned to go far deeper into the fallacies and traps of overreacting to theoretical load growth, until I realized Ric already wrote a fantastic short article that says it better than I could.

Energy Innovation Policy and Research – Let’s Stop Worrying About Load Growth and Get Serious About Solutions. A longer version of the overreaction trap, with a focus on demand side solutions. The best explanation of why near term supply side solutions cost us dearly for years to come.

GridStrategies - National Load Growth Report – The most useful data-scraping oriented on relative planned load growth. This has the most information on which US markets are likely to see the most new load.

Footnotes

[1] I use “power” instead of “energy” or “electricity” throughout given the outsized role of BESS and various backup technologies

[2] Hyperscalers provide cloud computing and data management services to organizations that require vast infrastructure for large-scale data processing and storage, e.g. AWS, Google Cloud, Microsoft Azure, etc.

[3] https://www.wsj.com/business/energy-oil/electricity-demand-coal-gas-retirement-charts-dd07029a

[4] https://www.ethree.com/wp-content/uploads/2024/07/E3-White-Paper-2024-Load-Growth-Is-Here-to-Stay-but-Are-Data-Centers.pdf

[5] https://www.spglobal.com/commodityinsights/en/market-insights/latest-news/electric-power/071724-experts-see-mismatch-between-us-utility-planning-cycles-and-data-center-builds

[6] https://www.dhirubhai.net/pulse/whats-rush-load-growth-ric-o-connell-ic0zc/

[7] CRS is especially knowledgeable about, and active in, the WECC markets

[8] https://www.ferc.gov/media/e-1-order-2023-rm22-14-000

[9] For the old folk. Webvan was a (now-infamous) Silicon Valley darling in the dotcom bubble that Sequoia, Benchmark backed with extremely high conviction…and capital intensity. It went violently bankrupt in 2001, and now serves as a cautionary tale.