As computation has exploded — whether for AI, Bitcoin, or general use — data center energy use is projected to double over just the next two years. In response, load shifting has emerged as a simple yet powerful strategy to unlock myriad benefits.

This focus on load flexibility has garnered more attention of late, from a New York Times investigative piece last year digging into whether Bitcoin mining operations truly modulate their load to soak up more renewables, to a recent Bloomberg article about the growing electricity consumption of the world’s data centers and their attempts to reduce the associated emissions and use more renewable energy through various forms of load shifting.

Load shifting can potentially drive many benefits, for example:

1. Load shifting away from times of extreme peak demand can alleviate strain on the grid, supporting greater reliability, reducing the risk of blackouts, and potentially lowering costs.
2. Similarly, load shifting away from times of dirtier electricity, such as when a more-polluting fossil peaker plant is the responding generator, can lower overall grid emissions.
3. Load shifting toward times of excess wind or solar generation that’s being curtailed (AKA thrown away), can both reduce emissions and also boost renewable energy’s grid integration. 

But whenever you see a story about load shifting, the key question is, which times is the organization’s electricity use shifting to and from? Or, in a question so critical we named our whole nonprofit after it: “Watt” time is the load being shifted to?

The promise (and pitfalls) of load shifting

As we just noted, load shifting is often touted for the beneficial things it can do. But load shifting is only good if it does do those things. And it only does those things if it shifts the load to the times that are best for a specific objective.

In fact, experts have long known — even since the late 2000s from research like this 2008 study — that many cases of load shifting that people thought helped the environment actually increased emissions, not decreased them. Then further research found it happening again and again. Why? Because whether load shifting helps or hurts any particular goal depends totally on what times you are shifting load to and from. 

This is because load shifting isn’t necessarily good or bad. It is simply a technique that can be leveraged toward various ends, to varying degrees of success (or not). Energy journalist David Roberts summed this up well in a 2019 article for Vox. His article was focused on battery energy storage, but the perspective applies equally to load shifting overall:

“It’s a mistake to deploy batteries, or energy storage in general, as though they will inevitably reduce emissions. They might or might not. Indeed, it’s probably a mistake to think of them as emissions-reducing technologies at all. Rather, it’s better to think of storage as akin to transmission lines. Wires can carry both clean and dirty energy; their impact on emissions depends on local circumstances. Their primary purpose is not to reduce emissions, though, but to make the grid run more smoothly. They’re a grid tech, not a decarbonization tech. The same applies to batteries.”

For load shifting to reduce emissions, the software intelligence driving the load shifting needs to be optimized (or co-optimized) for doing that: reducing emissions. And you need to use the right signals to do so.

Load shifting based on marginal emissions and system-level impacts

Make no mistake: load shifting — by time, by location, or both — can indeed help sop up excess renewables and reduce grid emissions. But what does it really mean to do those things?

For many years, people tended to assume that shifting load to times when wholesale electricity prices were lower must reduce emissions. But all three studies linked earlier in this article showed that the opposite is often true. 

Then, for many years people assumed that shifting load to times of low average emissions rates — rather than low marginal emissions rates — must reduce emissions. Then study after study after study proved that that’s wrong, too.

With load shifting, more than a decade and a half of peer-reviewed studies has clearly established that what affects emissions and excess renewables are the marginal generator(s). Which power plant(s) respond by turning on or off, or ramping up or down, in response to changes in demand from load shifting? That’s how you appropriately measure the real impact on the grid system and its emissions.

If you’re perhaps thinking about load shifting data center computing to minimize emissions, you might think that shifting to a time and location where the sun is shining and solar PV is cranking out clean energy would help. But if that grid’s overall demand is already using all the solar that’s being generated, then adding new demand via your load shifting could cause a polluting fossil-fueled peaker plant to respond. Oops!

Load shifting to soak up surplus renewables that would otherwise be curtailed thus requires looking at the marginal emissions rate and the marginal generators. When and where are wind and solar on the margin? When and where are they being curtailed, such that shifted load could help absorb more of that clean electricity for zero increase in overall grid system emissions? That’s what affects the environmental aspects of load shifting.

We’re excited to see software practitioners increasingly thinking about the best times and locations for their software to consume electricity — and developing approaches to turn theory into practice. Called Carbon Awareness by the Green Software Foundation, software developers can build these capabilities into their operations.

It’s undoubtedly an exciting time. Software and computing are often the “brains” behind load shifting other technologies’ electricity use to reduce its associated emissions, from smart thermostats to EV charging. More than ever, practitioners are also looking at how computing itself can tap into these same load shifting opportunities.

As ever, we’re strong proponents of load shifting as an emissions-reduction solution with gigatons of potential at scale. To get there, we just have to do it right.

At WattTime, we’re excited to see an increasingly large number of organizations asking, “When we take an action on the grid — whether that’s building a renewable energy project, shifting load to different times, or adding new load — what are the ways that action affects real-world carbon emissions?”

One way to think about that question is to break it into two parts: 1) What are the short-run effects of that project on real-world carbon emissions in the near term? 2) Will the long-run effect be pretty similar to that short-term effect, or somehow systematically different on a longer time frame?

Both of these questions can be answered by examining the marginal emissions rates of power grids, albeit through two different lenses: short-run marginal vs. long-run marginal. There are reasons to believe they might be systematically different.

The energy transition is causing near-term operational and long-term structural changes to electricity generation & emissions

Amidst the energy transition, power grids are changing in diverse and profound ways.

Renewable energy projects built today will be in operation decades from now, when the world — and the grid’s supply-demand interactions — may work differently. Or, it could be that the effect of an action now — such as the proliferation of data centers, electric vehicles, and industrial and residential electrification — also drives structural change in how power grids evolve in response, causing effects that might not show up until a long time later.

For example, in the short term, introducing a large load like a new data center or hydrogen electrolyzer in a given grid region will cause marginal generators to immediately ramp up and meet that new load. But over the longer term, this and other durable new demand might also nudge the grid operator to eventually build additional generating capacity. If that new capacity is much cleaner (or dirtier) than the generators that respond in the near term, the long-term change in emissions could be lower (or higher) than the short-term effect.

The challenge of long-run marginal insights

Long-run marginal emissions are not a topic that WattTime has previously weighed in on, as it’s different from our primary expertise in one key way. At WattTime, everything we do is rooted in scientifically validated, empirical, data-driven approaches that can be easily verified. We spent a lot of time comparing the predictions of different models to what actually happened in the real world, and rejecting models that failed to correctly predict real-world behavior.

We’ve had good success doing that for short-run marginal emissions. But for long-term models, that’s hard to do. How do you verify the accuracy of a model that makes predictions about 20 years in the future… without waiting 20 years to find out if you were right? That’s why in the past we have stayed out of debates about long-run marginal emissions rates, and left it to others who were more comfortable making estimates that are harder to verify.

But just because something is hard to measure doesn’t mean it’s not important. Long run effects may be significant, they may be systematically different from short-run marginal emissions, and we applaud those arguing that it’s smart to consider long-run effects as well as short-run effects when trying to make decisions about how to best reduce emissions. 

The importance (and opportunity) of long-run marginal insights

Given the growing willingness of companies and governments to actually make different decisions based on what experts like us say would be most impactful, we think this topic is becoming more important than ever. So, we’re starting to explore the existing and emerging research in this area from many different experts, and try to ascertain what we as a society can know with confidence about long-run effects.

Examples of projects we’re looking into are: seeing whether models can at least predict changes 5 years out successfully; whether models can predict structural change that happens quickly, but then lasts a long time; or gauging whether models applied to data from 20 years ago can reasonably predict successfully what’s going on today (without “cheating” and being fed the answer indirectly). 

Our hypothesis going in is that long-run models will rarely predict the future exactly, but often may give clear, robust directional evidence that certain decisions are almost certainly more impactful than others. But that’s a hypothesis; we’ll know more once we actually study the evidence. 

This is a new area of research for us and we’re very conscious we don’t have all the answers. We also don’t want to reinvent the wheel if others have already solved some aspects of this problem. If you’re looking at these topics too, we would be thrilled to collaborate with you. We’re looking forward to collaborating with other researchers in this area!

Here at WattTime we’re more accustomed to looking forward, rather than backward, with a focus on further impact we can help to catalyze. But today is a special date in our history. It’s our 10th birthday! February 21, 2024 marks a decade to the day since our official incorporation in 2014. And so in this article we’re going to be unusually introspective, taking a look back at some of the pivotal milestones and accomplishments of these past 10 years — and what we’re most excited about in the years ahead.

1. Behavioral economics academic research around choice.

 In the early 2010s, many of the first eventual WattTimers were grad students at UC Berkeley. We were behavioral economists, software programmers, data scientists. And we all shared a fundamental intellectual curiosity: What happens on the power grid when you flip on a light switch?

It seemed crazy that we, as everyday consumers, did not know. It was equally infuriating that we had no power over whether our electricity use caused more or less pollution. Yet we turned that sort of righteous indignation into opportunity via hackathons to try and figure out the answer.

2. Officially born in 2014 as a mission-centric nonprofit… with a software tech startup DNA.

As initial hackathons progressed and we rolled up our proverbial sleeves further, we soon discovered — to our surprise — that everyone else had this righteous indignation about it, too. They wanted the opportunity to voluntarily go green, if only given the choice to do so. A/B consumer testing strongly confirmed this hypothesis. (Subsequent consumer sentiment and behavior research, such as with our partners at the Great Lakes Protection Fund, have further affirmed our initial findings.) All of which prompted us to found WattTime as a mission-driven nonprofit, even though the solutions taking shape would have a high-tech software aspect to them.

3. Pioneering the idea of AER, powered by v1 MOERs.

Those first hackathons eventually evolved and matured into our first flagship solution: Automated Emissions Reduction (AER). AER provides a signal for smart devices to schedule their electricity use for times when they will cause less emissions and pollution.

We began with direct-to-consumer ideas such as smart plugs. The first adoption by an external user was four golf carts at UC Merced. Then things started to snowball with major tech companies and automakers, spanning technologies such as smart thermostats, battery energy storage systems, EVs (and their charging), and beyond.

v1 of our marginal operating emissions rate (MOER) powered this capability. We upgraded to v3 MOERs in 2021, also now available in a new-and-improved v3 API, including expanding geographic coverage for power grids around the world.

4. Championing the importance of marginal emissions.

When we started out with AER, as academics we knew that the best way to measure the impact of interventions (i.e., academic speak for things like load shifting) was to use marginal emissions, such as our MOER signal. This built upon the established, peer-reviewed literature that came before us.

More recently, though, we have found ourselves in important industry discussions (and sometimes, heated debates) about using average vs. marginal emissions rates. We didn’t set out with any expectation of getting involved in such debates; it has simply come with the job description.

The commercial tides are now turning in favor of the long-established academic findings. The likes of Microsoft, TimberRock, Brainbox AI, and others building WattTime and other marginal emissions signals into their energy and carbon intelligence platforms. Now there’s also, VERACI-T, a cross-industry collaborative group validating marginal emissions datasets.

5. 2017–2018: WattTime’s “Oscars party” collective moment.

For any idea or solution, there’s a time when it starts to gain real traction and recognition in the market. For us, these years were that moment — both for WattTime as an organization and for individual members of our team.

Our co-founder and executive director Gavin McCormick was named a climate “fixer” in the 2017 edition of the Grist 50, an annual list of emerging green leaders and bold problem solvers. One year later in 2018, he was named a finalist to the Pritzker Emerging Environmental Genius Award at the UCLA Institute of the Environment & Sustainability, which focuses on “uncovering promising young innovators and boosting their careers as champions for the environment.”

That same year, ‘emissionality’ was recognized as a finalist in the 2018 Shorty Impact Awards and AER was recognized as a finalist in the Emerging Technology of the Year category of S&P Global Platts’ annual Global Energy Awards. 2018 became an even bigger year when AER was named a winner of the 2018 Keeling Curve Prize, an initiative that recognizes and rewards the most promising projects that effectively reduce greenhouse gas emissions or increase carbon uptake.

6. An emissions signal for battery energy storage.

A different level of credibility came into play when government agencies and programs began incorporating some of our emissions signal work.

In California, for example, battery energy storage systems under the Public Utility Commission’s Self-Generation Incentive Program (SGIP) were supposed to help the state’s grid reduce its carbon emissions. That wasn’t happening — until SGIP began using WattTime to develop their program signal, ensuring battery energy storage programs achieved their actual emissions-reduction goals.

Now other states and jurisdictions are exploring similar approaches, using more direct measurement of the target metric (e.g., marginal emissions), rather than proxy signals and assumptions (e.g., price or roundtrip BESS efficiency).

7. A shift toward Impact Accounting.

Carbon accounting standards — especially the GHG Protocol’s prevalent Scope 2 guidance around the indirect emissions associated with electricity use — have motivated sweeping clean energy investments from corporations and institutions worldwide.

But best practices evolve with the times. Which is why we’ve teamed up with companies such as REsurety and written joint position papers with organizations such as Electricity Maps. It’s why we cheer on our corporate partners at the Emissions First Partnership and why we’ve written our own insight brief on the idea of Impact Accounting.

These and other efforts all aim to help better align corporate actions with true real-world impact and authentic emissions reductions, and to combat a rise in greenwashing concerns and skepticism around hollow actions that don’t achieve their proclaimed benefits.

8. Expanding from climate to health damages. 

Although we started our work years ago focused primarily on carbon emissions, we also recognize the importance of mercury and other forms of power plant air pollution — including their impacts on human health and environmental justice. So after much hard work, we unveiled a new health damages signal, which ties electricity use (and its associated grid emissions) to human harm.

9. Surpassing 1 billion watts of emissionality. 

Toward the end of the previous decade, we popularized emissionality as a next evolution of and complement to additionality.

As a strategy for clean energy procurement, the idea behind emissionality is simple: Not all renewable energy is created equal. The avoided emissions of a new wind or solar farm can vary, by a lot, depending on where that project gets built and what power plants its generation displaces. The size of the prize is literally gigatons of avoided emissions opportunity on the table.

Boston University was one of the first organizations to adopt the strategy. Others soon followed: steelmaker Nucor, tech giant Salesforce, solar developer Clearloop, advisory Edison Energy, and others have also leaned into an emissionality strategy for their clean energy procurement.

Toward that end, last year we were thrilled to surpass 1 GW of renewables procured via this strategy. Less than 6 months later, we’re already closing in on the next gigawatts of wind and solar procured in part with emissionality in mind.

10. Co-founding Climate TRACE and incorporating satellite-based emissions monitoring.

In 2019 we announced a new project to measure emissions of the world’s power plants from space, launched with grant support from Google.org’s AI Impact Challenge and covered by the likes of Vox. By 2020, that initial effort had expanded in a big way into Climate TRACE, a global coalition of NGOs, tech companies, universities, and climate leaders including Al Gore using satellites and AI to measure human-caused GHG emissions from essentially all of the major sources on the planet.

Across the three years since then, Climate TRACE’s data have progressed by leaps and bounds, rapidly advancing from country-level annual data to facility-level data for 350+ million assets in the world’s most-comprehensive and granular such dataset, which we unveiled in December 2023 on the mainstage at COP28.

Along the way, Climate TRACE has been named to Fast Company’s “most innovative” list and TIME’s “100 best inventions.” We received the Sierra Club’s Earthcare Award and our executive director Gavin McCormick gave a TED talk on Climate TRACE that’s been viewed nearly 1.8 million times.

But it’s the use of the data for faster, deeper decarbonization that makes us most proud. From national, regional, and local governments to major companies such as Tesla, GM, Polestar, and Boeing. 

What’s next: scaling further impact together

Whew! It’s been a busy (and positively impactful) 10 years. But after today’s celebration of our official 10th birthday, that’ll be enough reminiscing in the rearview mirror. We’re far more excited and motivated about the work ahead of us, and the even greater impact we can achieve together. Won’t you join us?

Everyone knows you can’t manage what you don’t measure. Less often pointed out? You can’t manage what you measure incorrectly. 

Corporate net-zero targets are at an all-time high, per reporting from The Economist. In fact, fully 75% of the world’s largest corporate greenhouse gas emitters have set net-zero by 2050 (or sooner) targets, as of an October 2022 benchmarking analysis by Climate Action 100. This is good news.

Or… it should be. Of course, these targets will only genuinely decarbonize the atmosphere if they measure the real thing. And, unfortunately, that’s not always what happens.

From South Korea to Europe to the United States, corporations are under more scrutiny for potential greenwashing than at any other time in recent memory. 

At WattTime, we care about this not because we care about catching bad guys. In our experience, the vast majority of corporate emissions miscounting is a genuinely well-meaning mistake. But such scrutiny is also good news nonetheless. Why? Because it is forcing corporations to re-examine their sustainability efforts to better align with true impact that corresponds to real-world emissions reductions, not merely on-paper-only green claims.

And as companies allocate growing sustainability budgets, a heightened focus on actual impact empowers them to identify and pursue strategies that yield the highest real-world decarbonization return on investment (ROI) — and, reciprocally, to avoid strategies that cause a real-world increase in total global emissions.

Using the Right Math Matters

How companies measure the emissions they cause and which math they use to do so matters. A lot. That’s because, let’s face it, climate change is starting to claim lives. And the only thing that will save lives is impact — whether and how much a company’s actions genuinely cause total global emissions to go up, down, or stay the same.

Historically, much carbon accounting was done in terms of average emissions factors (AEFs). AEFs take the overall electricity generation mix for any given power grid, then apply it to a specific company’s load for their facilities. This was a fine solution in the early days, when carbon accounting didn’t actually do much, and most companies were not taking meaningful real-world actions based on these emissions factors. 

Times have changed. Today, companies are actually meeting GHG targets, optimizing their actions, and taking sustainability seriously. This is fantastic news, but it means that today, the connection between carbon accounting and reality actually matters.  

But there’s one big problem. AEFs are the wrong math for measuring impact, because they ignore the basic physics of how power grids operate — including how power grids respond to various influences. Using AEFs assumes that all generation sources on a power grid equally share in outcomes. They don’t. Nuclear power plants are not going to turn on and off in response to what one electricity user does. Neither will always-on baseload plants.

Moreover, simply making AEFs more granular, such as hourly, doesn’t solve the problem, either, because it still ignores fundamental power grid operations.

When a company chooses to site a new facility (and its electricity load) — a data center, a factory, a new corporate campus — in a particular region because that region has a “green” power grid… When a fleet of electric vehicles (EVs) uses smart charging to modulate when those EVs do and don’t charge… When smart thermostats and building energy management systems modulate the flexible portion of a commercial building’s electricity demand to shift load across hours… 

All of these and other examples don’t impact the entire generation mix. Most of the power grid’s generation stack merrily chugs along unaffected, blissfully unaware of these influences.

But the common corporate decarbonization strategies mentioned above do impact a specific subset of generators that respond to the corresponding increases or decreases in electricity demand. It’s precisely these generators — and their emissions — that matter for understanding impact.

They are known as marginal generators. Their associated emissions intensity is known as the marginal emissions factor (MEF). And their emissions are the marginal emissions: those emissions that specifically result from marginal units responding (e.g., turning on, ramping up) in order to meet the next incremental megawatt of electricity demand.

If a company chooses to site a new facility in a particular power grid, it’s the marginal units that must meet that demand — and therefore, the marginal emissions that best measure the impact of that load-siting decision. If a smart thermostat or EV charging software shifts the timing of power demand, it’s the marginal units that are impacted — and also therefore, the associated increase or decrease in marginal emissions that best measure the impact of that load shifting.

The temptation to use AEFs is understandable: they are widely available and the calculations are easy to run.  But this is a well-established area of research. Scientists and grid experts agree that AEFs do not accurately measure impact. The GHG Protocol is clear that one may not use AEFs to measure avoided emissions; rather, they specify use of MEFs for such Scope 2 calculations. The list goes on and on.

Widespread Agreement to Use MEFs for Impact Assessment

More than a decade of robust research and widespread agreement among scientists and grid experts support using MEFs as the right way to measure the environmental impact of electricity system interventions. For example:

• “We compare marginal and average emissions factors, finding that AEFs may grossly misestimate the avoided emissions resulting from an intervention… Both supply- and demand-side interventions will displace energy and emissions from conventional generators. MEFs give a consistent metric for assessing the avoided emissions resulting from such interventions.” — Stanford’s Azevedo et al. in Environmental Science & Technology (2012)
• “These [marginal emissions] estimates have important implications for understanding the environmental consequences of many electricity-shifting policies… [they] are also relevant for understanding the impacts of activities and policies that increase electricity demand.” — Yale’s Kotchen et al. in the Journal of Economic Behavior & Organization (2014)
• “If the load under analysis represents an incremental increase or decrease, marginal ERs can provide an emissions measurement specific to the change. Thus, for consequential LCAs of new loads, marginal factors are recommended.” — University of Michigan’s Ryan et al. in Environmental Science & Technology (2016)
• “...the use of AEFs, which reflect grid-average situations, to estimate the effect of an intervention may be problematic, because not all generating technologies would respond to changes in demand proportionally… AEFs could significantly misestimate the amount of emissions avoided by an intervention. By contrast, MEFs estimate the emission intensity of marginal power generation that responds to a change in demand, and are a more appropriate metric to assess emission implications of policy and technology interventions.” — University of Minnesota’s Smith et al. in Environmental Science & Technology (2017)
• “Increasing (or decreasing) demand at a given time affects the operation of only the marginal units in the dispatch order, and not the infra-marginal units. Therefore, emission implications of energy storage, or any other incremental demand- or supply-side resource that leads to load-shifting, depend on marginal operating emission rates, based on the emission intensity of the marginal units, and not the average operating emission rates, based on the average emission intensity of the entire grid.” — Columbia’s Shrader et al. in SSRN: Social Science Research Network (2020)
• “Marginal Emission Rate provides a mathematically sound and transparent way to quantify the carbon footprint of electricity consumption and production. Mathematically, the MER calculation is similar to the calculation of Locational Marginal Prices (LMP) and system lambda, which are used for economic dispatch of power systems across North America.” — Tabors Caramanis Rudkevich’s Hua He et al. in The Electricity Journal (2021)

Here at WattTime, we’re strong advocates for measuring whatever will affect real-world total emissions. In electricity, that means MEFs. (Within our datasets, they’re referred to as MOERs: marginal operating emissions rates. You can read more about our perspective in our 2022 insight brief about impact accounting.)

In the wake of the UN IPCC’s AR6 final synthesis report about the climate crisis — underscoring the need for rapid, deep decarbonization of the global economy — none of us can afford to base decisions, and impact assessments, on faulty math. We need to make authentic progress reducing global emissions. And for that, we need to use marginal emissions data to honestly and accurately reflect how power grids actually respond to the strategies we implement.