Ember analysis finds data center load shifting barely dents peak demand
Data center load shifting can cut Ireland's 2030 peak electricity demand by about 100 MW, or roughly 1.5%, according to an Ember analysis published on 3 November. Behind-the-meter generation and storage could deliver nearly five times more relief.
On 3 November, pv magazine reported on an Ember analysis. It puts a number on something the grid software industry has been selling hard: how much demand flexibility data centers can actually provide. The answer, at least for Ireland, is about 100 MW by 2030. That is 1.5% of the projected peak. Across the other markets Ember studied, reductions range from 0.1% to 0.6%.
The finding lands in the middle of a busy week for energy management software. Vendors are pitching platforms that promise to turn AI load into a grid asset. Ember's numbers suggest the software layer matters, but not in the way the pitch decks imply.
The reason is workload mix. Ember estimates that cloud services, enterprise workloads and AI inference will account for roughly 87% of global data center capacity by 2030. All of these need low latency. Training is the flexible part. Under Ember's assumptions, 35% of training can shift, 5% of inference, and nothing in traditional data centers. That yields about 6.5% of global capacity theoretically movable. An average utilization rate of about 48% cuts the effective figure to around 3%.
Ireland is the stress case. Data centers already account for approximately 19% of peak demand there, pv magazine reports. In January 2025, when system demand peaked at 6 GW, data centers drew an estimated 1.1 GW. Ember projects that share rising to 24% by 2030, with data center demand at 1.6 GW against a 6.5 GW system peak. A large data center can need 200 MW to 500 MW and be built in 18 to 24 months. Grid expansion takes four to 10 years.
What the software can and cannot move
The alternative Ember examines is bring-your-own-capacity: keeping the workload running while swapping grid power for on-site generation or storage. That is where the five times figure comes from. In Texas, behind-the-meter generation and storage could cut up to 2.7 GW, according to the analysis. The software job shifts from scheduling compute to orchestrating batteries, generators and grid signals in real time.
That is a harder engineering problem, and a more familiar one for the vendors in this market. Ireland's regulator has already put numbers on the distribution side. CleanTechnica reported on 29 October that the Commission for Regulation of Utilities' first national Flexibility Needs Assessment identifies about 234 MW and 58 GWh of potential upward flexibility requirements on distribution networks in 2028. At the Naas 38 kV/10 kV substation, the modeled requirement is 0.6 MW and 3 MWh in 2028, concentrated in January. It falls to zero once reinforcement lands by 2030.
The Naas example is the point. A battery elsewhere in Ireland can earn money in balancing markets and contribute nothing to that constraint. CleanTechnica notes that UK Power Networks' March 2026 assessment reports the utility stopped seeking flexibility tenders for certain single-transformer substations. Their constraints typically lasted 24 hours, and the market could not address them on technical and commercial grounds. Grid software that cannot see the electrical location of a constraint is, in that context, a dashboard.
"We see a very, very significant penetration of battery storage and looking ahead, we'll see this scaling up even more," Rystad Energy Renewables and Power Vice President Sally Bogle said at All Energy Australia, according to pv magazine.
The storage buildout is real. At the Melbourne trade show on 28 and 29 October, pv magazine reported that nine large-scale battery projects totalling 2.3 GW / 7.1 GWh were switched on in Australia in the first six months of the year. That beat the 2 GW / 5.1 GWh added during all of 2025. Another 13 projects totalling 1.6 GW / 6.5 GWh reached financial close in the same half. More than 550,000 residential battery systems have been installed in the 15 months since the federal Cheaper Home Batteries program launched. Organiser RX Global said more than 22,000 people attended, up about 5,000 on 2025, with more than 490 exhibitors.
Every one of those assets is a dispatchable unit that some platform has to aggregate, bid and settle. That is the commercial opening the software vendors are chasing. It is also where the location problem bites. A residential fleet in South Australia does little for a constrained transformer in Naas, and the same logic applies in Australia's own distribution networks.
Utilities move, slowly
The utility side is moving, but on infrastructure timelines. National Grid agreed last week to buy 29.5 hectares (73 acres) at Dorman Point on Teesworks land. The Guardian reported on 1 November that the site could become Britain's biggest substation, supplying a planned AI data center of nearly 500,000 sq metres. Ben Houchen, mayor of the Tees Valley, told Sky News that the site could offer up to 3.2 GW of capacity. "Everyone who has tried to build a datacentre in this country knows the biggest problem is getting enough power to the site quickly," he said, according to the Guardian.
That is copper and steel, not code. But the substation's value depends on how the load behind it behaves, and that is where energy management platforms enter. The UK government's draft national policy statement on data centers, published last week and quoted by the Guardian, calls them "as fundamental to our economy as the railways and electricity networks on which we have long depended". It also says the country must "take control over the development of datacentres".
Not everyone is building. The Guardian reported in October that AI companies told ministers they had 12 months to head off public opposition to data centers or risk US-style protests. Campaign groups have opposed projects from Auchtertool, north of Edinburgh, to Brick Lane in east London. Software can improve utilization and shave peaks; it cannot fix a planning dispute.
Across the Atlantic, the plans are going the other way. Tucson Electric Power and Arizona Public Service filed their 2026 Integrated Resource Plans on Friday, according to CleanTechnica on 2 November. APS plans more than 4,000 MW of new fossil gas by 2041, including the Desert Sun plant and eight gas turbines in Maricopa County under the Redhawk Expansion. It has extended coal operations at Four Corners to 2038 instead of exiting in 2031. APS is also seeking a 14% rate hike that would add roughly $240 a year to the average residential bill. TEP plans 300 MW of new gas and has said it will not meet its earlier target of an 80% carbon cut by 2035.
Demand growth of that kind is usually framed as a reason to build more peakers. It is equally a reason to measure how much flexible load is actually available. Ember's 3% effective figure suggests the answer is less than the market narrative assumes, at least for workloads that cannot move.
The software market fills in
Vendors are not waiting for the analysis to settle. On 30 October, The Robot Report covered Percepto's launch of Axion, a hardware-agnostic platform. It ingests inspection data from third-party drones and sensors and applies more than 160 AI models to predict equipment failure. The company is based in Modi'in, Israel, with US headquarters in Austin, Texas. It is shifting from drone hardware to software-as-a-service under new CEO Julien Bellanger. "When I looked at Percepto, I saw a company that had already done the hardest part: fully autonomous drones flying real missions every day at some of the most demanding industrial sites in the country," Bellanger told The Robot Report.
It is a different corner of the same market. Grid operators need condition data from transmission and distribution assets before they can trust a flexibility bid. Inspection platforms are one route to that data. The Robot Report notes that Axion is designed for upstream oil and gas, power transmission and distribution, and downstream processing. It also notes that manufacturing will be onshored in the US to support the American Drone Dominance Initiatives.
Capital is moving too. CNBC reported on 2 November that OpenAI signed agreements with Abu Dhabi entities spanning energy, government and investment. These include a deal with ADNOC and G42 to deploy its frontier technology across ADNOC's operations, and a five-year agreement with Inception42 covering government, enterprise and education. Sam Altman met Crown Prince Khaled bin Mohamed bin Zayed Al Nahyan, the Abu Dhabi Media Office said. CNBC also reported that OpenAI is seeking to raise up to $20 billion in a round co-led by UAE investors.
Compute deals of that size are, in effect, grid deals. Every gigawatt of new load needs a contract, a connection and a control layer. The question Ember's work raises is how much of that control layer can be software, and how much has to be built in the right place at the right voltage.
Older examples show the pattern. Eugene, Oregon, adopted a Home Energy Score disclosure ordinance on a 6-1 council vote on 28 October, CleanTechnica reported. That makes it the fifth Oregon city with such a rule and pushes the share of Oregonians covered past 1 million, or 24.4% of residents. The policy is disclosure, not dispatch, but it produces the kind of building-level data that demand response platforms need to aggregate residential load.
Electric transport is following a similar path. CleanTechnica reported on 30 October that Artemis's EF-12 electric hydrofoil pilot boat entered sea trials in 2026 after Brabo ordered one in August 2024. A Python model put the fifteen-year discounted cost at about €5.32 million against €3.81 million for a conventional diesel tender, a premium of about 40%. The vessel was modeled as foilborne for zero, roughly 6% or roughly 13% of its underway time depending on port speed permissions. Efficiency gains that only apply in a narrow operating window do not pay for themselves. That is the same trap facing flexibility software that only works when the workload can move.
Ford's plans offer a reminder that timelines slip. Electrek reported on 29 October that a Ford Authority claim the Escape would return as an EV on the Universal Electric Vehicle platform was called "pure speculation" by a Ford spokesperson. CEO Jim Farley confirmed on the Q3 earnings call that the Fathom pickup, the first UEV model, remains on track for the fourth quarter of next year. Preorders open in early 2027 at $28,350 plus a $1,595 delivery fee.
For grid software, the near-term test is less about launches than about proof. Ember's 1.5% figure for Ireland and 2.7 GW for Texas set a benchmark that vendors will have to beat or explain. Regulators in Ireland have already published the substation-level numbers that make the comparison possible. The next twelve months of deployments will show whether the platforms can move load to where the constraint actually is, or whether the grid gets reinforced instead.
Sources
10- 01Behind-the-meter generation could deliver five times more grid relief than data center load shiftingEN
- 02Key takeaways from All Energy Australia 2026EN
- 03More of the Same Expensive Dependence on Fossil Fuels in TEP and APS 2026 Energy PlansEN
- 04Ireland's Grid Flexibility Assessment Shows Why Location MattersEN
- 05Eugene Becomes 5th Oregon City to Adopt Home Energy Score RequirementEN
- 06Electric Hydrofoils Save Energy, But Do They Earn Their Price?EN
- 07National Grid to site biggest substation in UK near planned Teesside datacentreEN
- 08OpenAI deepens Abu Dhabi foothold as Sam Altman inks deals spanning energy, government and investmentEN
- 09Percepto launches Axion AI platform to unify inspection data analysisEN
- 10The Ford Escape may return as an EV based on the UEV platform [Update]EN
All figures and quotations in this text come from the sources listed below.
Content prepared by the editorial team with AI assistance.
Comments
0- No comments yet — be the first.