The diagram is the compact form of the model, and a diagram can only be argued with by someone who is looking at it. This page is the same system written out: sixteen quantities and twenty links between them, each link with its sign and the reason for it. It is meant to be read on its own, checked against the sources, and disagreed with in specific places.
A generating technology with no fuel cost bids into the market at close to zero and sets the price whenever it is running. Every additional unit of that technology therefore lowers the price in exactly the hours its own output arrives. The revenue a wind farm earns per MWh falls as wind capacity grows, not because the market has failed but because it is working, and the same holds for solar against sunny hours and for batteries against the spread they exist to arbitrage.
The claim the model makes is that this is a general property rather than a fault of any one technology. Any resource whose value comes from a price signal will compete that signal away as it scales, and it will stop expanding when the signal has fallen to its own marginal cost. What follows from that is not an argument against building the technology. It is an argument that the useful question is what absorbs the surplus, because the surplus is what carries the revenue away, and whatever drains it decides where the value lands.
The system is drawn as five coupled structures. Three of them are balancing loops, meaning they resist the change that sets them off. One is a reinforcing loop, meaning it accelerates it. The last is not a loop at all but a fork, a point where the same surplus can be taken up by two different kinds of demand with very different consequences downstream.
Each link carries a plus or a minus. A plus means the two quantities move in the same direction: raise the first and the second rises. A minus means they move in opposite directions. Signs multiply around a loop, so a loop containing an odd number of minus links is balancing and one containing an even number, including none, is reinforcing.
Nothing in the notation carries a timescale, which is the standard weakness of this kind of diagram. Capacity takes years to build, prices move within the hour, and a connection queue clears over a decade. Where the delay matters to the argument it is stated in the text rather than encoded in the picture.
The first loop has four links and closes on itself. A favourable investment incentive adds weather-dependent capacity, solar and wind, which is a plus. More capacity deepens the supply glut in the hours when the weather cooperates, another plus, since output is correlated across every plant of the same type in the same region. The glut pushes down the capture price, the average value actually realised per MWh produced, which is the loop's one negative link. A lower capture price weakens the investment incentive, a plus, and the loop is closed.
One negative link makes B1 balancing. It does not stop deployment; it sets the level at which deployment stops being worth doing under a given cost structure, and it moves that level down as costs fall. Germany is the clearest illustration in the evidence base, and the reason the loop is drawn first is that everything else in the model is a response to its output, which is a quantity of cheap energy arriving at times nobody chose.
If flexible demand appeared wherever power was cheap, B1 would be self-limiting in a benign way: the glut would be consumed, the capture price would recover, and the loop would settle. That is the case the model exists to complicate. Version 4 held that the obstacle was economic, a breakeven the surplus must clear before anyone switches fuel. Version 5 holds that there are two obstacles in parallel, and that a project can clear either one and still be stopped by the other.
The distinction matters because the two gates fail in different ways. A price gate rations by cost, clears continuously, and tells an applicant immediately whether the answer is yes. A connection gate rations by permit and by waiting time, and can leave a project that is economic on every hour of the year waiting several years for an answer that may be no.
The quantity that matters is not the average price but the number of hours below breakeven, where breakeven is the electricity price at which an electric boiler beats the gas boiler it would replace. A deeper glut raises that count, a plus. So does a higher breakeven price, since raising the threshold puts more hours underneath it.
The breakeven itself has two policy inputs pulling against each other. The electricity tax and network tariff enter negatively: every öre added to the delivered cost of electricity means the spot price has to fall further before the switch pays, so the threshold drops. The carbon price on gas enters positively, because a more expensive fossil alternative means a higher electricity price can still be tolerated.
Finland and Sweden run the cleanest natural experiment on this link, since the hardware is comparable and the tax treatment is not. Finnish electric boilers went from 16 GWh in the early series to 750 GWh in 2023 and 2,580 GWh in 2025, about 8% of district heat produced, on a levy near €0.63/MWh. Sweden's roughly 1.24 GW of boilers sits largely idle at 43.9 öre/kWh. The convex shape of the Finnish curve is what a policy step looks like when the capital takes about two years to arrive, which is the delay the diagram cannot show.
The tax has a second property that matters more than its level, and the model previously missed it. Fernqvist, Broberg, Torén and Svensson, writing in Energy Policy, point out that Sweden's electricity tax is static, "a set price per kWh (0.36 SEK/kWh in June 2022), independent of the electricity price". A charge that does not move with the market puts a floor under what a district-heating operator actually pays. However deep the glut goes, the delivered cost stops falling, and negative prices, which are the strongest signal a market can send to consume, do not exist at the meter. The tax does not only shift the threshold downward; it truncates the signal B2 depends on, and it truncates it precisely in the hours the model cares about most.
The same paper sets the Swedish scale. Of the 305 largest district-heating systems, 72 have heat pumps or electric boilers installed, 1,207 MW of heat pumps and 1,151 MW of boilers, a combined maximum electrical draw of 1,540 MW, and between 2017 and 2021 those assets delivered an average of 7.6% of heat production. Estimated power-to-heat potential runs from 0.2 to 8.6 TWh, a range wide enough to be an admission that the number is not known. Heat storage across Swedish district heating came to roughly 150 TJ in 2016 with about 96 TJ available for additional load variation, which converts to around 42 and 27 GWh. Set against the daily flexibility needs in the German study above, that is a day of shifting rather than a season of it, and the authors say directly that the opportunity is unexploited against Denmark and Germany and that the literature will not support an accurate national estimate.
Sweden has already run the experiment, which is the part of the argument that does not depend on any projection. Power Circle counts about 130 electric boilers in the district-heating system at roughly 10 MW each, some 1,300 MW installed, and dates their peak to 1990, when 6.3 TWh of electricity went into district heat. Today those boilers produce about 0.1 to 0.4 TWh a year. The load B2 depends on did not fail to appear. It was built, at scale, and then it stopped, and the same report notes that only around 30 MW of that capacity is prequalified to the ancillary-service markets, against roughly 300 MW of heat-pump capacity.
The wedge can be priced exactly. Sending an extra MWh into an electric boiler costs 1,264 SEK/MWh once the spot price, the energy tax and the network fee are all counted, and 694 SEK/MWh with the tax taken out. The tax is therefore close to half of what the operator pays, and because it is flat it does not shrink as the spot price falls. The static-tax argument above is not a qualitative point about incentives; it is most of the delivered cost in exactly the hours the glut is deepest.
What survives the gate is instructive. Swedavia's aquifer store at Stockholm Arlanda has operated since the summer of 2009 and reduces the airport's purchased energy by 19 GWh a year, 4 GWh of electricity and about 15 GWh of district heat, which Swedavia equates to 2,000 houses. It cools the terminals in summer, banks the warmed water underground, and returns it in winter to melt snow on the aircraft stands and pre-heat ventilation air. The reason it works where the electric boilers do not is that it moves heat between seasons rather than converting electricity into heat, so almost none of its output passes through the meter where the tax is levied. A gate that rations by a per-kWh charge does not bind on a technology that barely draws kWh.
The incidence of this gate is uneven, and in a way that matters for the rest of the model. Moving the threshold from €54 to €40 costs SE1 and SE2 about three points of qualifying hours and costs SE3 thirteen points and SE4 twelve. Where power is already very cheap the carbon price is close to irrelevant. Where it is marginal, which describes most of the European industrial base, the carbon price is close to decisive.
The second gate is the amendment that defines version 5. Passing the price test earns the right to want a connection, not the connection itself. Whether grid connection is granted depends on the state of the connection queue, and the link is negative: a longer queue means a lower probability, and a longer wait, for any individual applicant.
Outright refusal is the rare form. Energimarknadsinspektionen examined Svenska kraftnät and six regional network companies on 11 June 2026 and found connection times routinely exceeding the two-year statutory limit, especially for large projects. What it documented was delay rather than denial. Delay still functions as a filter, because it selects for whoever can hold capital idle the longest, which is a selection on balance sheet rather than on value created per MWh.
The clean binary case is Swedish. Power2Earth, the fossil-free fertiliser venture of Lantmännen, Fertiberia and Nordion Energi, was shelved in autumn 2025 after Svenska kraftnät refused the 4 to 5 TWh it needed. Per Arfvidsson of Lantmännen described the refusal as concerning "inkopplingen och kapaciteten i det svenska elnätet, alltså trängsel i infrastrukturen".
What makes that case structural rather than anecdotal is where it happened. Luleå sits in SE1, which spent 65.8% of 2026 so far below the €54 breakeven, the highest share of any bidding zone in the EU and Norway, with SE2 next at 64.5%. On the price gate these are the most favourable conditions in Europe. If cheap power were sufficient the plant would be under construction. The scarce resource is the right to draw power, not the power.
Gate 1 has policy inputs, so gate 2 needed one as well, and it is the maturity and allocation rule that governs the queue. Its link to the queue is negative, since a stricter rule shortens it.
Sweden allocated connections first come, first served until 2024. Under that rule the queue ranks applicants by how early they applied and by nothing else, so a category of new load with committed capital is structurally near the front. The maturity requirements introduced in 2024 replaced it and cut the queue by 14,300 MW on their own, with roughly 20,000 MW of speculative bookings cleared since. Svenska kraftnät announced capacity zones and business matching on 4 May 2026, which goes further.
Whether this removes the capital bias or formalises it is an open question rather than a finding, and the model should be read as posing it. Documented milestones, permit progress and a detailed energy analysis are easier to produce for a financed hyperscaler than for a municipal heat utility or a farmer-owned joint venture. The rule now selects for project maturity, and maturity is partly a function of access to capital.
Two kinds of load can take up the surplus, and the model treats them as different nodes because they behave differently in every link that follows.
Flexible electrified demand, meaning industrial steam, electric boilers and heat pumps, consumes when power is cheap and stops when it is not. That is what lets it fill the trough rather than lift the whole curve, and filling the trough is what the capture price responds to. Data centres run flat, close to 24 hours a day, so their load raises consumption in every hour more or less equally. A flat load does raise the average price, but it does not selectively remove the cheap hours that cannibalization is made of. This is a correction to version 3, which drew both kinds of demand as absorbing the glut equally.
The asymmetry in size is larger than the asymmetry in behaviour. The IEA puts the technical potential for electrified industrial heat in the EU at around 600 TWh a year, against a projection of about 115 TWh for European data centres in 2030, up from roughly 70 TWh in 2024. The 600 TWh is technical potential rather than a forecast, and should not be read as demand that is going to appear.
Data centres therefore enter the diagram mainly through the queue rather than through the glut. Their link to the connection queue is positive: they take slots. Svenska kraftnät received 43 applications in 2025 covering over 9,000 MW of withdrawal, about half of it data centres, and granted 3,695 MW of consumption across 22 positive decisions. Roughly two applied megawatts in five got through. Data centres hold 5,000 to 6,000 MW in the queue with about 80% of it in SE3, set against roughly 7,000 MW of new industrial demand expected across the whole of 2026 to 2030.
Data centres are the fastest-growing load in the system by a wide margin, and the IEA states the comparison directly: from 2024 to 2030 data centre electricity consumption grows by around 15% a year, "more than four times faster than the growth of total electricity consumption from all other sectors". Nothing else in the demand mix, including electrified heat, is expanding at that rate.
How fast a load grows and how much of the growth it constitutes are separate questions, though, and the second one has a different answer in different places. Data centres account for under 10% of global electricity demand growth between 2024 and 2030 in the IEA's base case, and for about 10% of demand growth in the European Union on today's policy settings. In the United States they are close to half of it. The European figure is the relevant one for this model, and it means the bulk of the growth in European demand still comes from electrification of heating and transport, recovering manufacturing output, and cooling. A version of the argument claiming data centres are the main source of European load growth would be wrong by a factor of several.
Sweden shows where the acceleration is actually visible. Applications to connect large data centres in SE3 went from roughly 1,310 MW in 2024 to 6,692 MW in 2025, close to five-fold in a single year. Consumption has not moved anything like that fast: Swedish data centres draw around 4 TWh a year now, and the published projections for 2030 disagree by a factor of three. RISE puts it near 5 TWh and Energimyndigheten at 4.4 to 5.0 TWh, that estimate dating from 2023 and flagged by Svenskt Näringsliv as probably low because it predates the AI build-out, while scaling the Nordic system operators' forecast of 7 TWh in 2023 rising to 25 TWh in 2030 gives Sweden something like 14 to 15 TWh. Nobody has narrowed that range, and the model should not pretend otherwise.
What accelerated, then, is the queue rather than the meter. That distinction resolves what otherwise looks like a contradiction in the model, since data centres are about half of Swedish connection applications and about a tenth of European demand growth, and both figures are correct. The queue is a biased sample of demand growth. It sees loads that arrive as large discrete blocks requiring new firm capacity, and it is close to blind to loads that arrive distributed and incrementally behind connections that already exist. A heat pump, a boiler retrofit or a depot charger mostly never appears in it. The same asymmetry applies to the fork chart: 600 TWh of technical potential against 115 TWh of projected data-centre demand is a comparison of stocks, while the queue is a flow, and a stock that arrives slowly can be much the larger of the two and still lose every contested slot.
The consequence for gate 2 is that it selects for lumpiness on its own, before anyone has behaved strategically. A load that must be granted its capacity in one block competes in the queue; a load that can grow behind an existing connection does not have to. That sits alongside the selection on who can wait longest and the selection on who can document maturity, and unlike those two it requires no bias on anyone's part and cannot be removed by fixing the allocation rule.
Nor is the queue a Swedish peculiarity. The IEA reports connection waits across the European Union ranging from two to ten years depending on the country, averaging seven to ten years in the FLAP-D hubs of Frankfurt, London, Amsterdam, Paris and Dublin. Gate 2 was drawn from one Swedish refusal, and it generalises.
The queue is not the only route by which one kind of load displaces another. Because a data centre is capital-intensive, electricity is a small share of its total cost, so it can pay a price that would close an industrial plant and still earn its return. Pär Holmberg of IFN makes the point directly: the build-out continues even as it raises the price, and the pressure falls on southern Sweden, where broadband and latency draw the halls and there is no local surplus to absorb them. The same channel is visible in PJM, where data centres drove 63% of the 2025/26 capacity-price increase, about $9.3bn.
The two channels have two geographies, which is the part of the model most easily missed. The connection gate bound in the north, at Luleå, where the glut is deepest and the queue is the constraint. The price channel binds in the south, where the data centres want to sit and the surplus that would blunt them does not exist. A version of the model drawing only one of them would describe half the map.
The strength of the two claims is also different, and blurring them would be the easiest way to make the model wrong. The concentration of applications is documented. The displacement is inference. A systematic search of Swedish sources covering 2024 to 2026 found no named non-data-centre load refused, queued or shelved in Mälardalen, Stockholm, Uppsala, Gävleborg or Västmanland with a data centre identified as the competing load. Counter-evidence points the other way as well: Uppsala's position is that connection is available but slow, Ellevio reported Stockholm's shortage easing in 2025 after 2.42bn SEK of investment, and the only capacity-paused facility found in those counties is itself a data centre, Microsoft's Sandviken build. Connection negotiations are commercially confidential, so absence of public cases is not proof of absence, but it is not evidence either.
The second balancing loop is the one that would resolve B1 if it were allowed to operate. The glut raises the count of hours below breakeven, a plus. Those hours, if the connection is also granted, bring flexible electrified demand into existence, another plus. That demand consumes precisely in the surplus hours, which reduces the supply glut, and this is the negative link that closes the loop and makes it balancing.
The single most important property of B2 is that its gain is set by policy rather than by physics. Both gates sit inside the loop. Tax and tariff decide how many hours qualify; the queue and its allocation rule decide how much of the qualifying demand is ever built. A jurisdiction can leave B1 running at full strength and hold B2 near zero without deciding to do so, simply by taxing electricity heavily at the meter and rationing connections slowly. Finland and Sweden differ on the first of those, and the boiler output series is the result.
B2 as described so far invites an objection that the model should answer rather than avoid. If the effect of the trap is to push prices down, that looks like a transfer from generators to buyers, and a reader is entitled to ask why anyone other than a wind investor should mind. The answer is that a system without flexibility does not deliver cheap power. It delivers a wider price distribution.
Geis, Lindner and Brown model the German sector-coupled system to 2045 under four flexibility scenarios, published open access in Advances in Applied Energy. Their constrained-flexibility case produces a very high count of zero-price hours, which is the trap running unchecked, and at the same time an average electricity price 14 to 17 €/MWh above the base case, because the peak still has to be covered by backup that has to earn its keep in fewer hours. Their high-flexibility case produces the opposite on both ends: few zero-price hours, 7% in both 2035 and 2045, and the lowest average price of any scenario, 3 to 5 €/MWh below base. System costs run 6.9% higher when flexibility is constrained and 3.3% lower when it is abundant, and the abundant case needs 22 GW less backup capacity.
So the two ends of the distribution move together rather than trading off. Throttling B2 does not buy cheap electricity at the price of a lower capture rate; it buys more worthless hours and more expensive ones at once. That is the strongest available argument for treating the gates as policy choices rather than as facts of engineering, and it is worth stating that this is a model result rather than a measurement.
The same paper qualifies who the model casts in the role of flexible demand. In their 2045 base case, daily flexibility comes mostly from stationary batteries at 38% and flexible electric-vehicle charging at 30%, with resistive heaters at 7% and electrolysis at 6%. A cost-optimising system does not reach for industrial heat first, and a model that writes electrified heat into the centre of B2 should say so.
Heat also enters on a condition rather than unconditionally. Heat pumps and resistive heaters without thermal storage raise winter flexibility needs in their results, because the heat is wanted when it is cold rather than when power is cheap. In district heating, where storage exists, resistive heaters supply positive flexibility through the year. Storage is the component that converts heat from a consumer of flexibility into a supplier of it, which is a sharper statement of what the Finnish boiler series demonstrates. One inversion is worth noting alongside it: the low-flexibility scenarios build more central resistive heaters, using them as cheap substitute flexibility where nothing else is available, while the high-flexibility scenario builds fewer and runs more heat pumps, which are more efficient but less flexible.
The third loop is reinforcing, with three positive links and no negative one. Flexible electrified demand keeps energy-intensive industry viable, because electrified heat is what lets a smelter or a chemical plant survive a fuel price it could not otherwise carry. Industry sustains GDP, tax revenue, exports and jobs. Value-added funds the grid investment and the flexibility that would let more electrified demand connect, which closes the loop.
Reinforcing loops run in both directions, which is the whole of the concern. Run forward it compounds; run backward it also compounds, and the backward direction is the one with a completed example. German production in energy-intensive branches ran 17.8% below 2021 in 2025 and has stayed at roughly that level for four years, with chemicals down 18.1% between February 2022 and March 2026. Earlier versions of the model described this as an ongoing shock being absorbed. Four years of flat depression is a transition that has already run once to completion in a large economy, and R1 should be read that way rather than as a risk under discussion.
The employment asymmetry at the fork feeds this loop and should be stated with its correction attached. A colocation site runs 20 to 30 permanent staff per 100 MW, and a large build is roughly 1,500 construction jobs against about 100 operational ones. Brookings also finds that counties receiving their first large data centre see total private employment rise 4 to 5% over five or six years. The effect is small, and naive estimates inflate it about threefold, but it is not zero, and a version of the argument that treats it as zero is attackable on that sentence alone.
R1 routes the value of cheap power to industry through one channel only, the flexibility that lets a plant reach the cheap hours. There is a second channel, the price level itself, and it is worth drawing as its own loop because it answers the oldest claim in the debate directly, that cheap energy is the backbone of an industrial economy. The quantity it needs is the industrial power price, the all-in delivered cost a smelter or a chemical plant actually pays, the spot price plus the electricity tax and the network tariff, and it is a different quantity from the capture price that B1 tracks. The capture price is what the producer realises; the industrial power price is what the consumer pays. The same supply glut lowers both, which is the part most easily missed: one quantity read at two meters, a loss to the generator and a subsidy to the buyer at the same instant.
The loop has three links. A deeper supply glut lowers the industrial power price, a minus. A lower industrial power price keeps energy-intensive industry viable, another minus, because for a smelter the power bill is a large share of cost rather than a small one. Viable industry sustains GDP, tax, exports and jobs, which funds the grid and flexibility that bring more flexible electrified demand into existence, and that demand consumes the surplus and reduces the glut. Three minus links make B3 balancing: the glut has a second natural absorber besides B2, the industrial economy that its own cheapness keeps alive.
What makes B3 more than a restatement is that its gain runs through the same two gates as B2. The industrial power price is floored by the flat electricity tax, the same truncation Fernqvist identifies, so however deep the glut goes the delivered cost stops falling and the first link is throttled at the meter. The return leg, new flexible demand, is throttled by the connection queue. So the backbone the argument points at is real and now has an arrow, and the diagram immediately says it conducts only when the gates are open. Germany is the case where the price rose from outside, the gas shock, so the cheapness link never fired and R1 ran backward to 82.2% of 2021. Sweden is the case where the glut is genuinely cheap but the tax floor and the queue keep B3 from conducting, so the surplus persists and industry is neither fed nor killed.
One caution belongs on the third link. Its strength is a property of the load rather than a universal. For energy-intensive tradables the power bill is a large share of cost, so the affordability link is strong. For a data centre electricity is a small share of cost, so the same link is close to flat, which is exactly why the fork places data centres on the queue and not on the price. Drawn into the data-centre node this loop would be wrong; it belongs to energy-intensive industry alone.
The instrument that several jurisdictions have converged on addresses the fork and both gates at once, without any of them appearing to have coordinated. Ireland's moratorium, in force from November 2021, was lifted in December 2025 and replaced by a requirement that new connections self-firm, installing on-site generation or storage covering their full demand. Holmberg proposes that southern Swedish data centres be required to sign long-term contracts for new unsubsidised generation. The US administration is encouraging data centres to fund their own supply.
Each of these converts a load that competes for existing firm capacity into a load that brings its own. If new supply arrives in step with new demand, the price channel closes on its own and the queue pressure falls with it. Whether the instrument survives contact with the projects it is aimed at is not yet answerable from public evidence.
The Swedish data-centre growth path is the least settled quantity in the model, spanning 5 to 15 TWh for 2030 across published forecasts, and the queue claim and the consumption claim should never be swapped for one another. Beyond that, the B1 link is the most exposed to counter-evidence, and the honest test is a market where weather-dependent share rose substantially over several years while capture rates held flat, with no offsetting change in fuel prices or storage. The German solar series is the model's main support here and a single market is thin support.
The gate 2 mechanism rests on one clean refusal. Power2Earth is a single case, and in Norrbotten the load competing for capacity is steel and hydrogen, named as Hybrit, Stegra and LKAB, rather than data centres. No source names a data centre as the cause of that refusal. A records request under offentlighetsprincipen to Energimarknadsinspektionen's dispute register or to Svenska kraftnät's list of connection cases is the only route the search identified to settle whether contested cases exist, and it has not been made.
The NEM battery revenue series is excluded from the model for now rather than published with a caveat. A figure of A$148k/MW for 2024 could not be reconciled against published six-month windows, one source puts that year at A$50 to 70k/MW, and it is ambiguous whether the published numbers are gross or net and in which currency. The CAISO series is clean by comparison, falling 103, 80, 51 and about 38 USD/kW-year with January 2026 pacing toward 21.
The complete link inventory
Every causal link in version 5 of the diagram, with its sign and the loop it belongs to. Signs multiply around a loop: an odd count of minus links makes it balancing, an even count makes it reinforcing. Reading this table is equivalent to reading the picture.
| From | To | Sign | Loop | Because |
|---|---|---|---|---|
| Investment incentive | Weather-dependent capacity | + | B1 | Returns attract build-out. |
| Weather-dependent capacity | Supply glut | + | B1 | Output is correlated across plants of the same type in the same region. |
| Supply glut | Capture price | − | B1 | Zero-marginal-cost bids set the price in exactly the hours the plant runs. |
| Capture price | Investment incentive | + | B1 | Realised value per MWh is what the investment case is made on. |
| Supply glut | Hours below breakeven | + | B2 | A deeper trough puts more hours under any fixed threshold. |
| Breakeven price | Hours below breakeven | + | gate 1 | Raising the threshold puts more hours underneath it. |
| Electricity tax + tariff | Breakeven price | − | gate 1 | Charges at the meter mean the spot price must fall further before the switch pays. |
| Carbon price on gas | Breakeven price | + | gate 1 | A dearer fossil alternative lets a higher electricity price still compete. |
| Hours below breakeven | Flexible electrified demand | + | B2 | More qualifying hours improve the payback on the equipment. |
| Data centres | Connection queue | + | fork | They take queue slots rather than surplus hours. |
| Maturity + allocation rule | Connection queue | − | lever | Milestone requirements clear speculative bookings. |
| Connection queue | Grid connection granted | − | gate 2 | A longer queue lowers the odds, and lengthens the wait, for any applicant. |
| Grid connection granted | Flexible electrified demand | + | B2 | Nothing is built without the right to draw power. |
| Flexible electrified demand | Supply glut | − | B2 | Load that follows price fills the trough instead of lifting every hour. |
| Flexible electrified demand | Energy-intensive industry | + | R1 | Electrified heat keeps plants viable at fuel prices they could not otherwise carry. |
| Energy-intensive industry | GDP · tax · exports · jobs | + | R1 | Value-added, directly. |
| GDP · tax · exports · jobs | Flexible electrified demand | + | R1 | Value-added funds the grid and the flexibility that let more demand connect. |
| Supply glut | Industrial power price | − | B3 | The glut that starves the capture price also cheapens the buyer's wholesale cost. One quantity, two meters. |
| Electricity tax + tariff | Industrial power price | + | gate 1 | Meter charges mark the delivered price back up, and the flat tax does not fall with the spot price. |
| Industrial power price | Energy-intensive industry | − | B3 | A lower all-in price keeps plants viable where power is a large share of cost; for data centres, where it is not, the link is flat. |
Sources for every figure on this page are listed with headline, publisher and date on the references page. The zone-level shares below threshold are computed from the same Energy-Charts day-ahead data behind Elpris EU+Norge, covering January to July 2026. The five figures marked as verified were checked in the verification pass. The diagram this page describes is version 5 of the model.