Electricity is the new bottleneck for AI
„AI expansion is now limited not by chips, but by network connectivity and available megawatts: where there is no power, there is no data center.”
Counter-narrative
Part of the scarcity may stem from over-announced demand rather than physical shortage. The stories only indirectly hint at this: in the USA, proposed data center connection requests exceed 700 GW, more than ten times the estimated current data center consumption. Flexibility also emerged as a form of network adaptation. The counter-narrative has no named, independent carrier in the examined material.
Carriers
Carriers: IEA and EIA, FERC and US grid operators (PJM), hyperscalers and Nvidia (Microsoft, Amazon), Moody's, Berkshire Hathaway (Greg Abel), data center industry leaders (NTT Data, Brookfield, Infosys)., US federal and state government, Hungarian economic press (Pénzcentrum)
Analysis
The narrative holds that the megawatt has become AI's currency, while grid access determines site decisions, investments, and political moves. The factual basis includes several elements. These are the reported grid-connection freezes in Texas and Norway, the FERC finding, and the delay of the British AI supercomputer, possibly until the mid-2030s. Included are October 18 and 19 reports on Microsoft's withdrawal from the UK, citing energy shortages and grid delays. The circle of stakeholders is broad. Utilities must rework grid planning, Nvidia and Amazon are investing in Texas power supply, and governments are resorting to extraordinary measures. The weakness lies in assessing future demand. Connection demand of over 700 GW is ten times current data center consumption, suggesting over-announcement. Thus, part of the scarcity may be artificial, and the narrative could swing to overbuilding.
What's at stake: Whether AI capacity is deployed where the grid can handle it, or whether other consumers bear the grid's cost and risk.
Narrative sub-threads
Connection freezes and delayed projects
Network operators and governments are halting or delaying the connection of new data centers. As a result, specific projects may be delayed by years or investors may withdraw; Microsoft has already pulled out of a UK project.
It is not primarily based on forecasts. The evidence: connections were reportedly halted in Texas, and according to one report in Norway; a project is delayed in the UK, and Microsoft pulled out of another. These are supplemented by reports of grid constraints in Europe, Sydney, Mexico and Israel, and by an IEA estimate.
→ lasting · evidence: strongDemand surge at the grid's capacity wall
Official and analyst estimates indicate AI-driven power demand is growing faster than the grid can expand. In the US, the EIA expects record consumption for 2026 and 2027. Proposed connection requests exceed 700 GW; data centers accounted for at least 60% of the power demand increase, and the Department of Energy declared grid emergencies in 13 states.
It relies on macro-level, quantitative demand and regulatory evidence (EIA, FERC, DOE), primarily with an American focus, not individual projects.
↗ rising · evidence: moderatePhantom projects and inflated megawatts
A significant part of the announced data center power demand exists only on paper. A report suggests that more than two-thirds of the US power required for the AI boom will likely not materialize, and the gap between announced and built demand complicates grid planning.
Representing a counter-voice: challenges the extent of the bottleneck by highlighting the gap between measured and announced demand.
↗ rising · evidence: moderateGas turbine shortage as a new constraint
The speed of on-site power generation depends on gas turbine manufacturing capacity. Orders are therefore facing long waiting lists, and developers are turning to stopgap solutions, such as retired aircraft engines; some of these, like the Crusoe Boom turbine plan, have failed.
The constraint is not the grid but the supply chain of power plant equipment; the actors are turbine manufacturers and suppliers.
↗ rising · evidence: moderateFlexible power draw as faster grid connection
According to one model, AI data centers could fit within existing network constraints with approximately 35 hours of annual load reduction. BCG estimates that flexible grid connection could shorten the wait time for power from 5–10 years to 2–4 years.
Response branch: the bottleneck would be avoided with software load management. Technology companies, an alliance, and a live demonstration are driving this (Luxor Energy and Bentaus reacted to the ERCOT signal); scaling remains questionable.
Sub-thread stories
Connection freezes and delayed projects
Texas halts data centre grid connections as Nvidia and Amazon invest in power supplyenergy specialistConnection queues and investment in on-site generation point to new rules for allocating grid capacity.supplierProjects that need their own generation and storage are worth bidding for.engineerDesigning solutions that do not depend on a grid connection, such as local generation and storage, becomes more valuable.Norway halts new grid connections over AI data centers' energy demandenergy specialistThe Norwegian move signals tighter regulation of grid connections.European grid bottlenecks are stranding billions in AI data center capitalenergy specialistBecause connection capacity is scarce, grid access must be secured at the start of an investment.IEA: data centres used 1.5% of global electricity in 2024, a fifth of planned projects may face delaysenergy specialistFor planned connections in clusters, grid delays must be built into project schedules and capacity planning.supplierA supplier serving data centre projects must account for the risk of project slippage in its order book.Grid connection becomes the key condition for AI data centres worldwideenergy specialistBecause of the new connection order and rules for large loads, grid development and connection plans must be aligned with readiness criteria.plant managerIndustrial plants' grid connection and capacity needs compete with those of data centres, so it is worth discussing expansion early in the connection process.SME ownerGrid scarcity and premiums raise electricity price and capacity risk, so long-term energy procurement is worth reviewing.Power supply problems delay launch of UK's largest AI supercomputerenergy specialistGrid connection determines the schedule of large data centre projects, so power supply planning must be settled at the start of the investment.supplierThe delay may require suppliers to renegotiate delivery schedules.UK's largest AI supercomputer may wait until the 2030s for sufficient powerenergy specialistPower supply is becoming a tighter bottleneck than hardware, so the grid connection deadline is the key parameter in data center plans.Sydney, Mexican and Israeli power grids hit limits as AI data centres expandenergy specialistIn several regions, delays in grid expansion limit AI data centre capacity.supplierDemand is growing for grid and storage solutions.Microsoft pulls out of Essex AI data centre project over power shortageenergy specialistA concrete example of how a lack of power supply can halt a project.supplierOrders for data centre projects may fall through if access to energy is not secured.Microsoft pulls out of major UK AI data centre project over grid delaysenergy specialistGrid connection timing can sink a project, so the grid expansion schedule should be agreed early.supplierWith the investor pulling out, related orders may be rescheduled or cancelled, so it is worth covering this in contracts.
Demand surge at the grid's capacity wall
AI infrastructure spending to reach $725 billion as power becomes the bottleneckenergy specialistThe bottleneck lies in electrical capacity, so grid and generation projects are gaining value.supplierThe investment wave is directed at electrical and cooling products.EIA: US power use to hit record highs in 2026 and 2027 as AI use surgesenergy specialistGiven record demand, data centre load must be treated as a separate scenario in generation and grid capacity plans.US electricity demand could rise up to 32% by 2030 because of AI data centresenergy specialistGrid capacity and reliability risks call for a longer horizon for connection and capacity planning.plant managerIndustrial power price and grid-connection risks should be factored into investment decisions.Studies: AI data centers mean a heavier grid-upgrade burden than previously expectedenergy specialistGrid-upgrade costs and forecasts point to rising connection charges and tariffs, which must be factored into capacity planning.plant managerGrid congestion may also affect industrial consumers' connection and electricity-cost plans.AI data centres' power demand is redrawing the grid: capacity wall, flexibility and build paceenergy specialistGrid expansion takes years, so flexible load and load management borrowed from Bitcoin miners are the short-term tools.plant managerIndustrial power supply and data centre demand compete for the same capacity, so it is worth filing your own connection application early.supplierFor transformer, switchgear and generator makers, demand is set by the grid constraint.FERC: US power grid still lags behind data center demandenergy specialistDelays in grid upgrades make it necessary to reassess connection plans and deadlines.plant managerWhen planning a new high-demand site or expansion, the time needed for grid connection must be factored in.US data centre connection requests top 700 GW, and the grid cannot copeenergy specialistGrid capacity has become a bottleneck, so connection requests and self-generation alternatives need to be scheduled early.plant managerIndustrial power consumers should review their own expansion plans in light of grid upgrade costs and connection queues.Power and skilled labour, plus cooling and GPU links, constrain AI data centre buildsenergy specialistAccess to power is the main constraint on project schedules.skilled workerA shortage of skilled construction and operations workers creates demand in data centre trades.engineerCooling and interconnect design is a key bottleneck for expansion.US power grid constraints could upend Big Tech's AI plans, report saysenergy specialistGrid capacity may become the limiting factor for scheduling AI data centers, which calls for grid expansion planning.US declares grid emergency in 13 states and national emergency to protect bulk power systemenergy specialistThe emergency measures may affect grid operating priorities and load management.plant managerThe power-supply risk at US plants is worth reassessing.Data centers accounted for at least 60% of US electricity demand growth last year, grid struggles to keep upenergy specialistGrid planning needs a separate scenario for data center load growth, and large connection requests should be handled in phased stages.plant managerGrid constraints may also slow the connection of factory expansions, so power requirements should be reported early.
Phantom projects and inflated megawatts
Most power sought for US AI data centres will not materialise, report saysenergy specialistMost grid connection requests do not represent real load, so investment decisions require filtering for serious projects.supplierOf the announced data centre pipeline, capacity is worth reserving only for funded, permitted projects.Texas grid strained by "ghost demand" as AI data centre power requests balloon on paperenergy specialistIn grid planning, the gap between announced demand and actual connection should be modelled separately.plant managerInflated demand requests can slow industrial connection queues, so your own application should be documented.supplierBecause of demand uncertainty, supplier capacity expansion should be tied to projects that are actually being built.Data center power demand: uncertain forecasts and the risk of an AI slowdownenergy specialistGiven uncertain demand forecasts, capacity expansion decisions need to be backed by scenarios.supplierVolatility in data center orders warrants caution in capacity and inventory planning.Ghost properties and fake megawatts distort the AI data centre market, report saysenergy specialistAnnounced capacities should be verified against grid connection commitments and permits.
Gas turbine shortage as a new constraint
Gas turbine shortage becomes the biggest constraint on powering AI data centresenergy specialistGas turbine projects have long delivery lead times, so early reservation and alternative generation options should be considered.supplierExpect stronger demand for turbine-related parts and servicing.Gas turbines become AI infrastructure's new bottleneck after semiconductors, with long order backlogssupplierThere is long-term demand for parts and capacity offers tied to the order backlogs of turbine, switchgear and UPS manufacturers.energy specialistGiven turbine lead times, delivery time should be the first parameter when planning on-site (off-grid) gas-fired solutions.plant managerProcurement lead times for high-voltage equipment may also stretch a plant's own expansion plans.Crusoe walks away from $1.25bn Boom turbine dealenergy specialistProcurement risk for on-site gas-turbine supply has increased, so turbine delivery and contract terms warrant close scrutiny.Crusoe abandons $1.25 billion plan to use Boom turbines at AI data centresenergy specialistOn-site power plans built on new, unconventional turbine types are risky, so a fallback scenario is worth keeping.US AI data centers turn to retired jet engines for power generationenergy specialistTemporary on-site gas turbine generation can be an alternative to grid waiting times, but emissions and permitting requirements must be checked.engineerWhen designing power supply independent of the grid, the output and maintenance needs of engine-based units should be assessed.
Flexible power draw as faster grid connection
AI GPU cluster responds to ERCOT peak-demand signal within a secondenergy specialistAvoiding peak-period (4CP) charges by curtailing load can be a practical cost-cutting tool for large consumers.engineerRapid GPU load reduction depends on the control layer, so receiving grid signals and automatic response logic must be designed in.Emerald AI raises $150 million at $1.05 billion valuation for flexible data center power useenergy specialistA flexible data center able to curtail its load can secure a faster grid connection, so it is worth offering this in connection negotiations.engineerShifting and curtailing compute load over time becomes a plannable feature that must be built into scheduling systems at the design stage.Nvidia, Google and Emerald AI form alliance for flexible data center power useenergy specialistA data center that responds to grid conditions and shifts its load represents a new connection and operating model, which should be built into capacity planning.engineerThe need for flexible consumption should already be taken into account when scheduling computing and cooling loads.supplierDemand is expected to grow for power-supply and control solutions that support grid-friendly operation.Model: AI data centers could fit within grid limits with 35 hours of reduced load a yearenergy specialistFlexible, curtailable data center load is a new parameter in peak-period planning, so demand-response terms are worth building into connection agreements.plant managerIf data center load can be reduced during grid peaks, that may also affect grid connection queues and tariffs.Flexible data centre power use could save grids billions, but scaling is uncertainenergy specialistFlexible load can be a way to secure grid connection faster, so load control should be built in at the design stage.Flexible grid connections could cut AI data centre power wait to 2–4 yearsenergy specialistDemand-flexibility services, such as load shifting and distributed generation, may offer new business opportunities.plant managerIf a plant's consumption can be shifted in time, such programmes can provide a source of revenue, so it is worth assessing your flexibility.OpenAI's new data centers to cut power draw at peak grid demand under Grid Protectors programenergy specialistFlexible, curtailable load is a new bargaining basis for large-customer connections, so it is worth assessing which loads can be reduced at peak times.engineerIn data center design, load reduction (storage, scheduling, reserve capacity) may now become a design requirement and should be built into specifications.
Google search interest
Hungary
| Topic and search term | Weekly | Average |
|---|---|---|
| Electricity is the new bottleneck for AIdata center power supply | 0 | |
| Connection freezes and delayed projectsNetwork connection | 1 | |
| Demand surge at the grid's capacity wallpower consumption growth | 0 | |
| Phantom projects and inflated megawattsdata center construction | 0 | |
| Gas turbine shortage as a new constraintgas turbine | 0 | |
| Flexible power draw as faster grid connectionflexible consumption | 0 |
World
| Topic and search term | Weekly | Average |
|---|---|---|
| Electricity is the new bottleneck for AIdata center power | 19 | |
| Connection freezes and delayed projectsgrid connection delays | 0 | |
| Demand surge at the grid's capacity wallelectricity demand forecast | 0.3 | |
| Phantom projects and inflated megawattsdata center boom | 0.3 | |
| Gas turbine shortage as a new constraintgas turbine shortage | 0 | |
| Flexible power draw as faster grid connectiondemand flexibility | 0.7 |
Google Trends relative search index over the period studied, weekly average. A narrative and its sub-threads appear in a single comparison: 100 is the group's most-searched day, and the figures can only be compared within the group. The search term was chosen by machine analysis; it measures how much the term is searched, not how widespread the narrative is.
Who is affected and how
- energy specialist
- The main work will be qualifying grid connection requests and setting the conditions for flexible connection. It is advisable to tie capacity reservation to milestones and guarantees to filter out speculative demand.
- plant manager
- Industrial consumers compete with data centers for the same grid capacity and must expect connection delays when expanding. It is advisable to report the power demand of planned expansions now and to build reserves with on-site storage or local generation.
- supplier
- According to Moody's, the US AI boom alone requires 110 billion dollars in new power plant capacity, and utilities must expect costly grid expansions. Long-term framework agreements are advisable to prepare for demand for power plant and grid equipment.
Hungary
The domestic footprint of the topic during the period examined is minimal. Only one domestic story relates to it: according to a Pénzcentrum report, Árpád Rab suggested limiting energy consumption in a critical situation. The article states that energy supply and data center capacity are already limiting the expansion of the AI industry. It is worth monitoring what connection conditions the domestic system operator sets for major AI loads.
Weekly trend
Turning points
Texas halts data centre grid connections as Nvidia and Amazon invest in power supplyReports say Texas is suspending data center connections, turning the limit from an abstract forecast into an administrative stop.FERC: US power grid still lags behind data center demandFERC states that grid development is not keeping pace with data center demand, and the narrative receives regulatory confirmation.Microsoft pulls out of Essex AI data centre project over power shortageMicrosoft exits an Essex project due to energy shortages, showing that limits are already cancelling concrete investments.US declares grid emergency in 13 states and national emergency to protect bulk power systemThe US Department of Energy declared a grid emergency in 13 states, followed by a presidential executive order declaring a national emergency to protect the high-voltage system. The issue has thus been elevated to the level of national security.
Related narratives in the News magazine
The AI investment bubbleSovereign AI: own capacity, own modelSocial backlash: children, schools, data centers
Supporting stories
Texas halts data centre grid connections as Nvidia and Amazon invest in power supplyenergy specialistConnection queues and investment in on-site generation point to new rules for allocating grid capacity.supplierProjects that need their own generation and storage are worth bidding for.engineerDesigning solutions that do not depend on a grid connection, such as local generation and storage, becomes more valuable.AI infrastructure spending to reach $725 billion as power becomes the bottleneckenergy specialistThe bottleneck lies in electrical capacity, so grid and generation projects are gaining value.supplierThe investment wave is directed at electrical and cooling products.Norway halts new grid connections over AI data centers' energy demandenergy specialistThe Norwegian move signals tighter regulation of grid connections.European grid bottlenecks are stranding billions in AI data center capitalenergy specialistBecause connection capacity is scarce, grid access must be secured at the start of an investment.IEA: data centres used 1.5% of global electricity in 2024, a fifth of planned projects may face delaysenergy specialistFor planned connections in clusters, grid delays must be built into project schedules and capacity planning.supplierA supplier serving data centre projects must account for the risk of project slippage in its order book.EIA: US power use to hit record highs in 2026 and 2027 as AI use surgesenergy specialistGiven record demand, data centre load must be treated as a separate scenario in generation and grid capacity plans.US electricity demand could rise up to 32% by 2030 because of AI data centresenergy specialistGrid capacity and reliability risks call for a longer horizon for connection and capacity planning.plant managerIndustrial power price and grid-connection risks should be factored into investment decisions.Studies: AI data centers mean a heavier grid-upgrade burden than previously expectedenergy specialistGrid-upgrade costs and forecasts point to rising connection charges and tariffs, which must be factored into capacity planning.plant managerGrid congestion may also affect industrial consumers' connection and electricity-cost plans.In Europe, land shortages and grid limits push data centres farther from citiesenergy specialistConstruction shifting toward rural and industrial areas creates new grid connection demand.plant managerIndustrial sites compete with data centres for grid capacity.AI data centres' power demand is redrawing the grid: capacity wall, flexibility and build paceenergy specialistGrid expansion takes years, so flexible load and load management borrowed from Bitcoin miners are the short-term tools.plant managerIndustrial power supply and data centre demand compete for the same capacity, so it is worth filing your own connection application early.supplierFor transformer, switchgear and generator makers, demand is set by the grid constraint.Grid connection becomes the key condition for AI data centres worldwideenergy specialistBecause of the new connection order and rules for large loads, grid development and connection plans must be aligned with readiness criteria.plant managerIndustrial plants' grid connection and capacity needs compete with those of data centres, so it is worth discussing expansion early in the connection process.SME ownerGrid scarcity and premiums raise electricity price and capacity risk, so long-term energy procurement is worth reviewing.FERC: US power grid still lags behind data center demandenergy specialistDelays in grid upgrades make it necessary to reassess connection plans and deadlines.plant managerWhen planning a new high-demand site or expansion, the time needed for grid connection must be factored in.Moody's: US AI boom requires $110 billion of new power plantsenergy specialistThe substantial need for new generation capacity argues for speeding up investment and procurement plans.supplierThe wave of power plant investment points to rising demand for turbines, transformers and construction capacity.Rab Árpád says energy use should be restricted in the current critical situationenergy specialistThe energy-restriction debate also raises the question of priority between data centre and industrial consumers, which needs preparation.Power supply problems delay launch of UK's largest AI supercomputerenergy specialistGrid connection determines the schedule of large data centre projects, so power supply planning must be settled at the start of the investment.supplierThe delay may require suppliers to renegotiate delivery schedules.US data centre connection requests top 700 GW, and the grid cannot copeenergy specialistGrid capacity has become a bottleneck, so connection requests and self-generation alternatives need to be scheduled early.plant managerIndustrial power consumers should review their own expansion plans in light of grid upgrade costs and connection queues.Power and skilled labour, plus cooling and GPU links, constrain AI data centre buildsenergy specialistAccess to power is the main constraint on project schedules.skilled workerA shortage of skilled construction and operations workers creates demand in data centre trades.engineerCooling and interconnect design is a key bottleneck for expansion.Sydney, Mexican and Israeli power grids hit limits as AI data centres expandenergy specialistIn several regions, delays in grid expansion limit AI data centre capacity.supplierDemand is growing for grid and storage solutions.US power grid constraints could upend Big Tech's AI plans, report saysenergy specialistGrid capacity may become the limiting factor for scheduling AI data centers, which calls for grid expansion planning.Microsoft pulls out of Essex AI data centre project over power shortageenergy specialistA concrete example of how a lack of power supply can halt a project.supplierOrders for data centre projects may fall through if access to energy is not secured.Microsoft pulls out of major UK AI data centre project over grid delaysenergy specialistGrid connection timing can sink a project, so the grid expansion schedule should be agreed early.supplierWith the investor pulling out, related orders may be rescheduled or cancelled, so it is worth covering this in contracts.US declares grid emergency in 13 states and national emergency to protect bulk power systemenergy specialistThe emergency measures may affect grid operating priorities and load management.plant managerThe power-supply risk at US plants is worth reassessing.Data centers accounted for at least 60% of US electricity demand growth last year, grid struggles to keep upenergy specialistGrid planning needs a separate scenario for data center load growth, and large connection requests should be handled in phased stages.plant managerGrid constraints may also slow the connection of factory expansions, so power requirements should be reported early.
Machine analysis based on story titles and summaries, with a verification pass. The source of the facts is the article linked at each story.