Trends›The grid keeps pace with AI
The grid keeps pace with AI
Generation, baseload (nuclear/SMR/gas) and grid/transmission build out fast enough that energy infrastructure does not hard-cap datacentre construction through 2030.
22 confidence, computed 23 Oct 2026
- Environmental
- High weight
Rests onfacts underneath
What rests on thismarkets it underwrites or threatens
Underwrites
Evidencethe schedule the figure draws
The evidence, newest first
- Jul 2026
against · primary Private source - Jul 2026
for · primary Private source - Jul 2026
against · primary Musk's APR Energy acquisition is a hyperscaler routing around the grid entirely: trailer-mounted gas turbines chosen because they deploy in days with no interconnection queue, while the nuclear PPAs the rest of the industry signed (Microsoft/Constellation Three Mile Island, Google/Kairos… The Once and Future Kimi (State of the Future) - Jun 2026
against · primary US industry will add less than 10GW of gas per year in 2026 and 2027, with nameplate capacity shortfall requiring behind-the-meter solutions. US Grid Constraints: Towards 40GW+ of Behind-The-Meter Datacenter by 2028? - Jun 2026
against · primary Customers placing an order for a gas turbine at Siemens Energy today will wait up to four years for delivery. Siemens Energy's Record Backlog Creates Four-Year Wait for Turbine Delivery - Jun 2026
against · primary Grid supply cannot keep pace with AI DC load on the SemiAnalysis five-layer stack: Layer-1 grid-connected reliability headroom falls 18.3GW (2025) to negative in 2027 and a ~-40GW deficit by 2030, interconnect queues run ~7yr (PJM/N.Virginia), and industrial constraints (GOES… SemiAnalysis — To Boldly Go: The Case for Space Datacenters (Space DC TCO) - Jun 2026
for · observable Private source ~9.8 GW hyperscaler nuclear committed across 13 deals (SMRIntel) - Jun 2026
neutral · observable Announced agreements to date for nuclear power could provide up to 13 GW total, split roughly equally between PPAs and direct partnerships, though deployment timelines remain uncertain. Beyond the Hype: Assessing Hyperscaler Nuclear Commitments Against U.S. Energy Realities - May 2026
against · primary Data center incremental demand reaches 13.6 GW in 2026; planned generation additions limited relative to demand in Mid-Atlantic, Mid-Continent, Northwest. US Data Center Power Demand Projected to Double by 2027 - May 2026
against · primary Over 300 GW of generation queued in PJM and ERCOT, but data center projects face 3-4 year waits with only 60% (ERCOT) to 76% (PJM) of queued projects unlikely to come online. AI Meets the Grid: Interconnection Queue Analysis in PJM and ERCOT - May 2026
neutral · observable Fuel cells 40% higher LCOE than simple-cycle gas turbines at current gas prices; Bloom claims scalability but manufacturing expansion still years behind demand. Behind-the-meter gas turbine workaround cannot scale: fuel cells offer potential but at 40% higher LCOE than gas turbines currently - May 2026
against · primary As of end-2025, 2,060 GW of generation/storage capacity actively seeking interconnection; median IR-to-COD time has doubled to over 4 years for 2018-2024 projects. U.S. Interconnection Queue Data Through 2025: Complete Interconnection Request Dataset and Summarized Data Workbook - Apr 2026
against · primary As of 2026, US interconnection queue swelled to 2,600 GW backlog with median wait time approaching 5 years; data centers facing delays up to 12 years. Grid Interconnection Delays 2026: A Threat to US Energy - Apr 2026
against · primary Production slots through 2030 remain limited at roughly 10 GW of annual capacity, which means the 100 GW backlog represents a decade of output. GE Vernova Gas Turbine Backlog Reaches 100 GW, Driven by Data Center Orders - Apr 2026
against · primary GE, Siemens, and Mitsubishi are each reporting order backlogs that stretch roughly five years, with many delivery slots already sold into the next decade. The Gas Turbine Bottleneck Reshaping Energy Infrastructure - Mar 2026
for · observable Google paid $4.75B for Intersect Power (2.2 GW operating solar + 2.4 GWh storage) to build behind-the-meter solar+long-duration-storage energy parks that power datacentres directly, using the grid only as a backstop — routing around interconnection queues. Google acquires Intersect Power $4.75B for behind-the-meter solar+storage 'energy parks' - Mar 2026
against · observable Near-term firm power is met by GAS (Microsoft/Chevron ~5 GW West Texas, Google/Crusoe 933 MW, Meta Hyperion toward 7.46 GW), pushing Microsoft emissions +30%+ since its 2020 pledge and a ~195% gas-turbine price spike — the 'firm' bridge is gas… Hyperscaler gas bridge vs net-zero: +emissions, ~195% turbine price spike (Stand.earth) - Feb 2026
for · claim NRC docketed Holtec's Part-1 construction-permit application for two ~340 MWe SMR-300 units at Palisades — the first US SMR construction-permit in NRC technical review, but first power is ~2030 at the earliest (Mission 2030), delivering no firm MW in the… NRC dockets Holtec's twin Palisades SMR-300 construction-permit application - Feb 2026
against · primary Average interconnection wait times 5 years; grid limitations are the defining constraint on whether new datacentres can come online at all. Interconnection queue and grid constraints hard-capping datacentre construction timelines - Jan 2026
for · observable Developers plan to add a record 86 GW of utility-scale capacity in 2026 (43 GW solar, 24 GW battery, 6.3 GW gas), up from 53 GW in 2025 — the largest single-year build since 2002, a fast aggregate supply response. EIA: New US generating capacity to reach record high in 2026 - Jan 2026
against · observable Of 53 GW added in 2025, ~34 GW was solar and ~16 GW battery while firm capacity was only ~5.7 GW gas and 60 MW nuclear — the firm 24/7 capacity datacentres need is a small fraction of the headline… EIA 2025 additions: 34 GW solar, 16 GW battery, only 5.7 GW gas - Jan 2026
neutral · observable IEA projects renewables expanding ~1,000 TWh/yr to 2030 (solar >600 TWh) and record nuclear, ~half of generation by 2030 — but grid investment must rise ~50% from ~$400B/yr, with 1,200-1,600 GW of projects stalled on grid build-out. IEA Electricity 2026: renewables+nuclear ~half by 2030 but grid capex must rise 50% - Dec 2025
against · primary Grid modernization identified as key constraint on energy security and competitiveness, with AI's surging power demand testing grid limits in 2026. S&P Global Outlines 2026 Clean Energy Pressures Driven by AI Growth, Geopolitical Shifts - Dec 2025
against · primary Supply constraints push much new gas capacity into late 2020s and early 2030s; power crunch most severe through 2028-29 when supply begins catching up. Gas turbine capacity constraints: delivery windows pushed 5-7 years, peak power crunch 2028-29 - Oct 2025
against · observable Private source ERCOT large-load queue 56→205 GW in a year (Texas Tribune) - Oct 2025
against · primary GE Vernova's gas-turbine backlog plus slot reservations reached ~62-80 GW with deliveries into 2029 and slots sold out through 2030 — firm new-build gas is supply-chain-capped for the rest of the decade regardless of capital. GE Vernova gas-turbine backlog ~80 GW stretching into 2029 (Utility Dive) - Oct 2025
against · observable IEA: data-centre electricity demand rose 17% in 2025 (AI-focused climbed even faster) vs ~3% global demand, flagging physical bottlenecks 'limiting the rate at which data centres can expand' with grid connections held up, pushing developers to on-site gas. IEA: data-centre electricity use surged 2025, grid struggling to keep pace - Jul 2025
against · primary The US interconnection queue holds >2.2 TW with average time-to-COD ~5 years, and historically only ~19% of 2000-2019 requests reached commercial operation by end-2024 — too slow to absorb tens of GW of new high-utilisation load by 2030. RMI/LBNL: ~2.6 TW interconnection queue, ~5yr median to COD - Jun 2025
for · observable Constellation targets ~2027 interconnection for the 837 MW Crane/TMI-1 restart (a year early) under a 20-year Microsoft PPA, backed by a $1B DOE loan — an existing-reactor restart delivering firm baseload faster than new build. Three Mile Island (Crane) restart ahead of schedule, ~2027 (DCD) - Jun 2025
for · primary Unsubsidised utility-scale solar fell ~4% YoY to US$38-78/MWh (midpoint ~$58, an 84% drop since 2009) and battery LCOS fell to $93/MWh (from $104 in 2024, $155 in 2023), while new combined-cycle gas hit a 10-year-high LCOE — solar+storage is the… Lazard 2025 LCOE+: solar cheapest, storage LCOS falling - Jun 2025
for · observable GE Vernova booked 29 aeroderivative turbines (~1 GW) for Crusoe/Stargate as on-site bridge power for 5-7yr until grid interconnection arrives — the behind-the-meter release valve letting construction proceed despite gridlock. - Jun 2025
neutral · claim AWS secured a 1.92 GW Susquehanna nuclear PPA and the big-four contracted >10 GW of nuclear in a year, but all credible new SMR online dates (Kairos/Google, TerraPower, X-energy) remain 2029-2035 — firm nuclear helps post-2030, not the near-term crunch. AWS-Talen 1.92 GW nuclear PPA; >10 GW nuclear signed but online 2029-35 - Apr 2025
neutral · primary IEA Base Case projects generation to data centres roughly doubling from ~460 TWh (2024) to >1,000 TWh by 2030, with renewables meeting ~half the additional demand and the firm baseload bridge being gas (not nuclear) before 2030. IEA Energy and AI: supply to datacentres doubling to >1000 TWh by 2030
What would move itthe measures watched
- watchUS firm generating-capacity additions (gas+nuclear GW/yr) vs datacentre incremental demand. Record 86 GW planned 2026 but most is intermittent solar+battery; only ~6 GW firm gas. Track the FIRM-capacity share against ~10-15 GW/yr incremental hyperscale demand. Firm additions lagging = the aggregate number is misleading.
- watchGas-turbine OEM backlog depth / earliest available slot (years out). GE Vernova backlog stretches to 2029-2030, slots tightening. Rising backlog with no slot relief = firm new-build is supply-chain-capped regardless of capital. Watch OEM output ramp closing the gap.
- watchInterconnection-queue volume (TW) and median time-to-COD (years). ~2.6 TW in queue, median ~5yr to COD, historically only ~13% reach COD. Falling median + FERC Order 2023 throughput = bull; queues swelling (ERCOT 56→205 GW) = bear. The single best leading indicator.
- watchBehind-the-meter / on-site gas + nuclear PPA capacity contracted (GW). Rising BTM/on-site volume is the RELEASE VALVE that lets construction proceed despite grid gridlock — bull for 'keeps pace' but signals the grid itself is NOT keeping pace. The workaround, not the resolution.
- watchDatacentre projects explicitly delayed/curtailed for lack of power. Rising explicit power-driven delay events = direct falsification of 'does not hard-cap'. The realisation metric the whole assumption turns on.