In Qatar, the world's largest LNG exporter is racing to lift output. In Japan, a new gas turbine can take up to seven years to arrive. The bottleneck for AI is no longer the chip. It is the machine that makes the power.
Gas turbines are sold out. GE Vernova now holds 100 gigawatts under contract, with only about 10 gigawatts of 2029 and 2030 slots left. The company is already booking reservations for 2031. Roughly a fifth of that contracted volume is tied directly to data centre load.
So AI campuses have stopped waiting for the grid. They are building their own power.
The opening problem is not silicon. Gas turbines are sold out through 2030

The chip is not the constraint. The generator is. GE Vernova reports its gas turbine slots are effectively sold out through 2030. About 80% of its contracted volume serves utilities and industry. The other 20% is data centre load.
The turbine itself is not always the gating item. Heavy-duty orders run about three years, but permitting, construction and fuel add more. Across the industry, manufacturer backlogs now stretch through 2030. The wait, not the design, sets the schedule.
Traditional grid planning fails when a turbine takes seven years to arrive

The old sequence was simple. Order a turbine, wait for a grid connection, build the plant. That sequence has broken. The wait for a large gas turbine has risen from about three years to as long as seven.
Demand is broad and global. Siemens Energy booked 14 gigawatts of gas turbine orders in a year, and 65% were for data centres. Mitsubishi is doubling its turbine capacity. Even so, one expert warned the expansion will not drastically cut wait times. Grid connection queues are long as well, so both clocks run at once.
The engineering response is behind-the-meter gas. US proposals tripled to 252 GW

Campuses answered the queue by self-generating. In 2025, the United States nearly tripled its gas capacity in development to 252 gigawatts, overtaking China. More than a third is designed to power data centres on site, behind the meter.
The deals are concrete. ExxonMobil and NextEra will build a 1.2 gigawatt gas plant with carbon capture and market it to a hyperscaler. NextEra calls the model one where data centres essentially fund their own generation. Texas alone has 80 gigawatts in development, half of it for data centres.
If grid power and a turbine slot are both years away, what is your next move?
The strategic disconnect: 252 GW is proposed, yet turbine backlogs run to 2030

Proposals and delivery have pulled apart. Global gas capacity in development rose 31% in 2025, to 1,047 gigawatts. Yet the same tracker warns the buildout is bottlenecked by turbine production, with backlogs running through 2030.
Three makers, GE Vernova, Siemens and Mitsubishi, hold more than 75% of the named-manufacturer market. Two-thirds of proposed projects have no named turbine supplier at all. Many will slip or stall. If demand disappoints, some become stranded assets.
Policy fast-tracks gas connections. It cannot mint a turbine

Regulators have moved to speed dispatchable power onto the grid. Fast-track rules now favour shovel-ready gas projects in some markets. The paperwork can be accelerated. The forging cannot.
This is a manufacturing constraint, not a permitting one. China narrowly leads the world in gas capacity actually under construction, at about 31 gigawatts. A permit does not build a turbine, and a national target does not shorten a seven-year queue.
LNG is the fuel behind the bridge. Qatar lifts capacity from 77 to 142 MTPA

Self-generated gas needs fuel, and that supply is expanding. Qatar is raising LNG capacity from about 77 to 142 million tonnes a year. Its anchor North Field West phase, 16 MTPA across two mega-trains, has slipped to first cargo in 2031. The fuel timeline now echoes the turbine one. In the United States, Golden Pass shipped its first cargo in April 2026, the ninth US export terminal.
The scale is easy to miss. Here is a rough rule of thumb. One million tonnes of LNG a year can back nearly a gigawatt of continuous power. Plant efficiency sets the exact number. Qatar's expansion alone could underpin tens of gigawatts of new generation.
The path forward pairs dispatchable gas with the nuclear baseload being built

Gas is the bridge, not the destination. It covers the gap now, while slower, cleaner baseload is built behind it. Dispatchable gas answers a five-year need. Nuclear answers a thirty-year one.
The discipline is to sequence both. Secure the turbine slot and the fuel for the bridge. Plan the handover to baseload as reactors and small modular units arrive. Intelligence can model that sequence. Expert judgement still commits the capital and the machine.
The supply side is stressed too. Europe's 2026 drought cut river-cooled power

The demand side is not the only pressure. In summer 2026, Europe's rivers fell to record lows. By early August, the Rhine reached its lowest recorded level in Germany and at Lobith, and the Danube its lowest in 30 years. Hungary's only nuclear plant ran at reduced capacity. In Romania, crews blasted a Danube rock to keep cooling water flowing to the Cernavoda reactor. The World Weather Attribution group links a warming climate to higher evaporation and more frequent drought. When rivers cut nuclear and hydro output, the pull toward dispatchable gas grows, even as the turbine queue stays full.
The investment implication: the turbine slot is now the scarce asset

Capital is not the binding constraint. The slot is. Gas turbine prices are heading toward £451/kW ($600/kW, €528/kW) by the end of 2027, nearly triple their 2019 level. Building all the gas now proposed in the United States would cost above £313bn ($416bn, €366bn).
Building all of it would grow the US gas fleet by roughly half. The money is available. The machine is not. Two-thirds of proposed projects still lack a named turbine maker. Underwriting that prices only the megawatt, and not the slot and the fuel behind it, understates the real risk.
The bottom line…

The chip was never the ceiling. The turbine is. Sold out to 2030, so campuses self-generate, LNG bridges the fuel, and nuclear waits behind as the destination.
From Qatar to Texas to the Rhine, the lesson repeats. Dispatchable power, and the machine that makes it, is the scarce input. Intelligence can plan the sequence. The turbine still has to be built.
Next week: Japan restarts its biggest reactor. The chipmakers were already waiting.
The GTM Playbook Behind Warmly's Acquisition
Warmly ran pipeline, outreach, and lead scoring on autopilot for hundreds of startups — before a single sales hire.
HubSpot acquired them for it. Now the cofounders are walking you through the exact system, live, before they disappear into product. Join the Builder Session on August 12.
Leave with an agentic GTM stack you can replicate this week. Plus HubSpot Credits when you join HubSpot for Startups.



