Daylila

Climate & Energy · Wednesday, 12 August 2026

01 · Briefing · what happened

AI's power hunger hits records, and the grid runs into a hard truth about "capacity"

Climate & Energy 4 min 13 sources

US electricity demand is set to break records again as AI data centres multiply, and utilities from Texas to Kenya are colliding with the gap between a power source's nameplate rating and what it actually delivers.

4,391bn

kWh US power demand by 2027

a new record, up from 4,195bn in 2025

474 GW

in Texas's grid queue

~90% data centres, 5x its peak demand

34%

of Kenya's peak mix from wind and solar

past the 15% many grids treat as a ceiling

5.4 GW

nameplate at Gravelines nuclear

three reactors knocked offline by jellyfish

At a glance

  • US power demand will hit records through 2027 (4,391 billion kWh), driven by round-the-clock AI and crypto data centres.
  • Texas Republicans floated forcing a set share of new supply to be gas that can ramp on command, not intermittent wind and solar.
  • Kenya's utility warned variable renewables now top 20% of grid capacity and 34% at peak, straining stability.
  • Jellyfish shut three reactors at France's 5.4 GW Gravelines nuclear plant, proving even 'firm' power delivers below nameplate.
  • Utah drew more power from solar than any source for the first time in May, near a third of its electricity.
  • The honest fix for the gap is storage; the government keeps old coal plants open while batteries scale.

Forces in play

AI power demand High

data centres drive US demand to records through 2027

Renewables on the grid Building

70% of new Texas supply, a third of Utah's power

Reliability worry Building

Kenya warns of instability; Texas eyes gas mandates

Storage catching up Easing

cheaper batteries and new rental models scaling fast

In play EIA — forecasts record US power demand through 2027 ERCOT / Texas lawmakers — weigh forcing a share of dispatchable gas onto the grid Kenya Power — warns rapid wind and solar uptake is straining the grid EDF — shut three Gravelines reactors after a jellyfish swarm

How it unfolded

  1. Aug 3 Texas governor pauses new data-centre approvals
  2. This week EIA forecasts record US power demand through 2027
  3. This week Kenya warns renewables are straining grid stability
  4. Mon night jellyfish shut three reactors at France's Gravelines plant

Where this points

Watch whether Texas actually mandates a share of gas capacity - a rule that would pay plants for standing by, not just for power delivered, and reset how every US grid values reliability.

Full briefing

American electricity use will smash records in 2026 and 2027, the government’s Energy Information Administration said this week [1]. The driver is AI and crypto data centres that run around the clock [1]. Demand is set to climb from a record 4,195 billion kilowatt-hours in 2025 to 4,391 billion by 2027 [1]. Meeting that hunger has reopened an old fight the transition kept quiet. The question is not how much power a source can make flat out. It is how much it can be counted on to deliver, hour by hour.

That gap is now shaping policy on three continents. In Texas, solar and batteries are winning on price, and renewables make up 70% of new generation coming to the grid. Republican lawmakers floated a rule forcing a set share of new supply to be “dispatchable” - gas plants that can ramp up on command, unlike wind and solar [2]. The grid operator’s own chief made the mechanism plain. ERCOT pays plants for the energy they actually deliver, not for standing by [2]. Cheap zero-fuel renewables supply all they can whenever the sun and wind allow, leaving gas to fill what’s left [2]. In Kenya, the national utility warned this week that variable wind and solar now top 20% of grid capacity [3]. That is past the 15% many operators treat as a comfort ceiling. At peak they can hit 34% of the mix, “exposing the national grid to system vulnerability when wind and solar suddenly dip or rise” [3].

The reckoning cuts against the “firm” sources too. In northern France, a swarm of jellyfish clogged the cooling intakes at Gravelines, one of Europe’s biggest nuclear stations [4]. Its six reactors are rated 900 megawatts each, 5.4 gigawatts in all. The swarm knocked three reactors offline and cut a fourth to half power overnight [4]. Nuclear normally runs flat out about 90% of the year. For a night, one of the continent’s steadiest machines delivered a fraction of its nameplate [4]. The lesson runs the other way for renewables, which is why the milestones need reading with care. Utah drew more power from solar than any other source for the first time ever in May [5]. That was nearly a terawatt-hour, about a third of the state’s electricity - a real achievement, and one that says nothing about 2am in January.

The demand driving all this is concentrated and enormous. Texas’s governor paused new data-centre approvals on August 3, prompting the EIA to cut its 2027 Texas demand-growth forecast from 14% to 6% [1]. The state’s grid queue holds about 474 gigawatts of interconnection requests - more than five times its record peak demand [6]. Roughly 90% are data centres [6]. Analysts estimate AI computing earns around $1.76 billion per gigawatt per month, so delay is expensive [6]. Amazon is now backing a gas-burning plant for a Texas data centre that critics warn could become the most polluting power station in the US [7]. It is part of a wider build-out whose air-quality toll researchers are only beginning to price [8].

The physics has a name in the science too. A Nature Geoscience study this week modelled “renewable energy droughts” - stretches of low wind and weak sun [9]. In a high-renewables China of the 2050s, fossil plants would cover those gaps. Even with the best filters, they could raise deadly fine-particle pollution 15-40% during the shortfalls [9]. The gap between what renewables are rated to make and what they deliver on a still, cloudy week is not an accounting quirk. It is a public-health variable.

None of this is an argument against clean power, which keeps winning on cost. It is an argument about how a grid is built. The honest answer to intermittency is storage - batteries that soak up cheap midday solar and pour it back at dusk. New models are spreading fast. An Austin startup now rents households an oversized backup battery for about the price of a streaming subscription, then leans on the fleet to steady the local grid [10]. Cheaper sodium-ion cells are closing on the energy density that made lithium the default [11]. But building enough of it takes time the demand curve isn’t offering. That is why the Energy Department keeps ordering old coal plants to stay open past their retirement dates [12]. And it is why the same government that slapped fresh tariffs on imported solar panels [13] is presiding over a renewables boom it can slow but not stop. The grid of the next decade will be built in the space between a power source’s rated capacity and the fraction of it you can actually bank on.

02 · Lesson · why it matters

Why a gigawatt of solar and a gigawatt of gas are not the same thing

A power plant's rated size is a promise it can only sometimes keep - a grid runs on the fraction it can bank on each hour.

How it works

  1. A power plant's nameplate is its flat-out maximum
  2. Its capacity factor is the fraction of that it averages over a year
  3. Solar ~15-25%, wind ~35-45%, nuclear ~90%+
  4. So a gigawatt of solar and a gigawatt of gas are not the same thing
  5. Intermittent sources need far more nameplate, plus firm backup, to do the same job

The twist

Headline capacity is a promise a power source can only sometimes keep; what a grid actually runs on is the fraction it can bank on hour by hour.

Where you've seen this

Hiring

a team's headcount isn't its output - the question is how many show up and produce on any given day

Restaurants

a 200-seat dining room's real capacity is the covers it turns on a normal night, not the chairs

Broadband

the 'up to 1 Gbps' on the box is the nameplate; your evening speed is the capacity factor

The catch

Capacity factor cuts both ways: it exposes solar's gaps, but it also caught nuclear this week when jellyfish did what no forecast priced in.

Full lesson

The number on the box

Every power plant carries a headline number. A gas plant is “one gigawatt.” A solar farm is “400 megawatts.” A nuclear reactor is “900 megawatts.” That number is the nameplate: the most the machine can push out when everything is running flat out.

It is the number in the press release, the target, the pledge. Utah just drew more power from solar than any other source for the first time. Texas has the largest fleet of renewables in the country. These are real milestones, and they are measured in nameplate.

But the nameplate is a ceiling, not an average. It is what the machine can do at noon in June with the sun overhead - not what it does at 2am in January, or on a still, grey week. And a grid does not run on ceilings. It runs on what actually shows up, every hour, all year.

The fraction that matters

The honest number is the capacity factor: the share of its nameplate a plant averages over a full year. It is the gap between the promise and the delivery.

Solar sits around 15 to 25%. The sun sets, clouds pass, winter is dim - so a solar farm delivers roughly a fifth of its rated size across a year. Wind runs about 35 to 45%; the wind is stronger and blows at night, but it is fickle. Nuclear runs above 90%. It ignores the weather and the clock, humming at nearly full output nearly all the time.

So the arithmetic bites. A gigawatt of gas and a gigawatt of solar carry the same headline. Over a year, the solar delivers about a quarter as much energy - and none of it at night, exactly when demand can still be high. To do the same job as one gigawatt of steady gas, you need several gigawatts of solar nameplate, plus something to hold the power for the dark hours.

Why the whole grid fight is this fight

Once you see the capacity factor, the week’s news stops being a jumble and becomes one argument.

Texas lawmakers want to force a share of new supply to be “dispatchable” - plants that ramp up on command. That is not hostility to renewables. It is the capacity-factor problem written into law: the grid needs power on demand, and cheap solar, for all its nameplate, cannot promise a specific hour. Kenya’s utility hit the same wall from the other side, warning that variable wind and solar crossing a fifth of its capacity was straining the grid’s stability.

Even the steady machines only mostly deliver. This week a swarm of jellyfish clogged the intakes at one of France’s biggest nuclear stations and knocked three reactors offline overnight. Nuclear’s 90-plus capacity factor is a yearly average, not a guarantee. No source delivers its full nameplate all the time. The whole design of a grid is the art of stacking imperfect promises so that, together, the lights stay on.

Where else the gap hides

The trap - reading a ceiling as an average - is everywhere once you have the shape.

A restaurant’s real capacity is not its 200 chairs. It is the covers it turns on a normal night, with the tables that sit empty and the kitchen’s limits. A team’s output is not its headcount; it is how many people show up and produce on a given day. Your broadband is sold as “up to a gigabit” - the nameplate - while your evening speed, when the whole street is streaming, is the capacity factor.

In every case the headline number flatters. It describes the best possible moment and quietly implies it is the normal one. The useful question is never “how big is it,” but “how much of that can I count on, and when.”

The whole, held loosely

We are all inside this gap, whether or not we ever read a grid report. The price of electricity at dusk, the pollution on a windless week, whether the AI boom runs on gas or sun - all of it turns on one distance. It is the gap between what a plant is rated to make and what it delivers.

The reckoning is not clean power failing. It is a grid learning, in public, that reliability is a portfolio, not a headline. Solar’s low capacity factor is a fact to build around, not an insult - just as nuclear’s high one is a boast the jellyfish can still puncture. The people running the wires know they cannot see it all from any one seat. The forecast misses the still week; a plant trips; the demand curve outruns the batteries. Seeing the whole here means holding every big round number a little more loosely, and asking, always, what fraction of it is real.

03 · Lab · your turn

Power the grid

Build a grid to meet steady demand and feel the gap between a source's nameplate rating and what it actually delivers hour by hour.

04 · Hope · carry this

A working grid is a thousand imperfect promises stacked until the lights stay on. That we mostly manage it, hour after hour, is quiet proof of how much steady competence holds the world up.

Across the beats