Power & Grid · Daily Issue

The Thermostat Army: Google Just Bought a Power Plant Made of Other People’s Homes

While oil spiked and power stocks wobbled, the week’s smartest grid deal built no steel at all — it rented flexibility, by the gigawatt.

Isa Chen, Head of Editorial Systems

October 24, 2026

TL;DR — While oil spiked and power stocks wobbled, the week’s smartest grid deal built no steel at all — it rented flexibility, by the gigawatt.

The opening paragraph, up front — the full story follows below.

~9 min read

Executive thesis: The grid’s scarce product is no longer electricity in general — it is dependable capacity at the exact hour everyone wants it. This week proved the point three ways at once: crude priced a supply scare, gas priced a mild autumn, and Google paid households and businesses to become a power plant. Whoever can deliver flexible megawatts fastest captures the value; whoever waits years for steel captures the delay.

What Needs Your Attention Today

Verdict: one danger light, two warnings, and one genuinely hopeful build. Oil’s jump raises the cost of everything that burns fuel. Borrowing costs are knocking on a round number that changes budgets. But the largest new “plant” announced in the country’s biggest grid this week is a network of thermostats, batteries and factory controls — and it arrived without a single construction permit for a power station.
So what? If you pay a power bill in the mid-Atlantic, flexibility is now something you can sell. If you invest in power, the fastest-growing asset class this week was not generation — it was the ability to use less, on cue.

[DANGER]

Brent crude, Friday settlement

$116.72 a barrel, up 3.14%

Highest close since late April after attacks in Saudi Arabia; up 6% on the week

[WARNING]

Cost of borrowing

10-year Treasury at 4.99%

A whisker under 5% — the level where power-project financing gets repriced

[WARNING]

AI power stocks, five days

Constellation −7.84% · Vistra −9.09%

Investors are questioning how fast AI spending turns into power contracts

[WATCHING]

A power plant made of homes

1,070 MW of flexible capacity added

Google and Voltus expand their shared virtual plant in the PJM grid

[WATCHING]

Winter gas cushion

3,644 billion cubic feet in storage

Comfortably inside the normal range heading toward heating season

[WATCHING]

The grid’s fuel got cheaper

Natural gas $2.626 per MMBtu, down 6.08%

Mild weather beat geopolitics in the one market power plants actually burn

Green means watching — tracked and normal. Amber means warning — look closer. Red means danger — act. Every card states its meaning in words, not colour alone.

60-Second Regime Matrix

Friday closed a week of two tapes. Anything tied to a barrel rose; almost everything else fell — and the split inside the stock market tells you which fear is winning hour by hour.

Friday, October 23	Close	Move
S&P 500	7,623.68	−0.52%
Dow Jones Industrial Average	52,215.16	−1.23%
Nasdaq Composite	25,955.34	−0.21%
Volatility gauge (VIX)	19.45	+2.21%
WTI crude	$103.27	+3.8%
Gold	$4,007.90	−1.29%
S&P 500 energy shares	Record high	+2.5%

Read it plainly: oil producers had their best day while the average stock fell. That is a supply scare, not a boom — buyers are paying for barrels today, not for growth tomorrow. Sources: Reuters and Associated Press closing reports, October 23, 2026.

The Macro Vector — Demand Is Becoming a Business Story

Here is a milestone that would have sounded odd a decade ago: this year, American businesses are on track to buy more electricity than American homes for the first time. The Energy Information Administration expects total power use to climb from last year’s record to new highs this year and next, with shops, offices, factories — and, fastest of all, data centres — doing the growing while home use stays roughly flat.

That shift changes who the grid is built for. A household’s demand peaks on a hot evening and forgives delay. A computing campus runs around the clock and signs twenty-year contracts. McKinsey’s latest global outlook, a scenario rather than a promise, has data-centre electricity demand growing at 24% a year through 2030 — and warns that wires, transformers and connection queues, not power stations alone, set the real speed limit. Its Asia work makes the same point at continental scale: the region may need to spend on the order of a trillion dollars a year on energy for years, because new demand is arriving faster than the systems meant to serve it.

What is priced in: wholesale power is already expected to average $52 per megawatt-hour nationally this year, up 11% on last year — and up 41% in PJM, the mid-Atlantic grid operator’s territory. What would prove this wrong: a year of flat business demand in the official sales data — the milestone would quietly un-happen. Sources: EIA Short-Term Energy Outlook via Reuters, October 6, 2026; McKinsey Global Energy Perspective 2026 (scenario context).

Dispersion — The Market Just Split Power Into Two Trades

Thursday delivered the week’s most useful information. When Meta reported $34.9 billion of quarterly capital spending and negative free cash flow of $2.1 billion, investors did not sell “power” — they sold a specific kind of power. GE Vernova fell 10.6% in a single session, Vistra 9.8%, Bloom Energy 8.7%, Quanta Services 8.1% and Constellation 5.7%. The selling eased Friday — Constellation closed at $253.45, down 0.69%, and Vistra at $140.16, down 0.48% — but the week’s losses stayed.

Meanwhile the regulated giant next door barely blinked. NextEra Energy reported adjusted earnings of $1.24 a share, ahead of the $1.20 consensus, on revenue of $8.09 billion; its Florida utility’s profit rose 10.7% to $1.62 billion, it added 3.7 gigawatts of renewable and battery projects to a backlog now near 36.7 gigawatts, and it confirmed full-year guidance of $3.92 to $4.02 a share. Its stock closed Friday at $82.15, up 0.30%. Same industry, same week, opposite experience.

The lesson is about where the risk sits. A regulated utility earns an approved return on what it builds, so an AI spending scare is mostly weather. A merchant generator earns whatever a tight market will pay, so the same scare lands directly on its price. Neither is “better” — but they are no longer the same trade, and this week the gap became impossible to ignore. Sources: company results reported October 21, 2026 (Zacks, MarketBeat); CNBC move summary via market reports, October 22–23, 2026.

Tactical Asymmetry — What to Watch Next

The capacity bill comes due. PJM’s auction that guarantees plants will be available three years from now cleared at $325.00 per megawatt-day for 2028/29 — 11.2 times the $28.92 of 2024/25 — and the grid’s independent monitor attributes about 63% of recent capacity costs ($9.33 billion of $14.69 billion) to data-centre load. Watch whether states move first: the Federal Energy Regulatory Commission this month approved a registry for users of 50 megawatts or more (in force since October 12) but declined to make them bring their own power or accept first-in-line cuts, and it has paused a separate 6.8-gigawatt emergency procurement until February 28, 2027. PJM published results; FERC order, October 9, 2026.
The slow lane is very slow. Three retired nuclear plants are being brought back with up to $4.4 billion of federal loans — 2.25 gigawatts in total: Palisades in Michigan (800 MW, around March 2027), the Crane plant on Three Mile Island (835 MW, targeted 2027, supplying Microsoft under a 20-year deal) and Duane Arnold in Iowa (615 MW, by 2029, with Google). Set against grid connection waits of three to six years, that timing gap is exactly the opening on-site fuel cells are exploiting: Reuters reports one supplier delivering in under two months, and Rystad Energy sees fuel-cell spending for data centres rising from $2.8 billion in 2025 to about $30 billion by 2030, with roughly 40% of projected 2030 capacity potentially choosing on-site power. Associated Press, October 16, 2026; Reuters explainer, October 23, 2026.
The winter cushion has a soft spot. The latest report added 77 billion cubic feet against expectations near 80, leaving stocks 152 billion below last year but 35 billion above the five-year average of 3,609. The soft spot is regional: south-central storage sits 10.0% below last year, and its salt caverns — the fastest-flexing stores, closest to export terminals — lag their norms, so a cold snap would bite there first. The official outlook still expects storage to end October at 3,850 billion cubic feet, 2% above average. EIA Weekly Natural Gas Storage Report, week ended October 16, 2026.
Two dates carry the week ahead. The Federal Reserve decides rates on Wednesday, October 28, with futures pricing a 92% chance of no change — the risk is the language around energy-driven inflation, not the decision itself. The same morning, GE Vernova reports; after a 10.6% fall on Thursday, its order book is the cleanest public read on whether the building boom is accelerating or merely expensive. Federal Reserve calendar; CME FedWatch via market reports; company reporting schedule.

Value Creation vs Value Capture

End the week with a picture. Somewhere in Ohio, a family’s thermostat eases off for twenty minutes on a strained afternoon. They get a small payment. Multiplied across hundreds of thousands of homes and warehouses, that small act is now a gigawatt — the size of a respectable power station — and a technology giant is paying for it because it is faster than building anything.

That is where the value is moving. Creating electricity still matters enormously; the nuclear restarts and the record order books are real. But this week, capture belonged to whoever could be flexible: the Brattle Group estimates better grid use could save consumers more than $100 billion over the next decade, Duke University researchers put the capital saving from flexible data centres at $40 billion to $150 billion, and OpenAI has already agreed to cut up to 1 gigawatt from a planned 3.2-gigawatt Georgia campus when the grid calls. The request queues tell the other half of the story — Texas alone is tracking about 474 gigawatts of large-load requests, roughly 90% of them data centres, against a national request tally above 1,000 gigawatts that analysts say will mostly never be built. The scarce skill is choosing, sequencing and flexing what already exists.

The takeaway: the grid’s next decade will be won less by who owns the most megawatts than by who can move megawatts through time. Watch the flexibility deals. They are small, fast, and — as of this week — officially gigawatt-scale.