The Grid Space Nobody Mapped — 3 Tech Giants Need 100 GW ⚡

100 GW. That is every light, factory, and AC unit in France running at the same second. Google, Nvidia, and Anthropic need to find that much spare room on the US power grid — and they just hired a startup to locate it on a map nobody has drawn yet.

Three of the richest companies on Earth. One startup named Emerald AI. One target: 100 gigawatts of unused electricity.

The plot twist nobody expected: the thing slowing down AI isn’t chips. It isn’t money. It’s finding a wall socket big enough. And Google can’t find one.

Google, Nvidia & Anthropic Are Hunting for the Power Output of France

:electric_plug:

They have the money. They don’t have the map. So they hired someone to draw it.

Here’s the whole story in one breath. AI data centers — the giant warehouses full of computers that run ChatGPT and Claude — eat electricity like nothing before them. So much that the three biggest names in AI just teamed up with a startup whose only job is to find where on the grid there’s still room to plug in.

The number they’re chasing is 100 gigawatts. That’s roughly the entire electrical output of France.

ChatGPT was asked to help locate it. It crashed.

🧩 Dumb Mode Dictionary — 6 words, plain English
Term What it actually means
Gigawatt (GW) A unit of electric power. One GW runs about 750,000 homes. So 100 GW ≈ 75 million homes.
Power grid The giant web of wires that carries electricity from power plants to your plug.
Data center A warehouse packed with computers. This is where “the cloud” and every AI chatbot actually live.
Grid capacity How much extra power a spot on the grid can still deliver before the wires max out.
Interconnection queue The government waiting list to get a new power line hooked up. Takes 3–7 years. Yes, years.
Emerald AI The startup Google, Nvidia & Anthropic backed. It maps where spare power is hiding.
⚡ Wait — the bottleneck is a POWER SOCKET?

Everyone assumes the AI race is about chips (Nvidia) or cash (everyone). Both true, a year ago.

Now the wall is physical. You can buy all the chips you want. If there’s no spare electricity where you want to build, the chips sit in a box.

  • Chips? Available if you pay.
  • Money? These three have more than most countries.
  • Somewhere to plug in 100 GW? Nobody knows where it is.

That last line is the entire reason Emerald AI exists.

🗺️ Who is Emerald AI and why are THEY the story

Google has the money. It doesn’t have the map. Emerald AI sells the map.

Their whole job is scanning the grid to find spots that can take a new data center now — instead of waiting 3–7 years for fresh power lines. They squeeze more out of what’s already built.

Finding parking spots for God, basically.

The backers say the quiet part loud: this is a coalition to unlock grid space, not build new plants. The power mostly exists. It’s just unmapped and unclaimed.

⏳ The waiting list that changes everything (the interconnection queue)

Want to connect a big new load to the grid? You join the interconnection queue — the government waiting list to get a new line approved.

It takes 3 to 7 years.

By the time your data center is connected, the AI running inside it has three successors.

So the game flips. Nobody wants to wait for new power. Everybody wants land that’s already connected — old factories, closed plants, dead mines with fat power lines still attached. That real estate just quietly got more valuable, and most owners have no idea.

The best part? The entire queue is public data on FERC.gov. Free. Sitting there. Barely read.

🇫🇷 The receipts — what 100 GW really is
The number In human terms
1 GW ~750,000 homes powered
100 GW France. All of it. Every light on at once.
3–7 years Time to get a brand-new grid connection approved
$0 Cost to download the FERC waiting list nobody reads

France. They need France. :high_voltage:

💬 Overheard in the meeting (a joke, obviously)

Exec 1: So we’ve got the chips, the money, the demand curve — we’re ready to build.

Exec 2: Where do we plug it in?

Exec 1:

Exec 2: Call the startup. The one with the map.

Cool. So AI Just Ran Out of Outlets. Now What Do We Do?

( ͡° ͜ʖ ͡°)

Searching for a plug outlet

Here’s the secret the suits are paying millions for: the data is public. FERC filings, state utility reports, county permits — all free, all online, almost nobody reading them. Emerald AI serves the Google-sized clients. That leaves the entire small-and-medium world wide open. You don’t need an engineering degree. You need a spreadsheet and the nerve to filter it. Between you and me, this window is open right now because the mainstream news still thinks this is a “chips” story.

🕳️ The Abandoned Queue Grab

Projects drop out of the interconnection queue all the time — they get funding, then die. When they leave, they leave behind a known chunk of grid capacity that’s suddenly free. Land buyers don’t know this list exists.

Here’s what you do: pull the FERC generator interconnection queue, filter for slots marked “withdrawn” with big capacity, and sell that filtered list to land developers hunting for grid-connected sites.

:brain: Example: A researcher in Lisbon downloads the Texas queue CSV, filters for withdrawn slots above 50 MW, and sells the clean spreadsheet to three land-development firms for €400 each.

:chart_increasing: Timeline: First sale within 2–3 weeks. Window stays open until the data brokers wake up — call it 6–12 months.

📡 The Pre-Demand Zone Brief

Land near an underused substation reprices before the news hits that zip code. But the substation capacity numbers are already sitting in state utility filings today.

Here’s what you do: pull one state’s utility commission capacity report, mark every substation with big unused headroom, and sell a one-page county brief to rural land investors before the wave reaches them.

:brain: Example: A mapper in Nairobi pulls the Texas PUC and CAISO reports, flags counties with spare substation capacity, and sells $200 county briefs in a rural-land Facebook group.

:chart_increasing: Timeline: First brief sold in a week. The good zones get bought within a year — move early.

🔗 The Stranded Watt Finder

Old factories, closed mines, and dead power plants often keep their big grid connections. Worth a fortune now, because a new connection is 3–7 years away. The owners usually don’t know what they’re sitting on.

Here’s what you do: dig through state decommission filings, find these properties, and introduce the owners to data center developers who need power yesterday. Take a finder’s fee.

:brain: Example: A researcher in Manila searches Ohio utility filings for “plant decommission” (2018–2024), finds a shuttered site with a live 80 MW connection, and brokers the intro for a mid-five-figure fee.

:chart_increasing: Timeline: Slower burn — 1–3 months to a first intro — but a single deal can pay more than a year of the others.

📖 The Grid Atlas Flip

Grid operators like CAISO, PJM, and ERCOT publish capacity maps that are basically unreadable unless you’re a grid engineer. Mid-size companies that need 1–10 MW can’t read them and can’t afford Emerald AI.

Here’s what you do: translate one operator’s public map into a plain-English regional atlas PDF. Sell it to manufacturers and small hosting companies who need power but can’t decode the source docs.

:brain: Example: A freelancer in Kraków turns CAISO’s transfer-capability map into a plain-language 6-page PDF and sells it for $500 a pop to colocation hosts scouting the region.

:chart_increasing: Timeline: First sale in 2–4 weeks once the sample PDF exists. Refresh it quarterly, sell it forever.

🎯 The Mini-Emerald Play

Emerald AI serves clients who need 100 GW. Nobody serves the company that needs 5 MW for a new plant. That buyer exists, has budget, and has literally no one to call.

Here’s what you do: use the same free FERC and state utility data, packaged for one specific small buyer — a manufacturer, a mining op, a small host — who needs 1–10 MW and is currently flying blind.

:brain: Example: A researcher in Medellín finds three factories that just announced expansions, tracks down their facilities managers on LinkedIn, and sends each a one-page “available capacity near you” summary. One replies and pays $1,200 for the full report.

:chart_increasing: Timeline: First paid report in 3–4 weeks. Every new expansion announcement is a fresh lead — this one compounds.

🛠️ Follow-Up Actions
Want to… Do this tonight
See the waiting list Download the FERC interconnection queue CSV
Read a real grid map Grab the CAISO or ERCOT capacity report
Understand the term Skim interconnection queue on Wikipedia
Follow the money Read the original TechCrunch report

:high_voltage: Quick Hits

Want Do
:electric_plug: The one-line version AI ran out of outlets, not chips
:france: The scale 100 GW ≈ all of France
:hourglass_not_done: Why old land matters New grid hookups take 3–7 years
:bar_chart: The free goldmine FERC’s public queue
:world_map: Who sells the map Emerald AI — for the giants only

Google has enough cash to buy several small countries. It still can’t find a place to plug in. The map that fixes that is free — and almost nobody is reading it.

1 Like