Is 700GW Real? Wall Street Is Finally Checking the Story Behind the AI Bubble

For the past two years, the AI bubble argument has been fought over one question: is Nvidia too expensive? In 2026, the market quietly moved on to a far more uncomfortable question — does the thing you’re being sold actually exist? The SEC is asking firms selling pre-IPO shares in SpaceX, OpenAI, and Anthropic to prove they actually hold the shares. Grid operators are asking whether the hundreds of gigawatts of reported demand will ever show up. And for a nine-figure contract, the market is now asking whether the counterparty has any ability to pay at all.

I’ll be honest — my starting assumption was wrong. I thought the fake hunters were dying out. The biggest short sellers have been retiring one by one. Then I actually pulled the numbers and got humbled: published short and activist reports this year aren’t down — they’re more active than last year. The old faces are fading, but new ones are appearing. The hunt for fiction didn’t disappear. It changed hunting grounds.

If you’re holding AI-related positions — or deciding whether to get in — this isn’t another “will the bubble pop” piece. Nobody can answer that for you. What I can give you is more valuable: three verification questions you can run yourself, and a 10-minute due-diligence routine. One goal: make a story pass verification before it takes your money.

What the 700GW Number Actually Is

Let me put the number in context first. Reuters’ Sept. 1 survey found that large power projects across the U.S. have queued more than 700GW of grid interconnection requests — roughly ten times the industry’s estimate of all existing data-center electricity use. Texas is the extreme case: interconnection requests to ERCOT jumped from about 48GW in 2023 to over 474GW.

Faced with that, Texas paused new data-center grid connections and launched an audit. Why? Because the queue is full of what the industry now calls “ghost demand”: the same project filing with multiple utilities at once; developers locking in interconnection before they’ve secured land, financing, or a customer; applicants with no real ability to build a data center at all.

In other words, 700GW isn’t demand. It’s thousands of developers saying “I’ll take a spot, just in case.” When reporting a number costs almost nothing, the number will inflate.

Why This Matters More Than “Will the Bubble Pop”

• Numbers inflate when they’re free. A promise that costs nothing is a wish; a promise backed by a deposit, penalties, and real money is closer to actual demand.
• Deposits expose the fakes. After Exelon raised its collateral requirements, it cut its high-probability data-center load forecast from 18GW to 11GW — down 40%. AEP Ohio’s project pool shrank by more than half after it introduced study fees.
• But real demand is still enormous. PJM’s recent capacity auctions added $29.4 billion in data-center-driven costs. So the takeaway isn’t “it’s all fake” — it’s “you have to filter.”

The deposit test is the part I care about most, because it matters a hundred times more than the 700GW headline itself. Any narrative big enough to make you want to act can be measured with one question: is it willing to pay a deposit? If not, it’s a wish. If yes, it’s demand.

Three Checks You Can Run Yourself

None of these requires money or special access. You can do all of it on your phone.

First: does the asset actually exist? The hottest companies are staying private longer — you can’t buy SpaceX, OpenAI, or Anthropic on any public market. So someone packages a “fund interest” and tells you they’ll cut you in. But what you’re buying is the wrapper, not the company — and there may be two or three layers between you and the underlying shares: the middleman, the fund they bought, the entity that actually holds the stock. Every layer you add is one more thing you can’t verify, but the money you pay is just as real. So remember two rules: if you can’t see proof of the underlying holdings, that’s the biggest red flag there is — and if they can’t produce it, walk away. This is exactly why the SEC started demanding proof across a $2 trillion market on Sept. 1. Even regulators now think it has to be checked.

Second: how much demand is left once it has to pay a deposit? Don’t trust the sum of the queue. Look for the number that survives a deposit and a signed contract. Exelon’s number lost 40% after collateral. AEP’s fell by more than half after fees. That surviving number is the real one.

Third: does the counterparty have money, people, and an actual business? This is the “secret weapon” I want to hand you — a 10-minute due-diligence walkthrough. When a company announces a “major contract,” don’t chase the stock. Open four pages:

1. Delaware’s Division of Corporations entity search — when was the counterparty formed (a few months ago?), is it in good standing, and is the registered agent a third-party address?
2. SEC EDGAR — was the deal disclosed as an 8-K, and are the payment and termination terms real or empty? A “mega deal” that exists only in a press release is worth a 30% haircut before you read further.
3. LinkedIn — how many employees does the counterparty actually have? A four-person company signing a $133 million power deal tells you everything.
4. Wayback Machine — was its website built last month, is it two pages, and does it even have a product?

Short sellers make a living running exactly this checklist. You’re not learning to trade — you’re learning to make a story pass verification first.

The Mistakes I Made (So You Don’t Have To)

Mistake 1: I chased a big-contract headline. A company announced a $133 million AI data-center power deal and the stock popped instantly. Someone then checked the buyer: formed in Delaware in January, no office address, four employees, and a barely functional two-page website — and this one contract was a third of the company’s entire sales pipeline. A short-seller report followed, the stock dropped 17% in a day, and a class action came after. My mistake was the moment the stock popped and I only thought “don’t miss out” instead of spending ten minutes checking who was on the other side. That feeling is the worst — opening the counterparty’s website at 3 a.m., seeing two pages and no product, and realizing what I’d chased all day was just a headline.

Mistake 2: I treated short sellers as heroes or villains. There’s a guy who made a two-decade career out of exposing liars — on TV constantly — who was convicted of securities fraud this year. He was telling the public one side of a trade while quietly trading the other against the people who trusted him, to the tune of $21 million. The lesson isn’t “shorting is bad.” It’s: don’t expect a permanently honest hero in this business. Judge motives and mechanisms, not personas.

Mistake 3: I only read the financials. For years, judging a company meant reading its accounts. But lately the value sits in things that haven’t hit the income statement yet — an unfulfilled contract, an unbuilt project, a not-yet-public stake. By the time it shows up in the filings, it’s often too late. Move your verification earlier: does the asset exist, will the demand materialize, can the counterparty pay.

FAQ

Q: Is the AI bubble’s 700GW of power demand real?
Partly. 700GW is the sum of interconnection requests — roughly ten times existing data-center use — and much of it is ghost demand (duplicate filings, speculative queue spots). But after you strip the water out, real demand is still very large.

Q: Why did Texas pause new data-center grid connections?
Because requests jumped from 48GW to 474GW and Texas can no longer tell which are real. The audit now requires owners to disclose final ownership, financing, subsidies, water use, and whether on-site power is actually planned.

Q: How do I verify pre-IPO shares (SPVs) are real?
Retail investors can rarely verify directly — there’s no public registry, and two or three layers sit in between. That’s exactly why regulators stepped in. Your move: demand proof that traces to the underlying holdings; if they can’t produce it, treat it as a red flag.

Q: Can I trust short-seller reports?
Treat them as a question list, not a verdict. A report’s value is telling you what to verify. It may also carry a bias — short sellers profit from declines, and some are frauds themselves.

Q: What does the move to semiannual reporting mean for retail?
Less transparency, a wider information gap between insiders and everyone else, and prices that reflect reality more slowly. The proposal isn’t in effect yet, but it’s worth watching.

Final Thoughts

Line up these seemingly unrelated stories and they’re actually the same story: the SEC checks whether the asset you sold exists; the grid checks whether the demand you reported will show up; short sellers check whether the counterparty on your contract can pay. Different methods, same goal — make a story pass one more layer of verification before it becomes a price.

So when stories turn into prices faster than they can be verified, who ends up winning? I’ll leave that question with you.

By the way, we run a trading channel on Telegram where we occasionally talks through this kind of market movement and trade thinking — worth a look if you’re interested.