The biggest infrastructure funds on the planet — GIP (BlackRock's $100B+ infra arm), EQT, CalPERS, Qatar Investment Authority — just spent months doing due diligence on AES Corporation, a utility with 12 GW of renewable backlog, a $481M data center development deal, and Bloomberg NEF's #2 ranking in corporate renewable power sales.

Their price: $15.00 per share. The stock was trading at $17.26.

The smart money looked at the AI power thesis with full access to the data room, ran the numbers on 12 GW of contracted renewables, $1.5B in IRA tax credits, and a "leading provider of renewable energy to data center companies" — and offered less than what retail was already paying.

That's worth understanding.

The Deal

AES signed a definitive merger agreement on March 1, 2026, one day before filing what will be its last standalone 10-K:

• Buyer: GIP (BlackRock) + EQT Infrastructure VI
• Price: $15.00/share cash, board unanimous
• Enterprise value: $33.4B (including $29.5B debt)
• Multiples: 11.6x Adj EBITDA, 7.6x EBITDA with tax attributes
• Breakup fee: $588M from buyer, $321M from AES
• Deadline: June 2027, extendable to December 2027

The stock dropped 18% on announcement week. Investors were angry. AES was trading above the offer price. The board took the money anyway — unanimously.

What They're Actually Buying

Not AI power. Tax credits.

AES generated $1,540M in IRA tax attributes in 2025, up from $1,313M in 2024. At $15/share, that's a 14.4% annual tax-attribute yield. Private owners can harvest these without quarterly earnings calls, without GAAP impairment noise (AES took $264M on a Bulgarian coal plant and $103M on a failed smart grid investment in 2025 alone), and without a public market that can't figure out if this is a coal company, a renewable developer, or an AI play.

The renewables backlog is real — 12 GW globally, 5.7 GW under construction, 4 GW signed in 2025. But the Texas data center deal everyone got excited about? It's a $481M development services contract that runs for two years. Not a 20-year PPA. Not recurring revenue. AES built some infrastructure for an unnamed customer and recognized $105M in 2025. The remaining $376M books through 2027 and then it's done.

Compare that to Talen-Amazon ($18B, 17-year nuclear PPA) or Google's unnamed $9.9B commitment through 2047. AES's data center deal is a rounding error in this market.

The Coal They Can't Sell

Buried in the 10-K: AES tried to sell its Mong Duong coal plant in Vietnam (1,242 MW, PPA running to 2040) and failed. Reclassified it from held-for-sale back to held-and-used. $862M in loan receivables sitting on the balance sheet.

GIP and EQT just bought a Vietnam coal plant that runs until 2040. At $15 per share.

Add Maritza (Bulgaria, 690 MW, $264M impairment, PPA expires May 2026), Southland (California gas, 1,436 MW, contracts expiring 2026 with no plan), and Uplight (smart grid investment written to zero) — and the gap between $17 and $15 starts making sense. The coal and stranded assets aren't free. The discount from market price is the market learning what private buyers already knew.

What This Means for CEG, VST, and NRG

Here's the comparison that should worry anyone long AI power utilities:

All three public names have less than 45% idiosyncratic variance. These aren't stock picks — they're factor bets on the AI power narrative, dressed up as individual companies. When the narrative fades, they move together. Down.

Yes, CEG's nuclear fleet is genuinely scarce and VST has real ERCOT pricing power. AES is a mixed bag of international assets, coal liabilities, and development contracts. The P/E comparison isn't apples-to-apples. But the directional signal is clear: the most informed infrastructure buyers on the planet, deploying from a record $175B fundraising year, priced AES's AI power story at 11.6x EBITDA. The public market prices CEG's at 43x earnings.

Someone is wrong.

The Take-Private Wave

AES isn't isolated. This is the fourth major infrastructure take-private of a US utility/IPP since 2024:

GIP appears in two of four deals. Brookfield launched a $100B AI infrastructure program. Private capital is hoovering up energy transition assets — but at private market prices, not public market premiums.

The pattern: regulated rate base + renewables backlog + data center positioning, bought at 7-12x EBITDA while the public market prices comparable narratives at 40-60x earnings.

The Arb

Current spread: $14.15 vs $15.00 = 6% gross over max 15 months. Annualized, that's 4.8% — barely above the risk-free rate.

The CFIUS question matters. GIP is BlackRock (US-domiciled), but co-investors include EQT (Swedish) and QIA (Qatari sovereign wealth). A foreign sovereign acquiring US utility infrastructure that powers AI data centers during Trump's national security posture is a real review. GIP cleared CFIUS for Allete (with Canadian pension money) in 19 months. Qatar is a US ally hosting Al Udeid Air Base. QIA is likely structured as passive LP. Deal probably clears — 80% — but you're getting paid almost nothing for the risk.

The arb isn't interesting. What the deal reveals about the sector is.

Evidence