AIANALYSISPODCAST
A 300 gigawatt queue, $1M per megawatt deposits, and a $1.4 billion real estate deal show who can still build.
A 750-megawatt allocation in the U.S. Northeast now requires an upfront deposit of $580 million, written years before any electrons flow. The constraint in AI infrastructure has moved. It is no longer chips. It is the grid, and the new price of admission is a balance sheet that can clear nine-figure utility deposits before construction starts.
Oncor's Q1 2026 results, published May 7, 2026, show that active point-of-interconnection requests, the formal filings builders submit to plug a new data center or generator into the transmission network, rose 19% year over year, that the 2026 capital budget runs about 25% above 2025 actuals at roughly $9.0 billion, and that as of May 6, 2026 the utility was holding about $4.0 billion in customer collateral earmarked for cancelled generation and large-load projects. That collateral number is the most direct read on the new gate: it is money builders have already written to utilities to prove they can pay for the substations, transformers, and lines that have to be built before a single data center can be commissioned.
Michael Levy put the queue at more than 300 gigawatts on a recent episode of Podcast Alpha, against a current Oncor delivery capacity of about 32 gigawatts to North Texas. His assessment, attributed as his own industry read rather than a utility figure, is that the queue is roughly ten times oversubscribed and that Texas is on track to add another 32 gigawatts over the next five years. Texas deposits, in Levy's first-person accounting, run about $1 million per megawatt. A separate 750-megawatt Northeast allocation carries a $580 million upfront check that has to clear before any wire is strung.
The operators clearing those checks no longer look like the standalone real estate developers who defined the last cycle. Crow's January 20, 2026 announcement of a 245-megawatt flagship data center campus in central Dallas is the playbook. The project sits on roughly 40 acres along the Stemmons Corridor, partners with CleanArc Data Centers on an initial 70-megawatt building targeted for late 2027, and is built around a dedicated on-site substation that pulls the constraint inside the fence line. Levy's characterization of the underlying land as held for 65 years signals the holding period the new math requires. The integrated operator is now the unit of competition: land, power, and shell under one capital structure, with the utility relationship priced in before the foundation is poured.
Bridgepoint's acquisition of Kayne Anderson Real Estate closed at an upfront enterprise value of about $1.393 billion, split between $759 million in cash and roughly 189 million newly issued Bridgepoint shares, against a Kayne Anderson Real Estate platform that manages about $22 billion in assets. Combined, Bridgepoint runs a private-markets platform of roughly $117 billion in AUM, with the real estate business operating under Al Rabil as Kayne Bridgepoint and the most recent flagship fund, KAREP VII, having closed in May 2026 at $5.12 billion in commitments. Levy frames the deal as Exhibit A for the death of the standalone allocator, a thesis about capital structure rather than a verdict on the assets themselves. A pure-play real estate generalist, on this read, becomes a component of a multi-strategy private-markets platform rather than a category of its own.
Texas added 32 gigawatts over the last five years, and the buildout is a function of transmission policy as much as it is of demand. If the Texas grid operator (ERCOT) and the mid-Atlantic grid operator (PJM) compress the next 32-gigawatt build under accelerated interconnection timelines, the deposit gate and the vertical-integration moat both erode, and the data center trade returns to one trade rather than two. The second-order risk sits next to the first: a chip or architecture breakthrough could shrink a $20 billion data center the way cloud shrank on-prem, which is why integrated operators on the ground treat "data centers" as two different businesses under one name. Hyperscale shells leased to a ten-year tenant are a real-estate trade. Inference-ready shells wired for the next accelerator generation are a venture-style bet on architecture, with the chip-obsolescence risk priced into the lease.
The next data point to watch is whether Oncor's $4.0 billion collateral figure moves materially when the utility files its Q2 disclosure. If it grows, the gate is tightening. If it softens, the queue is starting to clear, and the standalone allocator gets a longer runway than the current cycle implies.