AT&T Bought $23 Billion of Spectrum and Pledged Half Its Fiber to a Pension Fund: The AI Grid's New Financing Template

AT&T reported third-quarter 2026 revenue of $32.3 billion, up 5.3% and the best consolidated service revenue growth in more than four years, with adjusted EBITDA of $12.5 billion (up 5.4%), adjusted EPS of $0.65 against $0.54 a year earlier, and free cash flow of $5.0 billion. What the headline numbers hide is the balance sheet: total debt closed at $159.6 billion and net debt at $149.8 billion, up from $126.4 billion three months earlier, and management attributed the sequential jump to the closing of the EchoStar spectrum transaction. Leverage rose from 2.68x to 3.13x in a single quarter, and the promise is to be back in the 2.5x range during 2029.

On Oct. 6, 2026, AT&T agreed to combine Forged Fiber 37 — the subsidiary holding the fiber build engine and network assets it bought from Lumen Technologies for $5.75 billion — with Gigapower, its existing wholesale open-access platform, and place the combined business in a joint venture with Global Infrastructure Partners, the infrastructure manager AT&T first partnered with in 2023 and which BlackRock acquired in October 2024, and with CPP Investments, the Canada Pension Plan Investment Board. AT&T keeps 50%; GIP and CPP Investments hold 50% between them. The venture closes in the first half of 2027, AT&T expects proceeds at closing, and it will not consolidate the venture — it books its share of equity income instead.

Run the arithmetic and the constraint is explicit. Net debt of $149.8 billion at 3.13x implies roughly $47.9 billion of trailing adjusted EBITDA, so the 2.5x target implies net debt of about $119.7 billion — roughly $30 billion of debt to be removed by 2029. Free cash flow of $18 billion covers the roughly $7.6 billion annual dividend about two times over, leaving some $10 billion a year for debt repayment, which over three years is most but not all of the gap. Management said on the earnings call that repayment will come from excess cash flow plus the 2027 proceeds of the fiber venture and the planned sale of Mexico.

The venture price is not disclosed, but the tape around it is instructive. In eight months, infrastructure capital bought almost every large piece of US fiber that came to market. Crown Castle closed the sale of its fiber and small-cell businesses to Zayo and an EQT infrastructure fund on May 1 for $8.5 billion gross and $8.4 billion net, then used more than $7.0 billion of it to cut debt and $1.0 billion to buy back stock. Verizon went the other way, paying $20 billion — $38.50 a share with roughly $10 billion of debt assumed — to buy Frontier in a deal that closed on Jan. 20, 2026 and left it with nearly 10 million fiber subscribers. Private capital is now the marginal buyer of American fiber, and it clears at a price set by contracted cash flows rather than by telco balance sheets.

The build itself is being sold by companies whose balance sheets can carry it. Corning — the optical fiber and cable behind most US builds — grew second-quarter 2026 revenue 16.6% to $4.5 billion and carries 1.62x net debt to EBITDA. Quanta Services, which builds the transmission lines, substations and telecom plant for the same customers, took second-quarter revenue to $9.56 billion from $6.77 billion a year earlier and still runs 1.97x net debt to EBITDA. On the demand side, Digital Realty operates 309 data centers with roughly 3.0 gigawatts of IT capacity and about 6.3 gigawatts of buildable capacity, and Equinix sells the interconnection that turns fiber into cloud capacity.

The pattern is now visible enough to underwrite. AT&T spent $23 billion on spectrum it could not fund out of free cash flow, told investors leverage would reach 3.13x, and then placed half of its fiber engine with a pension fund and an infrastructure manager in exchange for cash and capital-light growth. The same construction is running on the power side, where Google is underwriting 890 MW of new nuclear through Constellation Energy and Meta has signed 20-year agreements with Vistra. The investable conclusion is narrow but important: the AI grid's capital intensity is migrating from telco balance sheets into vehicles priced by long-duration capital, so the fiber and grid-construction vendors capture the spend while the carriers capture the deleveraging.