AT&T commits more than $3bn to Corning fiber in multi-year supply deal

AT&T has committed more than $3 billion across several years to fiber and cable from Corning, a multi-year supply agreement the carrier is tying to broadband demand it says AI is accelerating and to an expansion framed as extending internet connectivity "as AI drives growing data demand."

The numbers AT&T attached to that demand are its own: the average fiber customer on its network consumes more than one terabyte of data a month, five times the level of a decade ago, and the carrier expects monthly use to reach two to 2.5 terabytes by 2030 as streaming, gaming, video calls, cloud services and AI take a larger share of daily life. Those are consumption figures for existing subscribers, not a revenue forecast; the investment case rests on whether the terabyte curve keeps bending after the build is paid for.

The agreement lands on top of AT&T's stated targets of passing more than 30 million locations and counting 60 million fiber-connected internet customers by the end of 2030, neither of which commits the carrier to a spend curve. A supply agreement of this size is the kind of contract a carrier signs when it expects to keep pulling glass for years rather than months; Corning gets multi-year volume, and AT&T gets a claim on fiber and cable capacity at a moment when fiber has become both exit capital and the AI buildout's tightest input, as Macquarie's sale of its Polish fiber assets to fund a data center push showed.

Verizon struck its own multi-year, multi-billion-dollar Corning agreement first, calling for more than 80 million miles of high-density optical fiber through 2032 and up to 50 million broadband passings as it ties fiber-to-the-home, cell towers, data centers and residential neighborhoods to the same high-capacity backbone, leaning on a 30-year relationship with Corning. AT&T's agreement, by contrast, carries the dollar figure Verizon's lacked publicly, though the Verizon deal has the longer dated horizon in the coverage.

Both carriers are buying from the same supplier. Corning is now the named counterparty to two of the largest US wireline supply commitments on record in this coverage, and the sequencing — Verizon, then AT&T within weeks — says something about how the two carriers are reading the same demand. What the disclosures do not yet establish is how much of Corning's capacity is spoken for, whether a third buyer would find the same terms available, or whether AT&T's agreement includes volume commitments, pricing mechanics or delivery schedules.

AT&T executives have been building the public case for fiber for a while, and the language has moved from product quality to network dependence. CEO John Stankey, speaking at the Goldman Sachs Communacopia + Technology Conference, called fiber and fiber broadband "a superior product," saying it drives the customer experience, makes the bundle stickier, and allows the carrier to grow accretion and benefit over a customer's lifetime value; CFO Pascal Desroches was blunter at the 2026 Mizuho Technology Conference, saying "AI doesn't exist without our connectivity … plain and simple," and pointing to the carrier's fiber-based architecture as the means of connecting AI-heavy workloads.

How the spend gets underwritten turns on that framing: AT&T is not selling this as a consumer broadband upgrade alone but positioning its fiber plant as the links between workloads in other companies' data centers. Whether that positioning converts into contracted wholesale or enterprise revenue is not established by the coverage, but it shifts the audience for the build from households to operators of AI-adjacent facilities needing diverse, high-capacity routes between them. AT&T's own wiring does not reach most of those facilities, the last leg is someone else's asset, and the interconnection economics between carrier fiber and data center campuses are not disclosed.

Hyperscaler- and AI-lab-anchored assets get infrastructure pricing while everything else fights for capital, and a carrier build funded off a retail subscriber base sits on the unanchored side of that line even when the demand story is AI — which does not make the fiber less useful, only means the return depends on AT&T's ability to fill the glass at rates that justify the construction schedule. The consumption curve is the only public evidence on that question.

A $3 billion-plus supply agreement is meaningful for Corning and modest against AT&T's multi-year capital plans, roughly the right way to size it. What the second half of the decade looks like if the two-to-2.5 terabyte forecast holds is the harder question: AT&T wants 60 million fiber customers by 2030, Corning will supply the cable, and who buys the capacity on the routes connecting them is still an open column.

The Verizon comparison is the one readers will reach for, and the agreements are not symmetrical in what has been disclosed. Verizon put a fiber-mile figure on the table — more than 80 million miles of high-density optical fiber — and a date, 2032, alongside up to 50 million broadband passings; AT&T put a dollar figure on the table, more than $3 billion over several years, with no mile count and no end date in the coverage. A reader can come away thinking AT&T bought less; a reader can just as easily conclude the disclosures were drafted for different audiences.

Corning sits upstream of both carriers, and two multi-year, multi-billion-dollar commitments from the two largest US wireline buyers read as much on demand for the glass industry as for broadband. Corning has appeared in four of our stories as of late September, fewer than AT&T's seven, but the direction is what matters: carriers paying a component maker for capacity at scale, years out.

That same week, AT&T also leased 55 spectrum licenses across 25 states, adding roughly 20MHz of midband in most markets over installed radios as a cheaper path to depth than buying at auction; wireless depth and wireline depth are being bought with different instruments in the same budget, both capacity moves but only one requiring Corning.