AT&T Inc.

AT&T sells connectivity. Its main business is a nationwide 5G wireless network — radio spectrum licensed from the government, equipment on tens of thousands of towers, and the fibre that links them — used by about 75 million postpaid phone customers and 110 million wireless connections in all. Its second business is home and business internet, increasingly over its own fibre: it passes 38.6 million locations with fibre and has 12.9 million fibre customers, plus 2.6 million on "Internet Air", which delivers broadband over the wireless network. A shrinking third business is the old copper telephone network, which AT&T is switching off by 2029. A small fourth is wireless service in Mexico. A ten-year-old's version: AT&T owns invisible radio highways and glass cables under the street, and charges you every month to use them.

The economics are those of a utility without a regulator. AT&T spends $22–24bn a year on its networks and billions more, from time to time, on spectrum; in return it collects about $125bn a year in revenue, earns ~$46bn of adjusted EBITDA and turns $16–18bn into free cash flow. Prices are set by competition among three national carriers — and, increasingly, cable companies that resell Verizon's network. Customers rarely leave: just 0.86% of postpaid phone customers cancel each month.

The shares, at $22.18 on 9 October 2026 — the day they went ex-dividend and the day after SpaceX agreed to buy nationwide spectrum to launch its own mobile service — are 24% below their March high. The dividend, $1.11 a year, yields 5.0%; the ten-year Treasury yields about 5.2%. This report asks whether the network is still a moat, and whether the dividend is still a good reason to own it.

Two lessons for an owner. The first is that AT&T's management has a record of paying too much for things outside its network. DirecTV went in at $67bn and came out at $16bn; Time Warner went in at $109bn and was spun off with a dividend cut. Buffett's 1981 letter, written about acquisitive managers in general, happens to name the target: managers "certain their managerial kiss will do wonders for the profitability of Company T(arget)". The second is that the returns have been poor even when the operations were fine. Over the ten years to 2025 AT&T returned about 5.6% a year with dividends reinvested, against nearly 15% for the S&P 500; over twenty years, 7.3% against 11%. A 5% yield has not been enough to make AT&T a good investment, because the business has needed almost everything else it earns.

The wireless arithmetic. A postpaid phone customer pays AT&T roughly $57 a month and stays for years — at 0.86% monthly churn, the average customer lasts about a decade. The network costs nearly the same whether it carries 70 million customers or 75 million, so each extra customer is very profitable — which is why all three carriers subsidise phones and promotions to win them. AT&T added 432,000 postpaid phone customers in the second quarter, more than analysts expected, and its EBITDA margin in the connectivity segment rose to 42%.

The fibre arithmetic. Fibre to the home is expensive to lay but then lasts for decades; once AT&T has fibre in a street, a rival rarely pays to build a second network there. Fibre customers pay about $73 a month, more than copper customers did, and a household with AT&T fibre is far more likely to take an AT&T phone plan: 42.5% of its home-internet households already do. The plan is 60 million fibre locations by 2030, partly through a new 50/50 venture with GIP and CPP Investments announced on 6 October.

Where the cash goes. Of ~$46bn of adjusted EBITDA in 2025, ~$22bn went back into the network, ~$7bn on interest and ~$1–2bn on taxes, leaving $16.6bn of free cash flow. Dividends took $8.2bn and buybacks $4.5bn. What the free cash flow figure does not include is spectrum: AT&T bought $23bn of C-band licences in 2021, $9bn of 3.45 GHz in 2022 and ~$23bn from EchoStar in 2026 — all outside "free" cash flow (Part VI).

Buffett named AT&T in his 2020 letter, as the American company with the second-largest stock of fixed assets after Berkshire — $127bn of property, plant and equipment — and added straight away that such leadership "does not, in itself, signal an investment triumph. The best results occur at companies that require minimal assets to conduct high-margin businesses." That is the AT&T problem in two sentences. Its moat is real: nobody can easily build a nationwide network, buy the spectrum and win 75 million postpaid customers. But the moat protects three companies, not one, and the three compete on price with subsidised phones; the result is a return on invested capital of about 6%, near the cost of capital. Buffett's 1990 line about commodity businesses fits: "it's impossible to be a lot smarter than your dumbest competitor." SpaceX would make the moat wider in one sense — another expensive entrant — and shallower in the one that matters, price. We score the moat 5.

Why the stock fell — twice. AT&T did not miss a single target in 2026. Its first fall, from $29.10 in March to $20.48 on 1 July, came from outside: Oppenheimer downgraded it on 3 June, calling satellite constellations a structural threat to broadband and mobile; on 4 June the Supreme Court ruled 8–1 for the FCC in FCC v. AT&T, upholding the agency's power to fine carriers; on 11 June the CFO announced his retirement; and on 26 June the Financial Times reported that SpaceX had told investors in its share offering that Starlink planned a retail mobile service. The shares recovered to $26.72 by mid-September as results beat expectations. The second fall came on 8 October, when SpaceX agreed to buy Grain Management's nationwide 800 MHz licences — up to 14 MHz of paired low-band spectrum, reportedly for about $8bn — saying it would "pave the way for Starlink Mobile to become a major mobile carrier in the US." The day before, the FCC had approved 15,000 direct-to-phone satellites. AT&T, Verizon and T-Mobile each fell about 7% after hours; AT&T closed at $22.18 the next day, down 9.8% after allowing for its $0.28 dividend.

Our reading: the threat is real but slow; the cost of defending against it is immediate. A Starlink mobile network will not take AT&T's city customers next year — satellites cannot carry dense urban traffic, and a terrestrial build would take years and tens of billions. But the market is right that something has changed: spectrum was the reason only three companies could compete nationally, and SpaceX — with a different cost of capital and no need to earn a return from wireless alone — has shown it can buy its way in. The more immediate point is AT&T's own: in July it paid $23bn, largely borrowed, to buy capacity at a price that assumed scarcity. Since 2021 AT&T has spent roughly as much on spectrum and fibre outside its free cash flow as it has kept after dividends. Free cash flow of $18bn+ is the right number for the dividend; it is the wrong number for valuing the business, because the spectrum bill recurs.

The pattern. US wireless has been a three-player market since T-Mobile and Sprint merged in 2020, and three-player markets tend to price rationally. The threats now come from companies that do not need wireless to earn a return: cable operators use cheap mobile lines to keep broadband customers, and SpaceX's mobile ambitions are part of a satellite business valued on very different assumptions. Buffett's 1990 warning about the dumbest competitor applies with a twist — the danger is not a foolish competitor but one with different goals. On our board, the comparison is Verizon, whose higher yield (6.1% in September, more after its fall) comes with more debt and less fibre, and SpaceX itself, which we analysed in September.

Integrity: no veto, but a long memory. AT&T reports plainly and has met its targets since 2023; but the CEO who promised in February 2021 to sustain the dividend presided over its cut, and the deals that forced the cut were made while he was the company's strategist. Capital allocation: improved, still expensive. The post-2022 discipline is real; the 2025–26 purchases took leverage back up just as the scarcity they paid for was being challenged. Owner mentality: weak. No insider bought a share in the open market all year. We score management and capital 4.

Which earnings to believe. AT&T's GAAP earnings over the last twelve months include a $5.6bn gain on the sale of its remaining DirecTV stake; its adjusted earnings ($2.12 a share in 2025, $2.25–2.35 guided for 2026) exclude it and other items. Adjusted EPS is the better guide to the business — but even it flatters slightly in 2026, because depreciation has fallen as old copper assets become fully depreciated, and tower sales produced gains. The number to hold on to is free cash flow, which AT&T defines after capital spending and payments to equipment vendors it finances: $16.6bn in 2025, $18bn+ guided for 2026. From that we subtract a normal level of spectrum spending — the real recurring cost of staying in the wireless business — to reach owner earnings.

The debt. After paying EchoStar, AT&T's net debt is roughly $149bn, about 3.2 times EBITDA at the peak by its own measure (rating agencies' adjusted measures are higher). Two term loans drawn at closing ($11.5bn and $3.0bn) mature within two years, in a market where the ten-year Treasury yields 5.2%. Interest already costs ~$7bn a year. AT&T guides leverage back to ~3× by the end of 2026 and to 2.5× about three years after the deal.

The verdict on the dividend: safe, frozen, and less special than it looks. At $1.11 a share the payout takes about half of free cash flow and is covered twice; buybacks of ~$10bn a year stand between it and any squeeze, and management has promised to hold it through 2028. In that narrow sense it is safer than it was before the 2022 cut. But it will not grow, and at 5.0% it now yields less than the ten-year Treasury (~5.2%), which carries no equity risk at all. An income investor is paid for owning AT&T only if the shares at least hold their value — something they have not done over five, ten or twenty years. The comparison on our board is Verizon, whose higher yield is also covered ~1.8× and has at least been raised for twenty years.

What is it worth? We discounted owner earnings of about $1.90 a share at 10% for ten years. If AT&T delivers its plan — free cash flow rising to $21bn+ by 2028, leverage back to 2.5× — and owner earnings grow ~3% a year for five years and 2% thereafter, ending at 10 times, it is worth about $22.70: almost exactly today's price. If SpaceX and cable force a price war — $1.50 shrinking 1% a year, 8 times — about $13, a level that would very likely come with a second dividend cut. If convergence works and spectrum spending stays low — $2.30 growing 5% then 3%, 12 times — about $33.

At $22.18 the price sits at our central value. On headline numbers AT&T looks cheap — under ten times earnings, an 11.7% free-cash-flow yield — and that is why so many income investors own it. After the spectrum bill and the debt, it is fairly priced for a business whose moat is being questioned. For a company we score below 5 on quality, our rules ask for a clear margin of safety: hold it for income if you own it, and buy below about $19, where the yield would be 5.8% and the price about 16% under our central value — near the July low.