QUALCOMM Incorporated

Qualcomm does two different things under one roof, and they have to be understood separately. QTL, its licensing arm, owns one of the largest portfolios of patents essential to 3G, 4G and 5G cellular standards. Any company that makes a phone, tablet, car or watch that connects to a mobile network needs a licence, and pays Qualcomm a royalty — a percentage of the device's selling price, capped per unit — whether or not it uses a single Qualcomm chip. QCT, the chip business, designs Snapdragon processors (the CPU, graphics, AI engine and modem in one piece of silicon) for premium Android phones, modems for Apple's iPhones, and a fast-growing range of chips for cars, PCs, smart glasses, industrial devices and, from 2026, data centres. It owns no large factories: TSMC and Samsung make its chips.

A ten-year-old's version: every time anyone buys a phone, Qualcomm gets a small fee for the inventions that let it talk to the network — and for many phones it also sells the chip that does the talking and the thinking.

In fiscal 2025, to September, Qualcomm sold $44.3bn, of which $38.4bn was chips and $5.6bn licensing. The licensing was 13% of revenue and 26% of segment profit, at a 72% margin. The shares, at $175.50 on 9 October 2026, have had a wild year: $124 in April, $260 in late May on excitement about AI data centres, $148 at the end of July on Apple and memory costs, and back to $175 now. This report asks what each half of the company is worth.

Two lessons for an owner. The first is that the toll has outlived every attack on it. Regulators in four jurisdictions and Qualcomm's largest customer spent the 2010s trying to cut its royalties; the FTC lost on appeal, the EU's €997m fine was annulled, a £480m UK class action was dropped this year without a penny paid, and Apple — which built its own modem precisely to stop depending on Qualcomm — has just signed up again. The second is that the chip business has never been a toll. It earns what its latest Snapdragon earns against MediaTek, Apple and Samsung, and its biggest customers have spent years designing Qualcomm out. Over ten years the shares have returned 13% a year with dividends, a decent result that still trails the market; since the 2000 bubble, 4.5% a year. Paying a high price for Qualcomm's growth stories has rarely worked.

The toll. QTL licences Qualcomm's cellular patents for a percentage of each device's wholesale price. In 2017 Qualcomm published its 5G terms: 3.25% of a multi-mode handset's price for the essential patents alone, 5% for the whole portfolio, with the price base capped (reportedly at $400 since 2018) — roughly $13–20 a phone at most. Royalties are paid quarterly by hundreds of licensees; the business needs almost no capital; in fiscal 2025 it earned $4.0bn before tax on $5.6bn of revenue. Its weakness is that it has stopped growing: revenue has been flat for three years, Huawei stopped paying when its licence expired in early 2025, and rates set in 2017 must be defended again for 6G.

The chips. QCT sold $38.4bn in fiscal 2025 at a 30% pre-tax margin. Its economics are those of a leading chip designer: enormous R&D (Qualcomm spent $9.0bn in fiscal 2025), a new flagship Snapdragon every year, and pricing set against MediaTek and the in-house chips of Apple, Samsung and Xiaomi. Handsets were 72% of QCT in fiscal 2025; by the June 2026 quarter, after a 20% fall, they were 60%, with cars and IoT 40%. The business also carries the cost of the inventions QTL licenses — which is why the two are run together, and why a smaller chip business would eventually mean a thinner patent portfolio.

Where the money goes. Qualcomm turns about 30% of revenue into free cash flow in a good year ($12.8bn in fiscal 2025) and pays a large share of employees in stock: $2.8bn of stock-based pay in fiscal 2025, a fifth of free cash flow. The company's adjusted (non-GAAP) earnings leave that out; we put it back.

Buffett's 1991 letter defines an economic franchise as a product that is needed or desired, "thought by its customers to have no close substitute", and not subject to price regulation. QTL meets the first two tests more completely than almost any business in technology — you cannot build a phone that talks to a 5G network without its inventions — and has spent fifteen years fighting regulators over the third, mostly successfully. That is a genuine toll. The chip business meets none of the tests reliably: Snapdragon is desired, but customers have substitutes and are building more. His 2007 letter supplies the warning for QCT: "A moat that must be continuously rebuilt will eventually be no moat at all." Qualcomm's own 10-K concedes the licensing programme must be renewed and renegotiated continually too, but on a ten-year cycle, not an annual one. We score the moat 7: wide around the patents, narrow and narrowing around the handset chips, unproven in data centres.

What Apple did. For five years after the 2019 settlement every iPhone used a Qualcomm modem. In February 2025 Apple shipped its own, the C1, in the cheap iPhone 16e; in September 2025 the C1X went into the iPhone Air; and in September 2026 the C2 went into the iPhone 18 Pro and the new iPhone Duo worldwide, leaving only the US version of the 18 Pro Max on Qualcomm's X80. Qualcomm had planned for about 20% of the 2026 iPhone launch; on its July call it said the share would be well below that. Qualcomm does not name its customers' shares, but its largest unnamed customer was 21% of fiscal 2025 revenue and 24% of the first nine months of fiscal 2026; most readers assume that is Apple, royalties included.

What Apple kept. On 24 September 2026 Qualcomm announced that Apple had renewed its global patent licence from 1 April 2027, when the 2019 agreement runs out. The terms were not disclosed. That is the single most important fact in this report: Apple can design its own modem but cannot design its way around the cellular standard, and it chose to pay rather than litigate again. The toll survives the defection of the toll-payer's biggest rival as a customer.

Can the rest replace Apple? Management says non-handset revenue growth will accelerate from 24% in fiscal 2026 to more than 60% in fiscal 2027, enough, by its account, to replace Apple's product revenue within a year. The June 2026 investor day roughly doubled the fiscal-2029 target for non-handset revenue to $40bn — but look closely: the old target for cars and IoT was $22bn and the new one for the same businesses is about $24bn. The rest is a new data-centre line, >$15bn, from a business that barely exists today. In September Amazon agreed a multi-generation collaboration and received a warrant for up to 25m Qualcomm shares at $161.26, vesting as it buys up to $60bn of Qualcomm server products; 3.75m vested on its initial commitments. That is a strong signal of a real customer, and also a reminder of the price of entry: Qualcomm is paying a hyperscaler, in shares, to buy from it.

Our reading: the patent survives the customer; the chip business is being rebuilt with someone else's money at risk — ours. The licensing toll is worth roughly $4bn a year of pre-tax profit with very high confidence, and Apple's renewal extends that confidence into the 2030s. The chip business is losing a large, low-margin piece (Apple modems carry less margin than full Snapdragon systems, Qualcomm says) and gaining faster-growing, more diversified ones. Cars are already a real success. Data centres are a bet against Nvidia, Broadcom and the hyperscalers' own chips, and the shares' doubling and halving this year shows how hard the market finds it to price. Fiscal fourth-quarter results in early November — guided at $2.05–2.25 of adjusted EPS — will show the first full quarter of the Apple step-down.

The pattern. Qualcomm's chip business faces a structural problem that its licensing business does not: its biggest customers have the money and the motive to replace it. Apple has done so; Samsung alternates; Huawei and Xiaomi build their own; Amazon, Google and Microsoft design their own data-centre chips. Qualcomm's answer is to sell to customers who are not chip designers — carmakers, PC makers, industrial firms — and to offer hyperscalers a custom-chip service, as Broadcom does. In licensing, by contrast, there is no competitor: the alternative to paying Qualcomm is not to make a cellular device. The comparison on our board is Apple, the customer that left — and that Berkshire owns.

Integrity: no veto. Qualcomm discloses its customer concentration, its Apple exposure and its input-cost problems plainly, and its fights with regulators were won in court rather than by obfuscation. Capital allocation: good when cheap, aggressive when not. The 2018 buyback was superb; the 2026 buybacks, at $125–200 a share and partly borrowed, are a policy — "a baseline of anti-dilutive stock repurchases", in Amon's words — rather than a judgement about price. Buffett's 2011 letter is blunt about that: "It doesn't suffice to say that repurchases are being made to offset the dilution from stock issuances", and "what is smart at one price is dumb at another." Owner mentality: weak signal. No officer or director has bought a share on the open market in the past year; the chief executive has sold at every round number on the way up. We score management and capital 6.

Which earnings to believe. Qualcomm's GAAP earnings over the last twelve months ($8.76 a share) are meaningless on their own: they include a $5.7bn non-cash tax charge booked in September 2025 after the US tax law changed, and the release of the same $5.7bn in March 2026. Its adjusted earnings (~$10.51 for fiscal 2026) exclude both — and also exclude about $3bn a year of stock-based pay, which is a real cost: shareholders pay it through dilution or through the buybacks that offset it. Our owner-earnings measure is free cash flow minus stock-based pay: about $10.0bn in fiscal 2025 ($9.10 a share) and $7.2bn over the last twelve months ($6.70), a trough depressed by higher capex and the handset slump. We use ~$8.50 as a normal year.

Margins are the number to watch. In the June quarter QCT's pre-tax margin fell to 26% from 30% because memory, wafers and advanced packaging all became dearer at once, and Qualcomm raised its own prices from 1 September to pass them on. If that works, margins recover; if customers resist, the chip business is worth materially less than its history suggests.

The verdict on the dividend: safe, modest, and not the point. At $3.68 a share the dividend yields 2.1% and uses only about a third of free cash flow; there is no plausible path to a cut short of another royalty war, and Apple has just signed up for the next round. But Qualcomm has chosen to make buybacks its main way of returning cash, and to grow the dividend at low single digits, so an income investor gets a slowly rising 2% yield, not a compounding one. The risk sits one level up: the buyback is now partly borrowed, at prices that were not obviously cheap, while the company is also paying for acquisitions and capex for data centres. The comparison on our board is Broadcom, which pays a larger share of its cash as dividends.

A sum of the parts first. QTL earns about $4bn a year before tax, ~$3.5bn after. A royalty stream that durable — Apple renewed, every legal attack beaten — deserves 18–20 times: $63–70bn, or about $60–67 a share. That leaves the market paying ~$115bn, or about $110 a share, for the chip business: ~13–14 times its after-tax segment profit in a depressed year, for a business that is losing Apple, gaining cars and betting on data centres. Put that way, the toll is cheap and the chips are priced as if the plan works.

What is it worth? We discounted owner earnings of about $8.50 a share at 10% for ten years. If the Apple loss is replaced as management says, margins recover and owner earnings grow ~7% a year for five years and 4% thereafter, ending at 16 times, Qualcomm is worth about $160. If handsets keep shrinking and data centres disappoint — $7.00 growing 1%, 11 times — about $78. If the fiscal-2029 plan is met — $9.00 growing 13% then 6%, 18 times — about $249, close to May's high. At $175.50 the price sits about 10% above our central value. For a business of this quality we want a margin of safety: buy below about $140, roughly 12% under our central case — a level the shares passed through in both April and July this year.