Texas Instruments

Texas Instruments makes the unglamorous chips that sit inside almost everything that plugs in or runs on a battery. Most of them are analog chips: they manage power, so that a battery charges safely and a motor gets the right current, and they translate the real world (temperature, sound, pressure, light) into the ones and zeros a processor can read. The rest are embedded processors, small computers that run a car's brakes, a factory robot or an electricity meter. TI sells more than 80,000 of them. Most cost less than a dollar or two.

Three things make this a different business from the chip companies in the headlines. First, no single product matters much: TI's sales are spread across tens of thousands of parts and a very long list of customers, and a part designed into a car or a machine tool tends to be bought for ten years or more. Second, TI sells directly: more than 80% of 2025 revenue came straight from TI, much of it through ti.com, rather than through distributors. Third, and this is the heart of the story, TI makes its own chips. It does most of its manufacturing in-house, and since 2011 it has been moving production onto 300-millimetre silicon wafers, on which, the company says, an unpackaged chip costs about 40% less than on the older 200mm wafers.

In 2025 TI took in $17.7bn: $14.0bn from analog (79%), $2.7bn from embedded processing (15%) and $1.0bn of other items. Industrial and automotive customers bought about a third each. A new end market, the data centre, was about 9% of revenue in 2025 and about 13% by the third quarter of 2026, growing at roughly 95% a year as AI servers need ever more power-management chips.

What makes TI interesting today is not the products, which change slowly, but the money. Between 2021 and 2026 TI spent about $22 billion building new factories in Texas and Utah. While it did, its free cash flow collapsed, and it kept raising its dividend anyway. In 2026 the building stopped and the cash came back faster than almost anyone expected. The share price followed.

Most companies that pay rising dividends protect them by spending carefully. In 2021 TI's management did the opposite. It announced what it called a six-year period of elevated capital spending: new 300mm fabs at Richardson (RFAB2), a working fab it bought at Lehi, Utah (LFAB1) with a second one rising beside it, and the largest project of all, a site at Sherman, north of Dallas, with room for four fabs. In June 2025 the company put a number on the whole plan: more than $60 billion across seven US fabs, with no deadline attached.

The spending came fast. Capital expenditure, which had been $0.5–1.1bn a year from 2015 to 2020, went to $2.5bn in 2021, $2.8bn in 2022 and then $5.1bn, $4.8bn and $4.6bn in 2023, 2024 and 2025. Free cash flow per share, the number TI tells its owners to judge it by, fell from $6.72 in 2021 to $1.47 in 2023. At the same time the analog cycle turned down: revenue fell from $20.0bn in 2022 to $15.6bn in 2024 as customers worked through the inventory they had hoarded during the pandemic shortage.

Here is the part an owner should not skip. Through all three lean years TI raised its dividend, from $4.69 a share declared in 2022 to $5.02, $5.26 and $5.50. In 2023 it paid $4.56bn of dividends against $1.35bn of free cash flow. Over 2023–2025 together it paid $14.35bn of dividends and $2.70bn of buybacks out of $5.79bn of free cash flow. The difference, about $11.3bn, came from new debt (net borrowing of $5.3bn), from running down cash and short-term investments (from $9.1bn to $4.9bn), and from employees' option exercises. Part X takes that apart.

Free cash flow here is TI's own definition: cash from operations, minus capital spending, plus the cash grants received under the CHIPS Act. The grants are real money (TI has an award of up to $1.6bn, of which $630m had arrived by September 2026, and a 25% investment tax credit that rose to 35% for equipment placed in service from 2026), but they are partly one-off. Without the $1.1bn of grants received in the last twelve months, free cash flow was $7.95 a share, not $9.14.

Two lessons for an owner. The first is that this management will sacrifice several years of cash to lower its cost of production for decades, and will not cut the dividend to do it. That is the behaviour of people who think like owners, and it is also a risk you take on: the next time they see a reason to build, they will build again. The second is that the bet was made with a strong balance sheet and is being repaid by the cycle. Had the analog downturn of 2023–24 lasted two more years, the dividend would have been paid with more borrowed money, and the story would read differently.

On 21 October 2026 TI reported the best quarter in its history. Revenue was $6.06bn, up 28% from a year earlier and well above the previous record of $5.24bn set in the third quarter of 2022. Analog sales rose 32% to $4.91bn with a 51% operating margin; embedded processing rose 21%. Gross margin was 64.2% and operating margin 47.5%. Earnings per share were $2.62, against $1.48 a year earlier. For the fourth quarter TI guided revenue of $5.80–6.30bn, about flat on the third at the midpoint, when the fourth quarter is usually seasonally weaker.

Two things are happening at once, and an owner should keep them apart. The first is the cycle: after two years of customers living off their own stock, they are buying again. Industrial revenue grew more than 40%, lead times are stretching and, for the first time in this upturn, TI is raising some prices. The chief executive said in October that customers hold less inventory than TI would like. That is how upturns look in their second year. The second is secular: data-centre revenue, about 13% of the total, grew roughly 95%, because AI servers need large numbers of power chips, and TI's US-made capacity is a selling point to customers worried about supply from Asia.

The depreciation wedge. The lower half of the chart above is the most useful picture in this report. Reported gross margin fell from 68.8% in 2022 to 57.0% in 2025, which looks like a business losing its edge. It was not. Depreciation, the accounting charge for the new fabs, rose from $0.9bn in 2022 to $1.9bn in 2025 and is running at about $2.2bn a year in 2026. Add it back and the margin before depreciation was 73.4% in 2022, 67.9% in 2025 and 73.4% again in the third quarter of 2026. The cash economics of a chip sold today are what they were at the last peak. The difference is that TI now carries much more factory, and with it the ability to sell far more chips at low cost when demand allows.

That cuts both ways. Management says each extra dollar of revenue brings 75–85 cents of extra operating profit. In a downturn the same arithmetic runs backwards, and the fixed cost is now twice as large as it was in 2019. The third quarter's 47.5% operating margin is a peak-cycle number, and we value the company on something lower in Part XI.

By end market, TI now names five: industrial, automotive, data centre, personal electronics and communications equipment. Industrial and automotive were each about a third of 2025 revenue; together with the data centre they were about 75%, against about 43% in 2013. That shift was deliberate. A factory or a car uses chips for ten to fifteen years and its makers value a supplier that will still be shipping the same part in 2035; a phone maker redesigns every year and squeezes on price. TI has been leaving the second kind of business for the first for more than a decade.

By geography, the picture needs care. Measured by where customers have their headquarters, 2025 revenue was 38% United States, 21% China, 21% Europe, Middle East and Africa, 11% the rest of Asia and 7% Japan. Measured by where the chips are shipped, TI says that about half went into China, because so much of the world's electronics is assembled there, including for American and European customers. A trade war that stopped chips entering China would hit TI far harder than the 21% suggests.

One customer bought 12% of revenue in both 2024 and 2025, mostly analog parts. TI does not name it. A supplier of 80,000 parts with one customer taking an eighth of the total has a concentration that its catalogue hides; it is not a fatal one, but it is the single largest relationship in the business.

Who subsidises whom. In this cycle, the US taxpayer is a partner. The CHIPS Act award of up to $1.6bn for the Sherman and Lehi fabs, paid as milestones are met and subject to clawback, and an investment tax credit of 35% on new equipment from 2026, lower the effective cost of TI's capacity. TI received $1.57bn of CHIPS benefits in the twelve months to September 2026, including credits used against tax. That is a genuine advantage over foreign competitors building elsewhere, and it is also a dependence on policy that can change.

We look for moats by testing claims, not by admiring companies. TI makes five claims about its competitive advantages, and they hold up to different degrees.

Put together, TI's moat is mostly a cost advantage plus switching costs on a very wide front. Neither is dramatic. No customer is locked in the way a software customer is; any single TI part has a competitor. But a competitor who wants to take TI's place has to match tens of thousands of parts, at TI's cost, through TI's channel, at a time when TI has just built the capacity to cut prices if it chooses. That is why we rate the moat 8, and why we think the trend is widening: the fabs that hurt the free cash flow for three years are the reason TI's cost per chip should be lower than its rivals' for the next twenty.

Where it cracks. At the low end, in commodity power and interface parts, Chinese analog makers are improving quickly with government support, and China opened an anti-dumping investigation into US analog chips in September 2025, naming exactly the kind of parts TI sells in volume. If China's own makers take the cheapest half of the market inside China, TI loses volume that its new fabs were built to fill. The moat is wide in Detroit and Stuttgart; it is narrower in Shenzhen.

The pattern. In analog, the competition is not a duel but a crowd, and TI has chosen to win it on cost and breadth rather than on the most advanced design. ADI is the better-performing business on growth this year, partly because its customers are buying high-end parts for AI and communications that TI makes less of. What differs most is the capital. ADI spends less on factories and lets foundries carry the risk; TI has chosen to own the cost curve. If the next decade rewards scale and dependable domestic supply, TI's choice looks wise. If it rewards the highest performance and the lightest balance sheet, ADI's does.

The irony worth noticing. On 5 November, the day AMD fell 17.3% (see our AMD X-Ray of 14 November), TI fell 10.7% with no news of its own, because the semiconductor index fell 12.2%, its worst day since March 2020. TI makes almost nothing that goes into AI accelerators. It is valued, for the moment, as part of a basket it does not much resemble. That is one of the things that can create an opportunity in a business like this, and one of the reasons we would wait for one.

Integrity. We find nothing to question. TI's reporting is plain, its definitions are stated, its capital-management calls each February explain what it intends to do and then report on whether it did. It told owners in 2021 that free cash flow would fall for several years, and it did; it said capital spending would drop sharply in 2026, and it did. That is rarer than it should be.

Capital allocation. TI's long-standing rule is to judge every use of money by what it does for free cash flow per share over the long term, and to return all free cash flow to owners over time through dividends and buybacks. Its record on buybacks is good: shares outstanding fell from 1,261 million in 2009 to about 908 million early in 2026, and it bought $5.1bn of its own stock in 2018 when the shares were cheap. Its record on the fabs is, so far, vindicated. Its record on large acquisitions is short: National Semiconductor in 2011 is generally judged to have worked. Silicon Labs, at about 9.6 times sales for a company that lost money in 2025, is the large untested bet of this management (Part VIII).

Governance. Three points an owner should know. The roles of chairman and chief executive were combined on 1 January 2026, with Janet Clark as lead independent director. Shareholder support for pay fell to 83.2% at the April 2026 meeting, from 87% the year before, and the committee itself scored TI's relative shareholder return for 2025 below the median. And in one year TI has changed its chairman, its finance chief and a senior vice-president, while undertaking its largest deal in fifteen years. None of these is alarming. Together they mean the new team has not yet been tested by a downturn.

We would not have paid 9.6 times sales for a business that lost money last year, and we think it is fair to say so plainly. The case for the deal rests almost entirely on TI's own fabs: Silicon Labs pays a foundry to make its chips, TI can make them cheaper, and $450m a year of savings on a $7.5bn price is a 6% pre-tax return before any growth. With growth in connected devices and TI's channel behind the products, the return could reach the 10% the company asks of itself. It is a reasonable bet with a manufacturing logic. It is also exactly the kind of purchase that a company flush with cash after a long drought tends to make at the top of a cycle, and the approval from China, where TI ships about half of its chips, is a lever in a trade negotiation that neither company controls.

Buybacks have nearly stopped. TI bought back $1.48bn of stock in 2025 at an average of about $174 a share, and only $233m in the first nine months of 2026, at around $212. With $7.5bn set aside for Silicon Labs, it has $18.56bn of unused buyback authority and has used little of it this year. Meanwhile employees exercised options worth $770m in nine months, and the share count rose about 0.7% in 2026, the first rise in many years. For a company that judges itself on free cash flow per share, a rising share count is a small thing that should not continue.

Our rating of 7 for management and capital allocation reflects two strong decisions (the fabs, and holding the dividend through the trough) and one we have doubts about (the price paid for Silicon Labs), with the outcome of all three still to be fully proven.

The balance sheet is strong and getting stronger. At 30 September 2026 TI held $8.1bn of cash and short-term investments against $13.65bn of debt at an average coupon of 4.1%, so net debt was about $5.4bn, roughly half of one year's earnings before interest, tax, depreciation and amortisation (about $10.7bn over the last twelve months). Operating profit covered interest about 15 times. Moody's affirmed its Aa3 rating on 23 October 2026 and moved the outlook back to stable from negative, citing improving cash generation. After the $2bn bond of 9 November and the Silicon Labs payment, net debt would rise to about $13bn, or about 1.2 times EBITDA: still comfortable. The largest single maturity in the next five years is $1.4bn in 2029.

Returns on capital fell during the build and are recovering: our feed puts the trailing return on invested capital at about 21%. The accounting is clean, with one seam an owner should know: CHIPS Act grants are not recorded as revenue but reduce the cost of the fabs on the balance sheet, which lowers depreciation ($353m less in 2025) and flatters reported margins slightly, while TI adds the cash grants to its free cash flow. Both are disclosed; neither is a trick. But it means TI's free cash flow over the last twelve months includes $1.1bn that will not recur once the award is paid out.

On 17 September 2026 TI raised its quarterly dividend by 7%, from $1.42 to $1.52 a share, the 23rd consecutive annual increase. At $6.08 a year the yield is 2.0%, the lowest in at least ten years: at the end of each year from 2015 to 2025 it ranged from 2.03% to 3.17%. Over ten years the dividend declared per share grew from $1.40 to $5.50, about 14.7% a year, far faster than revenue (about 3% a year). That gap could only be closed by a rising share of cash being paid out, and by fewer shares, and for three years it was closed by debt.

Our reading. The dividend is safe, and it is safe partly because management has already shown what it will do under pressure: borrow rather than cut. That is admirable loyalty to income-seeking owners, and we would not want to own a company that made a habit of it. In the next downturn, with Silicon Labs' debt on the balance sheet, the dividend would again be the last thing to go and the balance sheet the first thing to give. With net debt at half a year's EBITDA, it has room to give.

The weakness is the yield, not the safety. At 2.0% the dividend is no longer a reason to buy TI; it is a pleasant consequence of owning it. Compare Altria, which we called "own for income" at about 6.5% in August, or PepsiCo at about 4.1%. An owner who bought TI at the end of 2025 at $173.49 is receiving 3.5% on cost and has seen the price rise 75%. A buyer today receives 2.0% and is betting on the price.

Every lens gives the same answer: TI is priced at 26 to 39 times whatever it earns, depending on whether you take the good year it is having or the better one analysts expect. The three lenses converge, and that is unusual. With a heavy builder like Microsoft or Amazon, earnings, owner earnings and free cash flow disagree because capital spending distorts the cash; here the building is over, the numbers agree, and they all say expensive.

Our method. We value TI the way it asks to be judged, on free cash flow per share. We estimate revenue in 2030 (including Silicon Labs), apply a free-cash-flow margin, divide by the shares, put a multiple on the result, discount it 4.1 years at 10%, and add the dividends received on the way (about $22 a share in present value).

The cases. "The cycle turns" has revenue in 2030 near today's twelve-month figure, as if the next downturn arrives on schedule and the data-centre boom fades: TI's free-cash-flow margin of 30% would still be respectable, and 18 times is a low but not a panic multiple. "A normal cycle" has revenue growing about 4% a year from this year's ~$22bn, slower than the 6% a year TI managed from the 2018 peak to the 2022 peak because 2026 is itself a strong year, with Silicon Labs added and the margin at its pre-build average. "Secular growth on top" has revenue compounding about 9% a year to $32bn, margins at a new high of 40% as the fabs fill, and the market paying 25 times.

The middle case moves to about $146 at 18 times and $202 at 26 times; to $156 if 2030 revenue is $23bn and $197 if it is $30bn; to $157 at a 32% margin and $191 at 40%. Even generous combinations stay below today's price. On analysts' own 2028 consensus of $13.50 a share, at 22 times and discounted back, we get about $254.

What is priced in? At $303.60, with the margin at 40% and a multiple of 25 times, TI needs revenue of about $38bn in 2030: growth of about 14% a year from this year, for four years, in a business that grew 3% a year over the last decade. With our middle-case margin and multiple, it needs $48bn. The price does not pay for a normal cycle. It pays for the upswing of 2026 to continue to the end of the decade with no downturn in between.

Margin of safety. TI is a wonderful business that we would happily own for twenty years, in the second tier of our scale: "generally excellent", which asks for a discount of 30–40% to a central value. Today's price is about 74% above our central value and 16% above our best case. We will not ask for a bargain on a business this good: if the price came back to where it is fair, we would start. That is about $190: a 3.2% yield, the top of TI's ten-year range; about 16 times 2027 consensus; and, for what it is worth, roughly where the shares were in March 2026, eight months ago.