STMicroelectronics N.V. STM

Key data

The business

ST is a broad-line chipmaker: analog parts, power devices, microcontrollers, and the tiny MEMS sensors that feel motion and pressure. Roughly half of what it sells goes into cars, a big chunk into factory and industrial gear, the rest into phones, gadgets, and now data centers. The thing a customer holds is unglamorous and sticky: the power module inside an electric drivetrain, the sensor in a braking system, the microcontroller running a washing machine. These parts get designed into a product for years, so a socket won in 2024 is still paying in 2029. That is the moat, a catalog of tens of thousands of designed-in positions inside customers' own products, and ST owns its fabs rather than renting them, which is a strength when volume is high and an anchor when it is not. It is a European company, listed in the US and on Euronext, run from Geneva.

The numbers

The shape here is a classic broad-line semi cycle: a monster peak in 2023, a two-year hangover as customers burned inventory, and a bottom in 2025.

Revenue fell 32% from the 2023 peak to the 2025 trough and profit all but vanished, $166 million on $11.8 billion of sales.

Q2 2026 revenue rose 26% year over year and 12.7% from the prior quarter, and operating margin came back to 5.4% from a 4.8% loss a year earlier. The GAAP diluted print of $0.24 sits below the $0.31 the Street tracks on its adjusted basis, but the direction is not in doubt. The lens that matters for a company like this is how far margin has to travel to justify the multiple.

Here is the tension in the price. Analysts expect revenue to pass the old peak by 2028, about $19.4 billion, yet they pencil earnings of $3.87 a share, still short of the $4.46 ST earned in 2023. That gap is the market saying the margin does not fully come back: silicon carbide is still dilutive, pricing has slipped, and the euro is a headwind. At $52.24 you pay about 14 times those 2028 earnings and roughly 103 times the trailing year's trough. What this memo believes that the crowd is split on is narrower than the cycle: the recovery is visible to everyone, so the only thing that decides whether 14 times is cheap or a trap is the gross margin path, and the print that settles it is the next two or three quarters of gross margin against a still-heavy 13.6% of sales going to capex.

Management

CEO Jean-Marc Chery is spending the trough reshaping the factory base, moving silicon to 300mm in Italy and France and silicon carbide to 200mm in Italy and Singapore, while resizing the cost base, a program begun in 2025. That is the right thing to do into a recovery and the expensive thing to do if the recovery stalls, because idle new capacity turns straight into unused-capacity charges, exactly what crushed 2025 margins. Buybacks have run $346 to $367 million annually from 2022 to 2025, modest against a $46.6 billion company and dwarfed by capex. The vendor feed shows no open-market insider buys or sells in the last twelve months, so there is no signal to read either way. The test is execution on the fab transition, not the pay plan.

How it fails or surprises you

The recovery re-rolls. If auto and industrial customers pause again, or EV demand softens, revenue stalls near $13 to $14 billion and the new 300mm and silicon carbide lines run half full. Unused-capacity charges return and gross margin slips back toward 33%. The tell is the quarterly revenue guide and book-to-bill turning down, with margin following within a quarter.

Margins never make it home. This is the fact the case for owning it explains least. The Street's own 2028 numbers say revenue tops the prior peak while earnings do not, a structural loss of about six margin points. If gross margin stalls in the mid-thirties while silicon carbide stays a money loser, 2028 earnings land a third light and 14 times becomes 20-plus times on what you actually get.

Data center scales (right tail). ST raised its 2026 data center revenue ambition to about $1 billion, from "nicely above $500 million," and said it could double in 2027 on AI server power, analog, and cloud optical interconnect. That is under 10% of sales today and priced as a maybe. If it tracks to $1 billion and doubles, it adds a high-growth leg the multiple ignores. The tell is quarterly data center revenue confirming the $1 billion run rate.

Closing thoughts

The market has priced the trough and some of the recovery, so the edge is not in betting the cycle turned, everyone can see that Q2 turned. The edge is in the gross margin. If ST rebuilds it toward the high thirties as volume fills the new fabs, the 2028 earnings the Street pencils in are real and 14 times them is genuinely cheap for a socket-durable analog and power franchise. If margin sticks in the mid-thirties because silicon carbide keeps bleeding and pricing keeps slipping, those earnings come in a third short and you overpaid at the top of a fake recovery. Two or three gross-margin prints tell you which, and the downside tail is fatter today because the fab spend is committed while the demand is not.

The bet is still that the world's cars and factories start buying STMicroelectronics chips again, after two years of working off old stock, and that a fresh pile of orders for AI data center power and connection parts lands on top of that. What breaks it is a margin that never climbs back while the company keeps pouring cash into 300mm and silicon carbide capacity. Watch gross margin against capex as a share of sales: if the first is stuck near 34% while the second stays above 13%, the recovery is funding someone else's price war, not ST's earnings.