STMicroelectronics N.V. (STMPA.PA) - Deep Dive Research Report

Prepared 24 July 2026. Listing: Euronext Paris (STMPA.PA), Euronext Milan (STMMI.MI), NYSE (STM). Reporting currency: US dollars. Fiscal year ends 31 December.

1. What the Company Does

STMicroelectronics makes the semiconductors that let physical things sense, move, decide, and communicate. If you press a fingerprint sensor on a phone, adjust the electric power steering in a car, run the inverter that drives a heat pump, read a tyre-pressure sensor, or move data through an AI data centre over light instead of copper, there is a reasonable chance an ST chip is doing part of that work. ST is not a maker of the giant logic processors that Nvidia or TSMC are famous for. It sits one layer down and one layer to the side: analog chips, power transistors, microcontrollers, motion and image sensors, and, increasingly, silicon photonics. These are the unglamorous but unavoidable components that turn a design into a working product.

ST is one of a small handful of companies in the world that both designs these chips and owns the factories that build them. That combination - an "integrated device manufacturer," or IDM - is the spine of the whole business. Most of the semiconductor industry has split into fabless designers (who own no factories) and foundries (who own only factories). ST deliberately kept both. It runs wafer fabs in Crolles (France), Agrate and Catania (Italy), and other sites, and it decides itself which of its chips to build in-house and which to outsource. Owning the fabs is expensive and cyclical, but it gives ST control over specialised process technologies - silicon carbide power devices, BiCMOS for RF, MEMS sensors, and now silicon photonics - that are hard to buy off a menu from a generic foundry.

The company was born from a 1987 merger of the semiconductor arms of two national champions: Italy's SGS Microelettronica and France's Thomson Semiconducteurs. That Franco-Italian parentage still matters today. The Italian and French governments, through a holding structure, remain the largest shareholders, and the company's fabs and jobs in Crolles, Agrate and Catania are politically load-bearing. This is why ST's restructuring decisions - closing legacy 200mm fabs, resizing the workforce - are negotiated carefully and slowly rather than executed the way a US chipmaker might.

For most of its modern history ST was known primarily as an automotive and industrial chip supplier. It rode the wave of electric vehicles and factory automation through 2022-2023, then fell into a deep downturn in 2024-2025 as carmakers and industrial customers worked down bloated inventories. The most important thing happening at ST right now is a pivot layered on top of that recovery: the company is repositioning around the AI data centre. Its PIC100 silicon photonics platform - chips that convert electrical signals into light for 800-gigabit and 1.6-terabit optical connections inside AI clusters - has become the single most important growth story in the company, anchored by a multi-year, multi-billion-dollar commercial engagement with Amazon Web Services announced in February 2026.

"AI data center growth is accretive to our gross margin." - CEO Jean-Marc Chery, Q2 2026 earnings call, 23 July 2026

That one sentence captures the strategic hope: ST is trying to graft a high-value, structurally-growing AI franchise onto a broad, cyclical analog-and-power base, and to do it while shrinking its cost structure at the same time.

2. Business Segments

ST reorganised in 2024-2025 into two Product Groups containing four reportable segments. VIPower products moved from Power & Discrete into Analog products, MEMS and Sensors effective 1 January 2025, so any comparison across that boundary needs care. The two Product Groups are APMS (Analog, Power & Discrete, MEMS and Sensors) and MDRF (Microcontrollers, Digital ICs and RF products).

APMS Product Group - Segment 1: Analog Products, MEMS and Sensors (AM&S)

This segment holds ST's analog signal-chain chips, its MEMS sensors and actuators (accelerometers, gyroscopes, microphones, pressure sensors), its optical sensing solutions (time-of-flight sensors used for autofocus and proximity detection), and, since the transfer, its VIPower smart-power devices.

The core capability here is MEMS. ST is one of the two or three largest MEMS makers in the world, and MEMS is genuinely hard: it involves micromachining moving mechanical structures - tiny beams and membranes that flex - directly into silicon, then packaging them so they survive being dropped, heated, and vibrated for years. ST has decades of accumulated process knowledge here, and it is the reason ST chips sit inside iPhones and other flagship consumer devices. The February 2026 acquisition of NXP's MEMS sensor business deepened this segment specifically in automotive safety sensors (the inertial sensors that trigger airbags and stability control), adding design wins at major carmakers.

Why it exists separately: MEMS and optical sensing are a different manufacturing discipline from digital logic, and they serve a customer base - consumer electronics OEMs and automotive Tier 1s - that buys on sensor performance and reliability, not compute. Within the group this segment is a swing factor tied to the personal-electronics cycle: a strong smartphone build season lifts it, a weak one drags it.

APMS Product Group - Segment 2: Power and Discrete products (P&D)

This is ST's power semiconductor business: discrete transistors, diodes, IGBTs, and - the strategic jewel - silicon carbide (SiC) power devices. SiC is a wide-bandgap material that handles higher voltages and temperatures with lower losses than ordinary silicon, which makes it the enabling technology for EV traction inverters, fast chargers, and industrial power conversion.

The core capability is that ST is one of the few vertically-integrated SiC players: it went as far as controlling substrate supply and building dedicated SiC fabs, notably the 200mm SiC line in Catania, Italy. SiC took ST more than a decade to industrialise, and the barrier to entry is real - substrate quality, defect density, and yield are extremely hard to master. The segment is now moving from 6-inch to 8-inch (200mm) wafers, which cuts cost per device.

Why it exists separately: power devices are a distinct process (thick wafers, high-voltage design rules) and a distinct end market (EV powertrain, industrial, renewable energy). This segment is the one most exposed to the EV demand cycle, which cooled sharply in 2024-2025 as EV adoption growth slowed and carmakers pushed out orders. Management has guided SiC back to double-digit growth in 2026.

MDRF Product Group - Segment 3: Embedded Processing (EMP)

This segment is built around the STM32 microcontroller franchise - general-purpose 32-bit microcontrollers that are among the most widely-designed-in MCUs in the world - plus automotive microcontrollers, connected-security chips (secure elements for payment, identity, IoT), and custom automotive ADAS processing.

The core capability is the STM32 ecosystem. A microcontroller is only as valuable as the software, tools, and engineer familiarity around it, and ST spent fifteen-plus years building a developer ecosystem so deep that STM32 is a default choice for a generation of embedded engineers. That ecosystem is a genuine switching-cost moat: once a product is built on STM32 with its libraries and toolchain, moving to a competitor means re-engineering firmware. This segment is the closest thing ST has to a recurring, sticky franchise.

Why it exists separately: microcontrollers and digital ICs are a logic-centric business (design-tool-heavy, ecosystem-driven) rather than an analog/power one, and they touch a vast fragmented base of industrial and consumer customers rather than a concentrated set of Tier 1s.

MDRF Product Group - Segment 4: RF and Optical Communications (RFOC)

This is the smallest but fastest-changing segment, and it is where ST's AI future sits. RFOC contains space and satellite-hardened products, ranging and connectivity chips, digital audio and signalling, and - critically - the PIC100 silicon photonics platform for optical interconnect.

The core capability is combining silicon photonics with BiCMOS on a 300mm platform. Silicon photonics means building optical components - waveguides, modulators, photodetectors - directly into a silicon chip so that data can be sent as light. ST claims PIC100 is the only 300mm silicon platform supporting 200 Gbps per lane. This is the technology feeding the AWS engagement and the 800G/1.6T optical transceiver market that AI data centres are consuming voraciously.

Why it exists separately: photonics and space products are exotic, small-volume-high-value businesses with completely different customers (hyperscalers, satellite makers) from the rest of ST. Management is treating this segment as the strategic option that could re-rate the whole company.

3. Products and Business Detail

Silicon photonics (PIC100). The headline product family. PIC100 combines ST's silicon photonics and next-generation BiCMOS technologies on a 300mm silicon platform, supporting 800 Gbps and 1.6 Tbps optical modules for AI clusters. Reported waveguide losses are as low as 0.4 dB/cm for silicon and 0.5 dB/cm for silicon nitride, which matters because every fraction of a decibel lost is bandwidth and power wasted. ST is building PIC100 on its 300mm lines and plans to expand production capacity roughly fourfold by 2027, with further increases in 2028. It is developing through-silicon-via versions to enable denser near- and co-packaged optics. The named anchor customer is AWS; the broader market ST is chasing (pluggable optical modules for data centres) is projected by LightCounting to grow from about $15.5 billion in 2025 to more than $34 billion by 2030.

Silicon carbide power devices. ST's SiC MOSFETs and modules go into EV traction inverters, on-board chargers, and industrial power systems. The manufacturing story is the 200mm SiC fab in Catania and the transition from 6-inch to 8-inch wafers, which lowers cost per die and is the lever ST is pulling to defend margins as SiC pricing normalises.

STM32 microcontrollers. A vast catalogue spanning ultra-low-power, mainstream, high-performance, and wireless MCUs, plus automotive-grade Stellar MCUs for the software-defined vehicle. These are designed into everything from industrial sensors and appliances to drones and medical devices. The moat is the tool ecosystem (STM32Cube), the community, and the enormous installed base of engineers.

MEMS and sensors. Accelerometers, gyroscopes, microphones, pressure sensors, and time-of-flight optical sensors. The consumer versions go into smartphones and wearables; the NXP-acquired automotive line goes into safety systems. ToF sensors also feed proximity and gesture applications.

Automotive and industrial analog and power. VIPower smart-power switches, gate drivers, and analog signal-chain devices for body electronics, powertrain, and factory automation.

Space and connectivity. Radiation-hardened components for satellites (relevant to the low-Earth-orbit constellation build-out management flagged), and ranging and connectivity chips.

Manufacturing footprint. ST runs front-end wafer fabs in Crolles and Rousset (France), Agrate Brianza and Catania (Italy), and Singapore, plus back-end assembly and test in Asia (Malaysia, Malta, Morocco, the Philippines, China). The strategic manufacturing programme now under way has three pillars: accelerate 300mm silicon at Crolles and Agrate (Crolles is being pushed toward 15,000 wafers per week to serve AI data centre and advanced-node demand, with 18/19nm and 14nm qualified on the 300mm platform), scale 200mm silicon carbide at Catania, and close legacy 200mm silicon fabs, with the full build-out targeted before 2028. A 14nm technology has been qualified with a key partner at a China facility. This reshaping is the source of both the restructuring charges hitting recent results and the cost savings promised for 2027.

4. Customers

ST sells to five broad buckets: automotive OEMs and their Tier 1 suppliers, industrial equipment makers, personal-electronics OEMs (smartphones, wearables, accessories), communications-equipment and computer-peripheral makers (the "CECP" end market that now includes hyperscaler AI infrastructure), and, increasingly, the hyperscale cloud providers directly.

The buying relationship differs sharply by customer type. In automotive, the decision-maker is a Tier 1 supplier's or carmaker's engineering and procurement team, the qualification cycle runs years, and once a chip is designed into a vehicle platform it stays there for the model's life. Cars must meet AEC-Q100 reliability grades and functional-safety standards, and re-qualifying a substitute part is so costly that incumbency is a powerful advantage. This is why ST's automotive revenue is sticky but also slow to grow or shrink. In personal electronics, the customer is a small number of very large OEMs (ST is a long-standing supplier of MEMS and sensors into flagship smartphones), the design cycles are annual, and volumes are enormous but the relationship is more contestable and price-sensitive. In industrial, the base is huge and fragmented, and STM32 microcontroller design-ins create long-tail stickiness through firmware lock-in.

The newest and most consequential customer relationship is with hyperscalers. The February 2026 AWS engagement is described as multi-year and multi-billion-dollar, tied to PIC100 silicon photonics for cloud and AI data-centre interconnect. This is a different animal from ST's traditional business: a concentrated, strategically-negotiated commitment with a single dominant buyer, where ST is being designed into the customer's own infrastructure roadmap. It is a quality signal (a hyperscaler validating ST's photonics) and a concentration risk at the same time.

Switching costs across the portfolio are highest in automotive (qualification and safety) and microcontrollers (ecosystem lock-in), moderate in power (design-in but more substitutable at the device level), and lowest in commodity discretes. Contract structure is a mix: automotive and hyperscaler business runs on multi-year supply agreements and design-win pipelines that give revenue visibility, while much of the industrial and consumer catalogue is book-and-ship through distribution, which is exactly what made the 2024-2025 inventory correction so violent - distributors and customers slammed orders shut when they found themselves overstocked.

5. Competitive Landscape

ST does not have one competitor; it has a different set in each segment, and it is a mid-to-large player in most rather than the clear leader in any.

In automotive and power, the primary rival is Infineon, the German leader that has been gaining automotive share (reaching roughly 13.7% of the automotive semiconductor market, versus ST around 10.2% and NXP around 10.8%). Infineon is stronger in automotive microcontrollers and power, and it is ST's most direct structural competitor. NXP is strong in automotive processors, radar, and connectivity. In silicon carbide, ST competes with Infineon, onsemi, and the struggling pure-play Wolfspeed, whose financial distress has actually improved the competitive backdrop for integrated players like ST. In analog, the giants are Texas Instruments and Analog Devices, both far larger and more profitable in catalogue analog, though less focused on ST's MEMS and photonics niches. In microcontrollers, ST's STM32 competes with Renesas, NXP, Microchip, and Infineon, but its developer ecosystem gives it an unusually defensible position. In MEMS, the key rivals are Bosch (the MEMS volume leader, private) and, historically, NXP - whose MEMS business ST just bought. In the new silicon photonics / optical arena, ST is the challenger against Broadcom (the scale leader in optical DSPs and co-packaged optics), Marvell, Coherent, Lumentum, and the GlobalFoundries Fotonix foundry platform that supplies many transceiver makers.

Where ST wins: vertical integration in SiC and MEMS, the STM32 ecosystem, and now a differentiated 300mm silicon photonics platform with a marquee hyperscaler customer. Where ST is exposed: it is sub-scale in pure analog against TI and ADI, it trails Infineon in automotive, and its owned-fab model carries high fixed costs that crush margins in a downturn (as 2024-2025 showed) even as it provides control in an upturn.

Barriers to entry are high in the specialised processes (SiC substrates, MEMS micromachining, 300mm photonics) and low in commodity discretes. The structural shift to watch is the AI-data-centre optical build-out, which is inviting new entrants and heavy investment from far larger competitors - ST's photonics lead is real but not guaranteed to hold against Broadcom-scale spending.

6. Industry

ST sits in the broad analog, power, sensor, and embedded-processing layer of the semiconductor industry - the part that interfaces with the physical world rather than the leading-edge logic that powers CPUs and GPUs. Demand is driven by three long secular forces plus one new one. The secular forces are vehicle electrification and the software-defined car (more silicon content per vehicle every year), industrial electrification and automation (motor drives, factory robots, renewable power conversion), and the proliferation of sensing and connectivity in consumer devices. The new force is the AI data centre, which needs vast quantities of optical interconnect and power management, and which is now ST's fastest-growing demand pool.

The automotive semiconductor market alone was worth roughly $107 billion in 2026 and is projected to grow at about 6.7% annually toward roughly $149 billion by 2031. The data-centre optical module market, ST's photonics target, is forecast to more than double from about $15.5 billion in 2025 to over $34 billion by 2030.

This industry is deeply cyclical, and ST just lived through a textbook cycle. The 2021-2022 shortage led customers to over-order and build inventory; when demand normalised in 2024, that inventory had to be worked off, and ST's revenue fell sharply (full-year 2025 revenue declined 11.1% to $11.80 billion, and the company called Q1 2025, at $2.52 billion, the bottom). The recovery from that trough has been powered less by a broad automotive/industrial snap-back (automotive stayed sluggish well into 2026) and more by the AI-driven communications and computing-peripherals demand, which is exactly why the mix shift toward photonics matters so much.

The industry is shaped by government policy as much as by end demand. Europe's Chips Act, national subsidies for fabs in France and Italy, and the strategic desire to keep advanced manufacturing on-shore all support ST's capex and constrain how quickly it can cut costs. Chips must meet automotive reliability and functional-safety certifications, which raises barriers and protects incumbents. ST occupies a mid-stream position in the supply chain: it buys wafers and materials, designs and fabricates its own chips, and sells to OEMs and Tier 1s, with less exposure to the leading-edge foundry bottleneck than the logic industry but full exposure to the capital intensity of owning fabs.

7. Growth Triggers

All triggers below are drawn from management statements on ST's earnings calls. Numbers here are management's own forward references, not analysis.

• AI data centre revenue scaling from above $1 billion in 2026 to well above $2 billion in 2027, driven by 800G and 1.6T optical transceiver adoption using silicon photonics and microcontrollers. (Q2 FY26 call, 23 July 2026)

  "Well above $2 billion" in 2027, driven by 800 gig and 1.6 terabit per second optical transceiver adoption. - management, Q2 2026 call

• Q4 2026 revenue expected above $4 billion, driven by AI data centres and low-Earth-orbit satellite communications. (Q2 FY26 call, 23 July 2026)

• Cumulative space revenue well above $3 billion for 2026-2028, tied to the LEO satellite build-out. (Q2 FY26 call, 23 July 2026)

• AWS multi-year, multi-billion-dollar commercial engagement for PIC100 silicon photonics in cloud and AI infrastructure, announced February 2026; high-volume PIC100 production began March 2026. (Q1 FY26 and Q2 FY26 calls; February/March 2026 announcements)

• Silicon carbide returning to double-digit growth in 2026 versus 2025, with the 6-inch to 8-inch wafer transition and book-to-bill well above one. (Q2 FY26 call, 23 July 2026)

• 300mm capacity ramp at Crolles toward 15,000 wafers per week, with 18/19nm and 14nm qualified on the 300mm platform and full build-out targeted before 2028. (Q2 FY26 call, 23 July 2026)

• NXP MEMS sensor acquisition closed February 2026, adding automotive safety-sensor design wins at major OEMs. (Q1 FY26 and Q2 FY26 calls)

• Restructuring cost savings in the high-triple-digit-million-dollar range by end of 2027, from reshaping the manufacturing footprint and resizing the global cost base. First detailed on the Q2 FY25 call and repeated on every call since. (Q2 FY25 call, 24 July 2025; repeated through Q2 FY26)

• 2028 revenue ambition of $18 billion, contingent on completing the manufacturing-efficiency programme. (Q2 FY26 call, 23 July 2026)

• 2026 net capex at the high end of the $2.0-$2.2 billion range, focused on cloud optical interconnect and advanced-node manufacturing. (Q4 FY25 and Q2 FY26 calls)

8. Key Risks

AI-photonics concentration and execution risk. The most important growth narrative rests heavily on a single anchor customer (AWS) and a single new product platform (PIC100). If hyperscaler optical roadmaps shift, if a larger competitor like Broadcom out-invests ST, or if PIC100 yields or capacity ramps disappoint, the entire re-rating thesis weakens. Management is guiding to "well above $2 billion" of AI data centre revenue in 2027; that is a large number resting on a young franchise and a concentrated customer base. The mechanism: photonics is being priced into the stock as a structural growth engine, so a stumble there hurts far more than a stumble in a mature analog line.

Owned-fab operating leverage cuts both ways. ST's IDM model means high fixed costs. In the 2024-2025 downturn this produced unused-capacity charges (roughly 290 basis points of gross margin drag flagged for Q4 2025) and drove operating margin to just 1.5% for full-year 2025 with an outright operating loss in Q2 2025. Any renewed demand air-pocket - a stalled automotive recovery, a consumer-electronics slump - would immediately re-load those under-utilisation charges. Management itself framed the manufacturing reshaping as necessary precisely because the cost base was too heavy for the revenue.

Automotive and EV cycle staying weak. Automotive is ST's traditional anchor, and it lagged through the recovery (Chery noted automotive was below expectations in Q4 2025). If EV demand growth stays soft and carmakers keep pushing out SiC orders, the P&D segment's promised double-digit SiC rebound could slip, and a large slice of ST's base revenue underperforms.

Restructuring execution in a politically constrained footprint. The savings depend on closing legacy 200mm fabs and resizing the workforce in France and Italy, where the French and Italian states are major shareholders and where plant closures are politically fraught. The savings target (high triple-digit millions by end 2027) is credible only if these socially and politically sensitive actions land on schedule. Delay is the base risk, not failure.

Margin normalisation in silicon carbide. SiC pricing is under pressure as capacity floods in and EV growth slows, with a distressed Wolfspeed and aggressive Chinese suppliers in the mix. The 8-inch transition is ST's cost defence, but if pricing falls faster than costs, the SiC margin story deteriorates.

Currency and cost base mismatch. ST reports in dollars but carries a large euro-denominated cost base (European fabs and workforce). A strengthening euro inflates costs against dollar revenue, a structural squeeze distinct from ordinary forex noise given how concentrated ST's manufacturing is in the eurozone.

9. Walk the Talk

The six calls used for this assessment: Q1 2025 (24 April 2025), Q2 2025 (24 July 2025), Q3 2025 (23 October 2025), Q4/FY 2025 (29 January 2026), Q1 2026 (23 April 2026), and Q2 2026 (23 July 2026). The most recent is within one day of this report.

The story across these six calls is one of a management team that called the bottom correctly, then delivered a recovery that ran ahead of its own conservative framing, while the profitability recovery lagged the revenue recovery - exactly as it had warned.

On the Q1 2025 call, with revenue at $2.52 billion (down 27.3% year-over-year), management explicitly called the quarter the trough of the cycle and confirmed the company-wide manufacturing reshaping programme. That was a specific, falsifiable claim. It held: every subsequent quarter came in higher, so the bottom call was correct.

On the Q2 2025 call, management put hard numbers on the restructuring for the first time, promising annual cost savings in the high-triple-digit-million-dollar range by the end of 2027, and took a $190 million impairment and restructuring charge that quarter. This is the commitment to track. It has been restated consistently on every call since without dilution or quiet abandonment, which is a good sign - management is not backing away from a hard promise as it becomes politically inconvenient.

Through Q3 2025 ($3.19 billion, book-to-bill above one) and Q4 2025 ($3.33 billion), the recovery in orders showed up roughly where management said it would, though the composition was not what they expected: personal electronics and communications/computing peripherals led, while automotive - the segment management had counted on - came in below expectations in Q4. That is a mild credibility ding: they got the direction right but leaned on the wrong end market.

The Q1 2026 call ($3.10 billion, up 23% year-over-year) is where the accuracy of management's guidance shows well. Recall that on the Q4 2025 call they had guided Q1 2026 to roughly $3.04 billion; the actual $3.10 billion came in above the midpoint, and Chery attributed the beat to engaged customer programmes in personal electronics and CECP plus the newly-closed NXP MEMS acquisition. Then on that same Q1 2026 call they guided Q2 2026 to about $3.45 billion at the midpoint.

Q2 2026 outlook: net revenues of $3.45 billion at the midpoint. - management, Q1 2026 call, 23 April 2026

The Q2 2026 result landed at $3.49 billion, again above the midpoint they had set three months earlier. Across the last three quarters, ST has beaten or met the top of its own guidance, which suggests the current management is guiding conservatively and delivering slightly ahead - the profile you want.

The one place to keep management honest is profitability. Revenue recovered fast, but the earnings recovery has been slow and lumpy: operating income was just $70 million in Q1 2026 and $187 million in Q2 2026, still well below the revenue trajectory, because the cost base is still being reshaped and unused-capacity charges linger. Management never promised a fast margin snap-back - they consistently framed 2025-2027 as a manufacturing-transition period - so this is a kept promise, not a broken one, but the $18 billion / margin ambition for 2028 remains the large unproven claim.

Assessment: this is a management team that does roughly what it says. It called the bottom accurately, has been steadily beating its own near-term revenue guidance, and has held to a hard restructuring commitment without walking it back. The credibility caveats are that the recovery came from a different end market than they forecast, and the big prizes (the 2028 $18 billion ambition and sustained AI-photonics scaling) are still promises, not results.

10. Shareholder Friendliness Index

Dividends. ST pays a cash dividend in quarterly installments. The annual dividend was set at $0.36 per share (paid as $0.09 per quarter), an increase from earlier years, and the company has held that $0.36 annual level flat across the 2024, 2025, and 2026 distribution cycles rather than cutting it through the downturn - notable given that full-year 2025 net income fell to $166 million and the payout consumed a large share of earnings in the trough year. Holding the dividend flat through a cyclical bottom, rather than cutting, is a shareholder-friendly signal, but it was not raised, so this is maintenance rather than growth.

Buybacks and dilution. ST has run share repurchase programmes alongside the dividend, executing buybacks in tranches through 2025 (for example, roughly $91 million repurchased in one third-quarter 2025 window at weighted-average prices in the roughly EUR 19-26 per share range, funded under a multi-year authorisation). The repurchases have been modest in scale relative to the company's size and were used partly to offset equity-compensation dilution rather than to shrink the share count aggressively, so shares outstanding have been broadly stable rather than materially declining. The MoatMap disclosure database recorded zero buyback filings for ST in the trailing ~90-day window (since 25 April 2026), consistent with repurchase activity being paced and programme-based rather than continuous; this 90-day gap is not evidence of any three-year absence, and the older 2024-2025 programme activity described above is the relevant multi-year record.

Verdict: Neutral - ST returns capital through a defended-but-flat dividend and modest, dilution-offsetting buybacks, prioritising heavy fab reinvestment (capex of $2.0-2.2 billion in 2026) over aggressive shareholder returns.

11. Insider Activities

The insider record for ST (Euronext, under EU Market Abuse Regulation Article 19 PDMR rules) over the last 12 months is dominated by a concentrated burst of selling by members of the ST executive committee in May 2026, with no open-market purchases at all. The MoatMap disclosure database records 26 transactions across 6 distinct insiders: 0 buys, 17 sells, and 9 "other" (grant/vesting) entries.

The pattern is characteristic of scheduled equity-compensation vesting, not a discretionary bet against the company. The "Other" rows at a €0 / $0 price are the delivery of vested restricted or performance share units, and the paired "Sold" rows immediately after are the executives selling a portion of those newly-vested shares. The timing - clustered tightly in mid-to-late May 2026, right after the annual shareholder meeting and equity-vesting window - and the fact that each sale is preceded by an equal-or-larger grant on the same day, both point to routine post-vesting disposals rather than conviction selling. The sales also happened into a rising share price as the AI-photonics story lifted the stock, the natural moment for executives to monetise vested equity.

(Source: EU MAR Article 19 PDMR notifications, via MoatMap multiverse disclosure database, all May 2026. AMF "Declarations des dirigeants" is the primary French register for these filings.)

Buys: There were none. No open-market purchases by any director or officer in the window.

Sells - the why: The most active seller by far was El Ouazzane, a senior member of the executive committee, who sold across five days in the 20-27 May window in both euro and dollar tranches - consistent with an executive diversifying a large vested-equity position, and the largest single seller. Cassis's 13 May block of about 70,000 shares (~$4.3 million) is the biggest individual disposal and, being un-paired with a same-day grant, looks like a discretionary sale of previously-vested stock. In every case the reason is not explicitly disclosed in the filing footnotes, but the structure - grants and sales on the same dates, tight clustering around the vesting window, no offsetting purchases - is the textbook signature of routine compensation-driven selling rather than a signal about business prospects.

Net assessment: Insiders were net sellers over the window, but the activity is narrow (six executive-committee members, one of whom accounts for eight of the transactions), grant-paired, and seasonally clustered in the annual vesting window. This is the noisy, low-information kind of insider selling, not a red flag. The genuine information would have been a buy, and there were none - so the honest read is neutral: no bullish conviction signal from insiders, but no alarming distribution either.

12. Scenarios

Bull case. The AI-photonics pivot works, and it re-rates the whole company. PIC100 ramps cleanly on ST's 300mm lines, the fourfold capacity expansion lands on schedule, and the AWS engagement proves to be the first of several hyperscaler relationships as 800G gives way to 1.6T optical modules across the industry. AI data centre revenue clears $2 billion in 2027 and keeps climbing, and the space and LEO-satellite business adds a second high-value leg. Underneath that, the traditional business recovers: automotive finally turns, silicon carbide returns to double-digit growth as the 8-inch transition lowers costs, and the STM32 franchise keeps grinding out sticky industrial revenue. The restructuring delivers its promised savings by end-2027, so the recovered revenue drops through to margin at a far higher rate than in the old cost structure. ST stops being a cyclical European analog house and becomes a structural-growth AI-infrastructure supplier with an owned-fab moat in photonics that Broadcom-scale spending cannot quickly replicate. The 2028 $18 billion ambition looks conservative.

Base case. ST continues the steady, guidance-beating recovery of the last several quarters without any single leg breaking out spectacularly. Revenue grinds higher quarter after quarter, led by communications and computing peripherals and the AI data centre ramp, while automotive recovers more slowly and unevenly than the bulls hope. Silicon carbide returns to growth but margins stay under pressure from oversupply. The restructuring lands roughly on time and roughly on target, so margins recover gradually as unused-capacity charges roll off, but the earnings recovery keeps lagging the revenue recovery. Photonics is a real and growing business but remains concentrated in a couple of customers and does not yet dominate the mix. Management keeps doing broadly what it says, the dividend stays flat, buybacks stay modest, and ST ends up a better-positioned, leaner version of the cyclical company it has always been, with a promising but still-unproven AI option embedded inside it.

Bear case. The AI-photonics story disappoints just as it has been priced in. Either PIC100 yields and capacity ramps slip, or Broadcom and the larger optical incumbents out-invest and out-compete ST in the transceiver market, or hyperscaler optical roadmaps shift in a way that strands ST's platform - and the AWS relationship, being concentrated, turns from a strength into a single point of failure. Simultaneously the cyclical base weakens: automotive stays soft as EV growth stalls, silicon carbide pricing collapses faster than the 8-inch transition can cut costs, and a consumer-electronics air-pocket re-loads unused-capacity charges onto ST's heavy fixed-cost base. The restructuring, dependent on politically-sensitive fab closures and workforce cuts in France and Italy, slips or gets diluted under state-shareholder and union pressure, so the promised savings arrive late and smaller. ST is left with a swollen European cost base, a cyclical downturn, and an AI growth story that failed to scale, and the operating losses of the 2024-2025 trough return.