STM | Earnings Review — Q2 2026

The quarter itself was a clean operating beat and a low-quality earnings beat. Revenue of $3,487m came in +0.8% above consensus ($3.46B) and +1.1% above the company's own guidance midpoint of $3.45B — the fifth consecutive quarter above the guided midpoint. Non-GAAP diluted EPS of $0.31 beat the $0.26 consensus by +19.2%, but gross margin landed exactly at the guided midpoint (34.8% U.S. GAAP, 35.2% non-GAAP) and non-GAAP OpEx of $960m was "in line with the expectation given in April." The entire EPS beat came from below the operating line: $41m of interest income on a $6.03B liquidity stack plus an 11.8% effective tax rate produced an $85m Q/Q below-the-line swing worth roughly $0.09/share against a $0.05 consensus beat. To management's credit, they never claimed an earnings beat on the call — they framed it as a revenue beat with in-line margins, which is the accurate framing.

The trajectory is the story, and it is unambiguous. Revenue Y/Y has gone (27.4)% → (14.4)% → (2.0)% → +0.2% → +23.0% → +26.1% across six quarters — +5,342bps of cumulative acceleration off the Q1'25 trough. Gross margin Y/Y inflected one quarter later, from (983)bps in Q3'24 to +136bps now. EBITDA compounded +105.1% Y/Y to $679m and operating income swung from $(133)m to $187m, a +1,017bps margin move. TTM revenue bottomed in Q3'25 at $11,791m and TTM Y/Y has now improved for five straight quarters to +10.5%. All four end markets are growing Y/Y for the first time this cycle, and three of four accelerated sequentially. The recovery is volume-led with pricing power returning — Q1'26 decomposition was volume +17% with ASP +6%, the first positive ASP quarter after (15)% in Q1'25.

New guidance: raised on revenue, hedged on margin. Q3'26 is guided to $3.70B ±350bps (+16.1% Y/Y, +6.2% Q/Q) at a 37.0% U.S. GAAP gross margin ±200bps — +220bps versus the prior guide and +377bps Y/Y. Q4'26 was newly framed as above $4.0B, H2-vs-H1 growth above the normal ~15% seasonality, and the FY2026 arithmetic floor is now ≥$14.28B (+21% Y/Y) against a $14.40B consensus. The Q3 revenue Y/Y decelerates 990bps — but that is a comp event (Q3'25 was 2025's strongest sequential quarter at +15.2% Q/Q), compounded by a deliberate, pre-announced Personal Electronics inversion to negative mid-single-digit Y/Y in both H2 quarters. It is not a demand event: book-to-bill was close to 2.0 overall and above 2 in Comm. Equipment & Computer Peripherals, backlog now covers 4.5–5.0 quarters of revenue, above 50% of Q2 bookings were for 2027, and distribution inventory fell below the standard target with DSI at 126 days versus 166 a year ago.

Eleven forward items were raised, one was hardened negative. The AI-datacenter ambition doubled on both horizons inside one quarter — CY2026 from $500m to $1,000m and CY2027 from $1B to $2B — with 2026 backlog 100% covered and a named mechanism (oversized MCU share on 800G/1.6T control planes, growing BiCMOS electronic-IC share, a steep 300mm silicon-photonics ramp at Crolles). SiC returned to Y/Y growth (+low-teens, +mid-30s Q/Q) with FY2026 guided to double-digit growth off booked design wins. Against that: FY2026 non-GAAP net OpEx was raised again to "slightly above $3.8B" (second upward revision in two quarters), net capex moved to the high end of $2.0–2.2B, and the ">40% gross margin at $4B/quarter" model was explicitly decoupled from the $4B trigger and re-dated to "end of 2027, not before."

That decoupling is the most important negative in the print, and it is a confirmed contradiction, not a nuance. In April, CFO Lorenzo Grandi said reshaping completion followed the 40% milestone. In July it became a precondition that must precede it — asserted as "we always said that there are 2 conditions," when the only prior on-record statement in eight quarters of transcripts says the opposite. Investors modelling 40% GM off the Q4'26 $4B revenue trigger are roughly six quarters early. Nine other contradictions were confirmed, including a "nonrecurring" reshaping cost now in its fifth consecutive quarter that is not excluded from the headline non-GAAP gross margin (60bps on $3,487m is ~$21m, but the entire GAAP-to-non-GAAP gross-profit bridge is only $14m), and a systematic pattern in which all four CEO-versus-press-release numeric disagreements ran in the flattering direction, zero exceptions.

Where the margin engine stands. The guided unused-capacity drag has deflated monotonically from 340bps (Q2'25) to 290bps to 220bps to 100bps to 70bps for Q3'26 — -270bps already harvested, only 70bps left, and Grandi said it will not fall further in Q4 because of a new China fab start-up, while the FX benefit turns neutral. From here gross margin has to be carried by mix (datacenter and LEO are both confirmed accretive) and by finishing the fabs — not by reloading.

Street Q&A was thin and asymmetric. Sixteen exchanges, eight analysts, 11 well answered / 5 deflected — and every one of the five deflections is a refusal to quantify magnitude or concentration on the new growth story: datacenter customer concentration behind "well above $2B in 2027," an unnamed LEO constellation "snag," Q4 gross-margin operating leverage, pricing timing, and the $20B/50% leg. Every question about the reported quarter got a straight numeric answer. Nobody asked about Power & Discrete, which grew +3.8% Y/Y and posted a fifth consecutive operating loss of $(99)m — a (21.4)% margin on the segment destroying the most operating income.

Upcoming catalysts. ~2026-10-22: Q3 print — the first test of the $3.70B / 37.0% guide and of the ~60% Comm/Computer and ~(5)% PE Y/Y claims. ~2027-01-28: Q4 — the above-$4B test and the first hard read on whether CY2026 datacenter cleared $1B. H2'26: new China fab start-up (the named reason unloading charges stop falling). End-2026: Sanan Chongqing SiC JV loading begins, gating the P&D turnaround. End-2027: reshaping program completion — two fab closures, 200mm→300mm silicon and 150mm→200mm SiC — the stated precondition for 40% gross margin and the single longest-dated, highest-consequence catalyst in the story.

| ($m, except EPS) | 2023Q3 | 2023Q4 | 2024Q1 | 2024Q2 | 2024Q3 | 2024Q4 | 2025Q1 | 2025Q2 | 2025Q3 | 2025Q4 | 2026Q1 | 2026Q2 | |---|---|---|---|---|---|---|---|---|---|---|---|---| | Net revenue | 4,431 | 4,282 | 3,465 | 3,232 | 3,251 | 3,321 | 2,517 | 2,766 | 3,187 | 3,329 | 3,095 | 3,487 | | Revenue Y/Y % | — | — | — | — | (26.6)% | (22.4)% | (27.4)% | (14.4)% | (2.0)% | +0.2% | +23.0% | +26.1% | | Gross profit | 2,109 | 1,949 | 1,444 | 1,296 | 1,228 | 1,253 | 841 | 926 | 1,059 | 1,172 | 1,045 | 1,215 | | Gross margin % | 47.6% | 45.5% | 41.7% | 40.1% | 37.8% | 37.7% | 33.4% | 33.5% | 33.2% | 35.2% | 33.8% | 34.8% | | GM Y/Y bps | — | — | — | — | (983) | (779) | (826) | (662) | (454) | (252) | +35 | +136 | | D&A | 396 | 414 | 430 | 438 | 441 | 451 | 428 | 464 | 482 | 480 | 454 | 492 | | Operating income | 1,241 | 1,023 | 551 | 375 | 381 | 369 | 3 | (133) | 180 | 125 | 70 | 187 | | Operating margin % | 28.0% | 23.9% | 15.9% | 11.6% | 11.7% | 11.1% | 0.1% | (4.8)% | 5.6% | 3.8% | 2.3% | 5.4% | | Op margin Y/Y bps | — | — | — | — | (1,629) | (1,278) | (1,578) | (1,641) | (607) | (736) | +214 | +1,017 | | EBITDA (OI + D&A) | 1,637 | 1,437 | 981 | 813 | 822 | 820 | 431 | 331 | 662 | 605 | 524 | 679 | | EBITDA margin % | 36.9% | 33.6% | 28.3% | 25.2% | 25.3% | 24.7% | 17.1% | 12.0% | 20.8% | 18.2% | 16.9% | 19.5% | | EBITDA Y/Y % | — | — | — | — | (49.8)% | (42.9)% | (56.1)% | (59.3)% | (19.5)% | (26.2)% | +21.6% | +105.1% | | Net income (loss) | 1,092 | 1,081 | 514 | 354 | 361 | 336 | 59 | (94) | 239 | (24) | 43 | 225 | | Diluted EPS (GAAP) | $1.16 | $1.14 | $0.54 | $0.38 | $0.37 | $0.37 | $0.06 | $(0.11) | $0.26 | $(0.03) | $0.04 | $0.24 | | Non-GAAP diluted EPS | — | — | — | — | — | — | — | $0.06 | $0.29 | $0.11 | $0.13 | $0.31 |

TTM cross-check: TTM revenue $13,098m versus prior-year $11,855m = +10.5%. H1'26 revenue $6,582m versus H1'25 $5,283m = +24.6%. Non-GAAP Q2'26 gross margin 35.2% versus 33.5% = +170bps Y/Y; non-GAAP diluted EPS +417% Y/Y. TTM FCF is still negative at $(261)m — management is spending into the acceleration.

Segment revenue and Y/Y — the growth is not uniform

| Segment ($m) | 2025Q1 | 2025Q2 | 2025Q3 | 2025Q4 | 2026Q1 | 2026Q2 | Y/Y | Direction | |---|---|---|---|---|---|---|---|---| | AM&S — Analog, MEMS & Sensors | 1,069 | 1,133 | 1,434 | 1,449 | 1,318 | 1,426 | +25.9% | accelerating | | P&D — Power & Discrete (incl. SiC) | 397 | 447 | 429 | 412 | 389 | 464 | +3.8% | inflected to growth | | EMP — Embedded Processing (STM32) | 742 | 847 | 976 | 1,015 | 975 | 1,147 | +35.4% | accelerating | | RFOC — RF & Optical Comms | 306 | 336 | 345 | 449 | 409 | 445 | +32.4% | high, marginally slower | | MDRF group (EMP + RFOC) | 1,048 | 1,183 | 1,321 | 1,464 | 1,384 | 1,592 | +34.6% | accelerating | | APMS group (AM&S + P&D) | 1,466 | 1,580 | 1,863 | 1,861 | 1,707 | 1,890 | +19.6% | accelerating |

MDRF is the growth engine and is now 45.7% of revenue versus 42.8% a year ago, compounding at +34.6% Y/Y and carrying $320m of segment operating income against APMS's $45m. P&D is the single genuine laggard: +3.8% Y/Y with a fifth consecutive operating loss of $(99)m — widening from $(84)m in Q1'26 and $(56)m a year ago, a (21.4)% margin. It is 13.3% of revenue but the entire reason group operating margin is 5.4% rather than double digits.

Why the trend is happening

1. The re-acceleration is volume-led with pricing power returning. Q1'26 decomposition: volume +17% with ASP +6% — versus volume (12)% / ASP (15)% in Q1'25. Distribution grew +33.1% Y/Y versus OEM +23.3% while distributor inventory fell below target — the highest-quality version of that mix shift, sell-through rather than channel fill.
2. Gross margin inflected because unused-capacity charges are unwinding. Actual charges: $103m (Q2'25) → $88m (Q4'25) → $69m (Q1'26) → $37m (Q2'26), i.e. 372bps → 106bps of revenue. Also suppressing reported GM: ~$11m/quarter of NXP MEMS PPA amortisation.
3. Operating leverage is the amplifier. D&A grew 77% over five years while revenue fell, so +26% revenue converted to +105% EBITDA and a $320m swing in operating income.
4. Mix is shifting to the better franchises — RFOC is where AI-datacenter silicon photonics and LEO satcom land, and the CE&CP end market is +50% Y/Y and still accelerating.
5. Caveat on the Q1'26 EPS optic. EPS was $0.04 versus $0.06 — down Y/Y despite +23% revenue — because Q1'25 carried an unusually favourable below-the-line/tax mix. Q2'26 is the cleaner data point.

Actuals: CE&CP, Industrial, Automotive, Personal Electronics for 2026Q2. Two markets accelerating hard, one holding at industry growth, one going negative on a memory-price shock ST does not control. Roughly half of the entire Q4 Y/Y dollar increase must come from CE&CP alone — that is the load-bearing assumption in the above-$4B number.

Where the EPS beat actually came from

| U.S. GAAP walk | Q4 2025 | Q1 2026 | Q2 2026 | |---|---|---|---| | Operating income | $125m | $70m | $187m | | Pre-tax income | $147m | $53m | $255m | | of which interest income | — | — | $41m | | Income tax (expense) | $(171)m | $(10)m | $(30)m | | Effective tax rate | 116% | 18.9% | 11.8% | | Non-GAAP net income | $100m | $122m | $291m | | Non-GAAP diluted EPS | $0.11 | $0.13 | $0.31 |

Q2'26 pre-tax income exceeded operating income by $68m, versus -$17m in Q1'26 — an $85m swing worth roughly $0.09/share on ~939m diluted shares, against a $0.05 consensus beat. Symmetrically, the (59.3)% Q4'25 EPS miss was a tax event, not an operating miss: a $(171)m tax charge on $147m of pre-tax income, in a quarter where revenue beat consensus by +1.2% and beat the company's own guide by +1.5%. Q2'26 GAAP operating income also absorbed $58m of impairment/restructuring/phase-out costs plus $24m of NXP MEMS PPA effect.

Beat rates and the pattern

| Basis | L4Q | L8Q | L12Q | |---|---|---|---| | Non-GAAP EPS vs consensus | 2/4 = 50% | 5/8 = 63% | 8/12 = 67% | | Revenue vs consensus | 3/4 = 75% | 5/8 = 63% | 5/9 = 56% (3 qtrs unsourced) | | Revenue vs own guidance midpoint | 4/4 = 100% | 6/8 beat + 2 in line | 83% at-or-above | | Gross margin vs own guidance midpoint | 3/4 at-or-above | 4/8 at-or-above | series starts 2024Q3 |

Two different companies depending on the yardstick — and that is the finding.

• Revenue: consistent beater, and improving. 100% of the last four quarters above the guided midpoint, five straight, average +1.2% L4Q versus +0.45% L12Q. Management sets a beatable revenue bar and clears it — exactly the guidance-accuracy signature you want.
• EPS: mixed, and the magnitude is deteriorating. 67% L12Q collapses to 50% L4Q. Average surprise walks +7.4% → +4.5% → (10.0)%, and this is not a small-base artifact: in absolute cents the L4Q net is -$0.10 versus +$0.04 in the prior four. The range across the last eight quarters is (59.3)% to +40.0%, driven almost entirely by tax rate and restructuring/PPA charges below the operating line.
• Gross margin: recently reliable. The last three quarters landed +21bps, +6bps, +4bps versus the guided midpoint, after averaging roughly -20bps in the four before that. Real improvement in forecasting discipline, consistent with the CFO now pre-flagging the 60bps reshaping drag rather than letting it surprise.

Read-through: EPS surprise is a poor signal for this name; revenue-versus-guidance is the honest one. Investors underwriting STM on EPS surprise are trading noise.

The new guidance, as issued (2026-07-23)

| Item | Guide | Source | |---|---|---| | Q3'26 net revenue, midpoint | $3.70B | Daloopa | | Revenue range (±350bps) | $3.571B – $3.830B | Daloopa | | Sequential growth at midpoint | +6.2% | Daloopa | | Q3'26 gross margin, midpoint (U.S. GAAP) | 37.0% ±200bps | Daloopa | | — of which unused-capacity charges | ~70bps | Daloopa | | EUR/USD assumption | 1.14 (vs 1.16 effective in Q2) | Daloopa | | Q3'26 non-GAAP net OpEx | ~$980m (~$920m ex start-up + share awards) | transcript | | Q3'26 non-GAAP gross margin | not given this quarter (≈37.3% implied) | disclosure dropped | | Q4'26 revenue | above $4.0B | Daloopa | | H2'26 vs H1'26 growth | above the normal ~15% seasonality | transcript | | Datacenter revenue, CY2026 | above $1.0B | Daloopa | | Datacenter revenue, CY2027 | well above $2.0B | Daloopa | | FY2026 non-GAAP net OpEx | slightly above $3.8B; like-for-like +high single digit | transcript | | FY2026 net capex | high end of $2.0B–$2.2B | transcript | | FY2028 model | $18B confirmed; 45% GM contingent on reshaping + FX | transcript | | GM above 40% at $4B/qtr | reaffirmed but decoupled from the $4B trigger — "end of 2027, not before" | transcript |

Structural note: ST guides exactly two line items one quarter ahead — net revenue (±350bps) and gross margin (±200bps). It does not guide EPS, operating margin or EBITDA. Every EPS or operating-margin figure below is either consensus or an explicitly-labelled bridge built from the guided components.

Q3'26 versus the prior guide

| Metric | Prior guide mid (Q2'26) | New guide mid (Q3'26) | vs prior | |---|---|---|---| | Net revenue | $3.45B | $3.70B | +$250m / +7.2% | | Revenue Y/Y at midpoint | +24.9% | +16.1% | -880bps (comp-driven) | | Revenue Q/Q at midpoint | +11.6% | +6.2% | -540bps | | Gross margin (U.S. GAAP) | 34.8% | 37.0% | +220bps | | — unused-capacity drag | 100bps | 70bps | -30bps (better) | | Gross margin (non-GAAP) | 35.2% | not guided (≈37.3% implied) | disclosure dropped | | Non-GAAP net OpEx | $950–960m | ~$980m | +$20–30m (worse) | | Operating margin | not guided | not guided (≈10.5% implied) | — | | Non-GAAP diluted EPS | not guided | not guided (≈$0.35–0.41 implied) | — | | EUR/USD | 1.15 | 1.14 | -0.01 |

Implied figures are bridges, not guides: $3.70B × 37.0% = $1,369m gross profit, less ~$980m non-GAAP net OpEx = ~$389m operating income (10.5% margin); on ~939m diluted shares and a 0–15% tax range that is ~$0.35–0.41 of non-GAAP EPS versus $0.31 delivered.

Consensus honesty flag. A Q3'26-specific street revenue or EPS consensus was not obtainable this run — FMP returned HTTP 402, Visible Alpha and Bloomberg are not connected. No Q3 consensus point estimate is asserted anywhere on this page. The street's Q4 split is derived from the FY2026E consensus below.

Revenue guidance ladder — prior guide to Q4 consensus

FY2026 — guidance-implied versus consensus

| Metric | Prior (at Q1'26) | New (at Q2'26) | Consensus | Direction | |---|---|---|---|---| | FY2026 revenue | "double-digit growth", no number | ≥$14.28B (arithmetic floor) | $14.40B | specified & raised | | FY2026 revenue Y/Y | "double digit" | ≥+21.0% | +22.0% | quantified | | FY2026 gross margin | "improve sequentially Q1→Q4" | ~36.0% (+~215bps Y/Y) | n/a | quantified | | FY2026 non-GAAP EPS | not guided | not guided ($1.29–$1.43 implied) | $1.34 | consensus mid-range | | FY2026 non-GAAP net OpEx | l-f-l +mid-to-high single digit | slightly above $3.8B, l-f-l +high single digit | n/a | raised (negative) | | FY2026 net capex | $2.0B–$2.2B | high end (~$2.2B) | n/a | raised (negative to FCF) | | Datacenter revenue CY2026 | $500m | $1,000m | n/a | +100% (2x raise) | | Datacenter revenue CY2027 | $1B | $2B | n/a | +100% (2x raise) |

FY2025 base: revenue $11,800m, gross profit $3,999m (33.89% GM), operating income $175m, diluted EPS $0.18. FY2026 consensus EPS of $1.34 less H1'26 non-GAAP actual of $0.44 implies $0.90 for H2, against a guided-component bridge of $0.85–$0.99. Consensus sits mid-range on guidance mechanics — achievable, not stretched. The FY2026 risk is margin execution, not revenue.

FY2027 deliberately not quantified. The circulating aggregator series is flagged LOW confidence and internally inconsistent (the same source carries FY2026E below reported H1 plus guided H2, which is arithmetically impossible), so it is not published as a number. What can be said without it: a Q4'26 exit rate above $4.0B/quarter held flat, with zero growth, annualises to ≥$16.0B — against management's confirmed $18B by 2028 and a datacenter line alone stepping from above $1B to well above $2B.

Y/Y growth: reported trend versus guided trend — the two lines cross

Reading the chart. Revenue Y/Y decelerates 990bps in Q3 (+26.1% to +16.1%), then re-accelerates at least 410bps in Q4. Two mechanical causes, neither a demand event: (i) the comp — Q3'25 was 2025's strongest sequential quarter, jumping $2,766m to $3,187m (+15.2% Q/Q), while Q4'25 added only +4.5% Q/Q; (ii) Personal Electronics inverts, guided to negative mid-single-digit Y/Y in both H2 quarters on a pre-announced change in the engaged-customer launch profile. Meanwhile gross-margin Y/Y expansion accelerates hard (+35bps → +136bps → +377bps) even as revenue Y/Y decelerates. The margin story runs on a different engine from the revenue story: mix plus the collapsing unused-capacity drag, not volume.

Guidance accuracy — the credibility test

| Guide issued | Applies to | Revenue guide mid | Actual | vs mid | GM guide mid | GM actual | vs mid | |---|---|---|---|---|---|---|---| | 2025-04 | Q2'25 | $2.71B | $2,766m | +2.1% | — | 33.48% | — | | 2025-07 | Q3'25 | $3.17B | $3,187m | +0.5% | 33.5% | 33.23% | -27bps | | 2025-10 | Q4'25 | $3.28B | $3,329m | +1.5% | 35.0% | 35.21% | +21bps | | 2026-01 | Q1'26 | $3.04B | $3,095m | +1.8% | 33.7% | 33.76% | +6bps | | 2026-04 | Q2'26 | $3.45B | $3,487m | +1.1% | 34.8% | 34.84% | +4bps | | 2026-07 | Q3'26 | $3.70B | pending | — | 37.0% | pending | — |

5 of 5 revenue guides beaten, by +0.5% to +2.1% (average +1.4%), with gross margin landing within ±27bps every single quarter and within ±10bps in the last two. Tight ranges, consistently modest beats, no manufactured blowouts. This materially raises the base-rate probability that the $3.70B Q3 midpoint is a floor rather than a stretch, and it is the strongest argument for reading the Q4 "above $4 billion" language as a genuine commitment.

The margin engine is deflating on schedule — and nearly spent

| Applies to | Guided unused-capacity drag | Actual charge | |---|---|---| | Q3'25 | 340bps | $102m | | Q4'25 | 290bps | $88m | | Q1'26 | 220bps | $69m | | Q2'26 | 100bps | $37m (106bps) | | Q3'26 | 70bps | pending |

A monotonic -270bps reduction in five quarters, against FY2025's total unused-capacity charge of $416m. This single line explains most of the +377bps Q3 GM Y/Y expansion — and only 70bps remains, with Grandi stating it will not fall further in Q4 because of a new China fab start-up.

Guidance-change ledger — eleven raises, one hardened negative

| Metric | Prior (Q1'26 call) | New (Q2'26 call) | Direction | |---|---|---|---| | Datacenter revenue, current year | $500m "nicely above" | $1,000m "above" | RAISED 2x | | Datacenter revenue, next year | $1B "well above" | $2B "well above" | RAISED 2x | | Q4 revenue | not given | above $4.0B | NEW | | H2 vs H1 growth | "the usual seasonality" (~15%) | above normal 15% | RAISED | | FY revenue | "double-digit growth" | ≥$14.28B, ≥+21% | SPECIFIED / RAISED | | Book-to-bill | "well above 1" (qualitative) | close to 2; above 2 in CE&CP; above 50% of Q2 bookings for 2027; backlog 4.5–5 quarters | NEW, MUCH STRONGER | | Distribution inventory | "normalized" | below standard target | IMPROVED | | DSI | 140 days | 126 days (vs 166 a year ago) | IMPROVED | | Pricing | "very low single-digit decline" expected | selected price increases, net neutral vs input costs | IMPROVED (but GM-neutral) | | SiC FY2026 | "will be a growth year" | double-digit growth; Q2 +low-teens Y/Y, +mid-30s Q/Q | FIRMED | | Supply conditions | ample | tight supply in several categories; rising MCU lead times | NEW (tightening) | | Reshaping GM cost | ~50bps/qtr, persisting | ~60bps/qtr, persisting | WORSE (modest) | | FY non-GAAP net OpEx, l-f-l | +mid-to-high single digit | +high single digit; slightly above $3.8B | RAISED (negative) | | FY net capex | $2.0B–$2.2B | high end of range | RAISED (negative to FCF) | | Q4 unused-capacity charges | expected to keep falling | will not fall further in Q4 — China fab start-up | NEW (negative) | | FX in guide | 1.15 assumed | 1.14; FX benefit turns neutral in Q4 | NEW (negative) | | GM above 40% at $4B/qtr | tied to $4B + reshaping complete | explicitly decoupled from $4B — "end of 2027, not before" | DECOUPLED (negative) | | FY2028 $18B model | $18B | $18B confirmed | UNCHANGED | | 45% GM at $18B / 50% at $20B | both cited | 45%/$18B confirmed (FX + reshaping contingent); $20B deflected | HEDGED |

Tone: a clear upgrade on demand, flat-to-more-hedged on margin

| Dimension | Q1'26 (2026-04-23) | Q2'26 (2026-07-23) | Shift | |---|---|---|---| | Demand framing | "despite the macroeconomic uncertainty, we saw improving demand" | "demand further accelerating, strong bookings in all end markets and improved visibility" | Materially stronger — the macro hedge is dropped entirely | | Forward specificity | "I will not guide on '26"; FY = "double-digit growth" | Q4 above $4B; H2/H1 above 15%; per-end-market Y/Y for both Q3 and Q4 | Much more specific — a confidence tell | | Visibility | book-to-bill "well above 1" | book-to-bill close to 2; backlog 4.5–5 quarters; above 50% of bookings for 2027 | Step-change in disclosed visibility | | Margin confidence | "will improve sequentially Q1, Q2, Q3 and Q4" — unqualified | Q4 improvement confirmed but triple-hedged: unloading flat (China fab), reshaping transfer costs persist, FX turns neutral | Flat to weaker — the notable divergence | | Capacity | ramping, no tightness language | "signs of tight supply", rising MCU lead times; admits a Q2 legacy-analog ramp delay that constrains Q3 | New constraint, candidly disclosed | | Long-term model | $18B/2028, 45% GM, 50% at $20B | $18B confirmed with "increased confidence"; $20B deflected | Core confirmed, stretch goal parked |

Confidence ranked by how much management will put behind it: highest on datacenter/optical (2x raise, named mechanism, 100% 2026 backlog coverage, "not right now gated by capacity expansion"); high on Q3 revenue and gross margin (5/5 record plus mechanically deflating unloading); medium on Q4 above $4B (needs CE&CP roughly doubling Y/Y and Industrial near +40% — supported by backlog but concentrated); lower on Q4 gross margin (guided only as "sequential improvement", immediately qualified by three named headwinds, magnitude deliberately declined); lowest on the 40% GM / $4B linkage, now conditional on a program that completes no earlier than end-2027.

Risk caveats management raised unprompted: tariffs excluded from the outlook; the Q2 legacy-analog ramp delay already constraining Q3; the China fab start-up holding unloading charges flat in Q4; customer qualification gating the 200mm→300mm and 6"→8" SiC transitions; and — new this quarter — "other constellations have faced a bit of a snag lately, to be fixed in the coming quarters" on LEO.

| Metric | Q3'24 | Q4'24 | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | |---|---|---|---|---|---|---|---|---| | Revenue Y/Y % | (26.6)% | (22.4)% | (27.4)% ← trough | (14.4)% | (2.0)% | +0.2% ← crosses zero | +23.0% | +26.1% | | Rev acceleration (bps) | (136) | +420 | (492) | +1,294 | +1,245 | +221 | +2,272 | +310 | | GM Y/Y (bps) | (983) | (779) | (826) | (662) | (454) | (252) | +35 ← GM crosses zero | +136 | | Op margin Y/Y (bps) | (1,629) | (1,278) | (1,578) | (1,641) | (607) | (736) | +214 | +1,017 | | GAAP diluted EPS | $0.37 | $0.37 | $0.06 | $(0.11) | $0.26 | $(0.03) | $0.04 | $0.24 | | EPS Y/Y % | (68.1)% | (67.5)% | (88.9)% | loss | (29.7)% | loss | (33.3)% | loss → profit |

TTM smoothing — the cleanest signal in the analysis

| | Q1'25 | Q2'25 | Q3'25 | Q4'25 | Q1'26 | Q2'26 | |---|---|---|---|---|---|---| | TTM revenue ($m) | 12,321 | 11,855 | 11,791 ← TTM trough | 11,799 | 12,377 | 13,098 | | TTM revenue Y/Y % | (25.3)% | (23.1)% | (17.1)% | (11.1)% | +0.5% | +10.5% | | TTM acceleration (bps) | — | +227 | +593 | +606 | +1,153 | +1,004 |

TTM revenue bottomed in Q3'25 and has expanded for three straight quarters; TTM Y/Y has improved for five consecutive quarters with no give-back. That is the single most reliable trajectory fact here.

Why the EPS row is the wrong lens

GAAP diluted EPS Y/Y swings ±100pp between adjacent quarters. Two specific non-operating items cause both artificial dips, and neither reflects demand:

| Quarter | GAAP EPS | Distortion | |---|---|---| | Q2'25 | $(0.11) | $(190)m of impairment, restructuring and closure costs — the largest charge of the eight quarters | | Q4'25 | $(0.03) | Pre-tax income was positive $147m, but a $(171)m tax expense (~116% effective rate) drove net income negative. Non-GAAP EPS was +$0.11. |

A full 8-quarter non-GAAP EPS Y/Y series does not exist in reported data — ST began non-GAAP disclosure in H2 2025 (Daloopa carries it from Q3'25 only). GAAP is used as the spine and the distortions are itemised rather than smoothed away.

Plain-English trajectory assessment

STM is one quarter past the peak rate-of-change of a textbook semiconductor cyclical recovery, and the recovery is broadening from a comp-driven rebound into demand-driven growth.

1. The trough was Q1 2025 — revenue $2,517m, Y/Y (27.4)%, maximum deceleration of -492bps. Management said so explicitly on the Q2'25 call: "we confirm that Q1 was a low point for Automotive revenues" and "confirming that Q1 was the bottom" for Industrial.
2. Five consecutive quarters of acceleration since, cumulatively +5,342bps in the Y/Y rate. Zero-crossing was Q4'25; the largest single step was Q1'26 at +2,272bps.
3. Q2'26's acceleration was modest in rate terms (+310bps) but the highest absolute growth rate of the cycle (+26.1%) and the highest revenue quarter since Q1'24. Crucially it is no longer a base effect — revenue is +38.6% above the trough level and TTM Y/Y finally reached +10.5%.
4. The next two quarters mechanically decelerate then re-accelerate. A headline Q3 Y/Y deceleration should not be read as a broken trajectory. But if Q4'26 fails to clear $4B, the five-quarter acceleration story breaks — that is the real forward risk, and it is a comp/execution risk, not a demand risk.
5. Margin, not EPS, is the confirming signal. GM Y/Y turned positive in Q1'26 (+35bps) and widened to +136bps — exactly the one-quarter lag you would expect as unused-capacity charges roll off. Q3'26 GM is guided ~+377bps Y/Y, a further widening. EPS acceleration should outpace revenue acceleration from here as operating leverage compounds on a still-depressed margin base.
6. The trajectory is not uniform. P&D grew +3.8% with a fifth straight operating loss. The acceleration is carried by EMP (+35.4%), RFOC (+32.4%) and AM&S (+25.9%). A quality-of-acceleration caveat, not a challenge to direction.

Verdict: accelerating, and accelerating for the right reasons. By the investing-principles test — a company growing 10% but accelerating beats one growing 15% but decelerating — STM is growing 26% and accelerating, off a five-quarter base of monotonic improvement, with margin Y/Y confirming the revenue signal one quarter later.

The one-line read: ST just did something management teams rarely do — it doubled the same forward ambition twice in three months. Against that, the sell-side reaction was that ST had "not done enough" (J.P. Morgan) and needed "more evidence of gross-margin improvement." That gap — management compounding a revenue catalyst while the street underwrites the margin catalyst instead — is the whole set-up. The catalyst that decides it is not the datacenter ramp (that is happening, and the backlog proves it); it is the manufacturing reshaping program, which management explicitly says does not finish until end-2027.

| # | Catalyst | Timing | Management position | Signal / evidence | Verdict | |---|---|---|---|---|---| | 1 | AI datacenter ramp — silicon photonics PIC100 + BiCMOS electronic ICs + MCUs for 800G/1.6T pluggables + power | Inflects Q4'26; step-function 2027 | CY2026 $1,000m (was $500m); CY2027 $2B "well above" (was $1B); 2026 backlog 100% covered | CE&CP Y/Y +50%, accelerating from +41% and -5% a year ago. Book-to-bill significantly above 2 in this vertical. PIC100 in high-volume production at Crolles on 300mm since Feb/Mar 2026, capacity to go 4x by 2027 | Highest conviction. Guided path +50% → ~60% → ~90% Y/Y = ~4,000bps of acceleration in two quarters, off a backlog, not a forecast | | 2 | Q4'26 revenue above $4B | reported ~2027-01-28 | $4B floor, not midpoint; H2/H1 above the normal 15% seasonality | Backlog 4.5–5.0 quarters of Q2'26 average revenue; above 50% of Q2 bookings for 2027; book-to-bill close to 2 | The near-term guide is the soft spot; the Q4 step is the payoff. Jefferies pinned the Q3 softness on a slower iPhone 18 ramp — mix, not demand | | 3 | Gross-margin recovery / reshaping completion | GM 37.0% Q3'26; program completes end-2027 | Two conditions for >40% GM: $4B revenue and completed 200mm→300mm silicon + 150mm→200mm SiC with two fab closures. "We are in the middle… it will be at the end of 2027, not before" | Unloading drag collapsed 340bps → 70bps. Reshaping cost ~60bps, held flat through 2026 | Most likely to disappoint. The easy tailwind is ~80% spent. Q4 GM improves off 37.0% but the slope flattens | | 4 | LEO satcom / new-space ramp | strongest 2027; cumulative through 2028 | "well above $3B cumulative space revenue over 2026-2028"; addressable market ~$3B by 2030, ~4x the 2025 level; confirmed GM-accretive | ST LEO-linked revenue ~$175m (2021) → ~$600m (2025) → approaching ~$1B (2026). 7.5bn Starlink chips delivered cumulatively. Second LEO customer ramping | Most under-modelled. Risk is disclosed not hidden: "other constellations have faced a bit of a snag lately" — launcher-dependent, so timing slips are the failure mode, not demand | | 5 | GP MCU tightness → lead times | live; runs through 2027 | Distribution inventory below standard target, POS "very strong", "tension on the supply and increasing lead time"; raising prices on selected products but input costs "more or less offsetting" | Industrial +34% accelerating. DSI 126 days vs 166. #1 worldwide GP-MCU vendor, 5th straight year (Omdia) | Genuine upcycle signal — but management called pricing net-neutral on GM. This is a volume catalyst, not a price catalyst. Do not double-count | | 6 | Personal Electronics reverses to a headwind | Q3 and Q4 2026 | PE Y/Y negative mid-single-digit in both quarters; FY low-to-mid single digit; blamed on low-end smartphone units "because of the memory price" | PE was +20% in Q2'26 — a ~2,500bps swing. Corroborated: mobile LPDDR5X ASPs +78–83% Q/Q in Q2'26; TrendForce cut 2026 smartphone output to -2% Y/Y | Only guided negative in the model. Well-telegraphed and quantified; risk is it worsens if memory inflation persists into 2027 | | 7 | SiC return to growth + 6"→8" / Sanan JV | double-digit CY2026; benefits land end-2027 into 2028 | SiC +low-teens Y/Y, +mid-30s Q/Q in Q2; book-to-bill "well above 1"; FY2026 double-digit growth off already-booked design wins | P&D is the laggard: $464m revenue (+3.8% Y/Y) and a 5th straight operating loss of $(99)m. Sanan Chongqing JV loading starts end-2026 | Cheapest optionality in the story. P&D going from $(99)m/quarter to breakeven is worth ~285bps of consolidated operating margin with zero revenue help. Not in anyone's model | | 8 | Physical AI / robotics — NVIDIA Halos | design-in now; revenue 2027+ | Named partner in NVIDIA Halos for Robotics (functional safety for industrial + humanoid robots) across MCUs, sensors, motor control, security | Announced 2026-06-23, building on the Q1'26 NVIDIA robotics collaboration. New industrial MEMS with embedded AI and a compact 3D LiDAR module launched in Q2 | Thematic, not near-term revenue. Evidence of design-win quality in Industrial, not a 2026-27 line item | | 9 | Agrate 300mm build-out; China-for-China | full build-out "pretty soon before 2028" | Agrate supports MCU growth "as soon as we have qualified the 19nm and 14nm technology"; China node qualified with the main partner | 2026 net capex now at the high end of $2.0–2.2B, raised at Q2 for cloud optical interconnect | Capex at the high end is a confirming signal on catalysts 1 and 4 — but it defers FCF (TTM FCF still $(261)m) | | 10 | Semiconductor tariffs (Section 232) | any time | Guidance explicitly excludes any further change to global trade tariffs | 25% Section 232 tariff live since 2026-01-15 on a narrow high-performance category; Taiwan/Korea carve-outs negotiated | The un-hedged tail risk. The guide is tariff-naked by management's own admission | | 11 | OpEx step-up | FY2026 | Non-GAAP net OpEx slightly above $3.8B, up low-double-digit Y/Y; like-for-like high single digit (raised from mid-to-high at Q1) | Prior Q3'25-era quarterly guide was $915m; Q3'26 guide ~$980m | A real, quantified negative. Some is discretionary AI/MEMS investment, some is transfer start-up cost that rolls off |

Dated catalyst calendar

| Date | Event | Why it matters | |---|---|---| | ~2026-10-22 | Q3 2026 results (quarter ends 2026-09-26) | First test of the $3.70B / 37.0% guide and of the ~60% CE&CP and ~(5)% PE Y/Y claims. One source prints 2026-10-29 instead — confirm with IR | | ~2027-01-28 | Q4 2026 results | The above-$4B test, the above-normal-seasonality test, and the first hard read on whether CY2026 datacenter cleared $1B | | H2 2026 | New China fab start-up | Named as the reason unloading charges stop falling in Q4 — a margin drag with a known cause | | End-2026 | Sanan (Chongqing) SiC JV starts loading | Gates the P&D turnaround | | Continuous H2'26 | LEO launch cadence | Revenue is launcher-dependent by management's own description; a launch slip is a revenue slip | | Any time | Section 232 scope changes | Guidance explicitly excludes tariff changes | | End-2027 | Reshaping program completion — 2 fab closures, 200→300mm silicon, 150→200mm SiC | The stated precondition for >40% GM. The single longest-dated, highest-consequence catalyst | | 2028 | $18B revenue model | Re-confirmed on this call; $20B/50% GM deferred |

Where consensus is most likely wrong

Applying the "you are not contrarian, only management is" test:

1. 2027 datacenter. Management has said "well above $2 billion" and has doubled the number twice in three months, with 100% backlog coverage for 2026 and named engagements for 2027 — and Chery said the raise is demand-led, not supply-led ("it is really driven first by demand"). El-Ouazzane's three-vector framing is specific and checkable. Jefferies models roughly 7pts of 2027 growth from it. This is the mispriced catalyst.
2. P&D / SiC inflecting off a $(99)m quarterly operating loss with double-digit CY2026 SiC growth already in backlog. Nobody models the loss going away.
3. LEO at ~$1B in 2026 heading to well above $3B cumulative through 2028, GM-accretive, and absent from every published model.

The honest counter. The GM bridge is now the whole debate and management hardened it, not softened it. With only 70bps of unloading recovery left, Q4'26 GM improves off 37.0% at a visibly flatter slope. Q2 EBITDA of ~$679m also missed the ~$798m consensus by roughly 15% even as revenue and EPS beat — the quality-of-beat question is legitimate and it sits directly on top of the raised OpEx guide. Sentiment is going worse → better on demand and better → worse on near-term margin.

| Item | Detail | |---|---| | Management on the line | Jean-Marc Chery (CEO); Lorenzo Grandi (CFO); Marco Cassis (President APMS + Strategy); Rémi El-Ouazzane (President MDRF); Jérôme Ramel (IR) | | Analysts who got the mic | 8 — Menon, Buchalter, Bluestone, Deshpande, Ghilotti, Scemama, Houri, Bouvignies | | Question exchanges | 16 (8 primaries + 8 follow-ups; IR enforced a one-question-plus-brief-follow-up limit) | | Well answered | 11 of 16 (69%) | | Deflected / avoided | 5 of 16 (31%) | | Notable | Unusually short Q&A for a $3.5B-revenue quarter — IR closed the call after the 8th analyst with no queue drain. Four presidents on the line and questions were routed down to Cassis / El-Ouazzane on the growth topics, which produced better granularity than STM's recent calls. |

Where the deflections cluster: every one of the five is about magnitude or concentration, not direction. Management was open about which way things move (revenue up, GM up, datacenter up, LEO up) and closed about how much and how concentrated. That is the single most useful read on this Q&A.

The five deflections, ranked by how much the missing answer matters

The best answers — what the Q&A actually gave you

• Chery's quarter-by-quarter Y/Y map by end market (see the end-market scoreboard above) — far more than STM normally gives, and it lets the above-$4B Q4 be reverse-engineered rather than trusted.
• 100% 2026 datacenter backlog coverage — the highest-quality de-risking datapoint in the entire Q&A, given in answer to whether the raise was demand- or supply-driven.
• A candid execution admission. Asked why Q3 capacity cannot be filled now if fabs are still underloaded, Chery said the residual underloading is "pure legacy analog", Q1 was "totally underloaded" so the Q2 ramp-back "was challenging", ST "delay[ed] a little bit this ramp-up", and that delay "has impacted also our capability on Q3 to fully fulfill the demand." Management is leaving Q3 legacy-analog demand on the table because of its own ramp execution, while carrying the unused-capacity charge for it. Both a demerit and a coiled spring for Q4/Q1.
• The 40% GM walk-back, delivered honestly. Ghilotti named the change; Grandi reframed the model rather than defending an unreachable milestone. Worth more than a dodge — but recognise what it is: the milestone effectively moved to 2028.
• Backlog and booking disclosures — above 50% of Q2 bookings were for 2027, and total backlog is now 4.5–5.0 quarters of Q2 average revenue. Caveat for the model: a backlog that large for a broad-line analog/MCU house includes long-dated AI and LEO program commitments and should not be read as near-term shippable demand.
• SiC quantified and forward-guided by Cassis, grounded in booked backlog rather than pipeline. SiC inflecting back to Y/Y growth is a genuinely new fact from this call.

What the Street did not ask

1. Power & Discrete is bleeding and nobody asked. The operating loss widened to $(99)m from $(84)m in Q1'26 and $(56)m a year ago, on $464m of revenue — a (21.4)% operating margin, disclosed in the prepared remarks. Sixteen questions and not one on the segment destroying the most operating income.
2. The capex raise. Net capex moved to the high end of $2.0–2.2B. Bouvignies circled the structural question (greenfield/brownfield) and got no answer at all; nobody followed up on the FCF consequence.
3. OpEx creep. FY26 non-GAAP net OpEx to "slightly above $3.8B", up low double digit Y/Y. Nobody asked what the incremental spend buys or whether it is permanent.
4. The $1.5B convertible and the early call of the 2027 notes. Zero questions on capital structure or dilution despite a net financial position of only +$2.01B against $4.02B of total debt.
5. Tariffs. The guidance explicitly excludes further tariff changes and no analyst tested the carve-out.

Management credibility check. Grandi was consistently straight, including on the 40% GM walk-back. Chery was the looser speaker — two wrong Industrial figures, a 1,000bps-ambiguous CE&CP guide, and the explicit refusal on the $20B leg. El-Ouazzane and Cassis both delivered quantified, backlog-grounded answers where asked. Analyst engagement was thin: eight analysts, 16 questions, call closed early, and the worst-performing segment drew no questions at all. The buy side is currently underwriting the AI/optical narrative and not policing the legacy franchise.

Documents examined: the Q2 2026 transcript against the Q1 2026, Q4 2025, Q3 2025, Q2 2025, Q1 2025, Q4 2024 and Q3 2024 calls; the SEC Form 6-K filed 2026-07-23. Reported figures cross-checked against Daloopa.

Two are materially thesis-relevant on their own: the 40% gross-margin condition set was rewritten while management asserted it had "always" been stated that way, and the $1B datacenter milestone moved forward four years in six months. A third — a "nonrecurring" cost now in its fifth consecutive quarter that is not excluded from headline non-GAAP gross margin — is an accounting-presentation issue, not a wording slip.

The pattern that matters more than any single card

Four of the ten confirmed contradictions are the same phenomenon: the CEO's unscripted Q&A number is more favourable than the press release or the CFO's number, every time.

| Metric | Press release / CFO | Chery in Q&A | Direction of error | |---|---|---|---| | Q2 Industrial Y/Y | 34% | 32%, then 30% | Lowers the base, inflating apparent Q4 acceleration | | Q2 CECP Y/Y | 50% | "close to 60%", "60% growth in Q2" | Overstates the H2 trajectory by ~10pts | | Automotive vs expectations | +16% Y/Y, guided in line with market | "above what we expect and what the market is expecting" | Upgrades in-line to a beat | | Distribution inventory | "below our standard target" | "well below our standard" | Escalates the restocking runway |

Zero instances run the other way. Individually each is a slip; together they are a bias.

Second pattern: retroactive reframing. Twice in one call management asserted that current guidance had been given previously when the transcript record says otherwise (cards 1 and 13). Per investing-principles, conflicting statements across transcripts is a named marker of a weaker management team — and it sits awkwardly beside an otherwise strong operational quarter.

Third pattern: new negatives emerged only under questioning. The Q2 legacy-analog ramp delay and its Q3 spillover, the China fab start-up as a Q4 gross-margin headwind, and the deferral of the 40% milestone past 2027 were all absent from the prepared remarks and the press release. Each came out in response to an analyst.

Consistency checks that passed

Recorded so the above is not read as a blanket indictment of disclosure quality. Q2 revenue versus its own guidance: guided 34.8% GAAP gross margin and ~$3.45B revenue; delivered $3,487m and 34.8% — revenue above the midpoint, gross margin exactly at it, and the Q1 guide's "about 100 basis points of unused capacity charges" for Q2 landed at 106bps. LEO cumulative ambition: "well above $3 billion cumulative over '26 to '28" stated identically in Q1'26 and Q2'26, no drift. Reshaping completion: end-2027 in Q2'26 is consistent with Q1'26's "benefits of Agrate at full speed will be more in the end of '27" and the Q3'24 framing of cost savings "exiting 2027." The $18B 2028 model narrows the prior '27-'28 range without contradicting it. Distribution inventory trajectory: promise made and kept. OpEx arithmetic: Q1 $885m + Q2 $960m + Q3 guide $980m implies ~$975m in Q4 to reach "slightly above $3.8 billion" — coherent.

Follow-up questions for management

1. On what call or in what filing were the two conditions for 40% gross margin at $4B both stated before 2026-07-23?
2. Is the Q3 37.0% gross-margin guide U.S. GAAP or non-GAAP, and why was the non-GAAP guide discontinued this quarter?
3. Why is the reshaping cost in COGS described as nonrecurring while remaining inside non-GAAP gross margin, when the $58m of restructuring charges is added back?
4. Was Q2 Industrial growth 34%, 32% or 30% Y/Y, and off which base is the "close to 40% in Q4" framed?
5. Net of price increases and input-cost inflation, is pricing a tailwind, a headwind or neutral to H2 gross margin?
6. Does the "well above $2 billion" 2027 datacenter figure assume current assembly-and-test capacity, or expansion still to be committed?
7. How does a book-to-bill near 2.0 with five quarters of backlog coexist with unused-capacity charges that do not improve in Q4?

The headline macro read-through is a removal, not an addition. In Q1'26 Chery concluded with "in the first quarter, despite the macroeconomic uncertainty, we saw improving demand." On the Q2'26 call that hedge disappears entirely — the concluding language is "in Q2, we saw demand further accelerating, strong bookings in all end markets and improved visibility." No reference to macroeconomic uncertainty appears anywhere in the Q2 call. A broad-line analog/MCU supplier with ~39% automotive and ~30-35% industrial exposure has stopped caveating the macro. That is a cycle-level signal for the whole analog/MCU/power complex, not an STM-specific one.

Macro commentary by axis

| Macro axis | What was said (Q2 2026) | Direction vs Q1 2026 | |---|---|---| | Interest rates / monetary policy | NOT MENTIONED. Zero references to rates, central banks, cost of capital or financing conditions — by management or by any of the eight analysts. | absent both quarters | | Inflation / input costs | "we see, let's say, in our input cost price increase… different materials or maybe, contractor activity that are increasing prices." (Grandi) | NEW — inflation re-emerging | | Pricing (output) | "in this context of higher input cost, yes, we're increasing the prices on selected products… an ongoing process that is expanding… at the end, the 2 impacts are more or less offsetting each other." (Grandi) | Improving — but GM-neutral | | Consumer sentiment | "the market of a smartphone that is decreasing because on the low end… there is an impact because of the memory price." (Chery) | Deteriorating at the low end | | Industrial sentiment | "the solid recovery of the overall industrial market… distribution inventory now well below our standard and the POS dynamic is very strong." (Chery) | Strongly improving | | Semi supply/demand | "book-to-bill close to 2 overall… significantly above 2 in Communication Equipment & Computer Peripherals… signs of tight supply in several product categories." (Chery) | Sharply tightening | | Order visibility | "out of the 100% of the booking we received in Q2, well above 50% were for next year… total backlog… 4.5 to 5 quarters of Q2 average revenue." (Chery) | Materially lengthening | | Automotive (industry-level) | "for semiconductor industry, the Automotive is expecting to grow about 13%, 14% year-over-year. This is what we will achieve." (Chery) | New industry data point | | Trade / tariffs | "This business outlook does not include any impact for potential further change to global trade tariffs." | Unchanged — identical language both quarters | | FX (EUR/USD) | Effective 1.16 in Q2; Q3 guide assumes 1.14. "we had also benefit from the FX… in Q4, this effect will not be there. It will be neutral." | Tailwind Q2→Q3, neutral in Q4 |

The analog pricing inflection — three consecutive quarters of improvement

| Call | Pricing expectation stated | |---|---| | Q4 2025 | "pricing decline on low to mid-single-digit expectation" | | Q1 2026 | "a very low single-digit price decline… a better situation in respect to what was a few months ago" | | Q2 2026 | "we're increasing the prices on selected products… an ongoing process that is expanding in terms of price increase" |

Trajectory: mid-single-digit decline → low-single-digit decline → selective increases, expanding. Positive read-through for every broad-line analog/MCU/power peer, since analog pricing moves as an industry. Caveat: the increases are offset by input-cost inflation — pricing is a revenue/mix positive but gross-margin neutral. Do not model pricing as a margin driver.

Companies and entities named

| Company / entity | Relationship | What was said | Read-through | |---|---|---|---| | NVIDIA | Ecosystem partner (robotics + datacenter power) | ST named a partner in NVIDIA Halos for Robotics — end-to-end functional safety for industrial and humanoid robots — contributing MCUs, sensors, motor control and security across chips, kits, software and reference designs | Better for both. Third consecutive quarter of escalating engagement (Q1: robotics ecosystem + 800V DC datacenter power; Q2: named Halos partner). Validates NVIDIA's physical-AI platform pulling in silicon partners. Worse for competing MCU/sensor vendors not on the Halos partner list | | NXP Semiconductors | Seller of the acquired MEMS business; automotive competitor | Integration "progressed well"; complementary portfolio "strengthening our Automotive Sensors business with awards at key players for active safety and tire pressure monitoring" | Better for STM, worse for NXP — STM is converting the carve-out into design wins at key players within ~5 months of close, i.e. NXP sold an asset now winning sockets against it. Cost: $24m PPA drag and ~$50m of added FY26 OpEx. Read-through for other auto-sensor incumbents: a newly-scaled competitor | | Texas Instruments | Closest broad-line analog competitor | ANALYST "you are seeing some tightness, as you say in the release, also TI suggested as well some tightness" (Bouvignies, UBS) | Better for the whole analog group. Two independent broad-line suppliers describing capacity tightness in the same window is an industry-level supply signal. Supports pricing power at TI, ADI, Microchip, onsemi, Infineon, Renesas. Analyst characterisation of TI, not an STM statement — verify against TI's own disclosure | | Hyperscalers (collectively) | Datacenter customers | "we see an evolution of our revenue that is pretty consistent with market share distribution between hyperscalers, which makes us actually quite confident about the composition of our revenue" (El-Ouazzane) | Presented as a de-risking disclosure: revenue spread across hyperscalers roughly in proportion to their shares, i.e. not single-customer concentrated. Read-through: silicon-photonics/optical content becoming a broad hyperscaler standard rather than one buyer's bespoke design. But see Deflection 1 — no number was given | | Amazon Web Services | Named anchor datacenter customer at Q1 | Q1 "we expanded our strategic engagement with AWS through a multiyear multibillion U.S. dollar commercial engagement"; "AWS will be a fantastic driver… But our ambition is well above" | Better for STM, but note the Q2 silence. AWS was named repeatedly at Q1 and is not mentioned once on the Q2 call, replaced by the generic hyperscaler framing. Two readings: (a) benign — the opportunity genuinely broadened, exactly as promised; (b) no progress update on a multibillion-dollar engagement announced one quarter earlier. The doubling of the ambition argues for (a) | | Hua Hong Semiconductor | China-for-China foundry partner | Q1 first batch of STM32 wafers fully produced in China by Huahong delivered; Q2 "we have qualified 14-nanometer technology in China with our main partner" | Better for Hua Hong and for STM's China access. Localised MCU supply is now a live competitive weapon in China industrial and optical. The Q2 quote says "our main partner" without naming — the linkage is inferred from Q1. The 14nm claim sits oddly with Hua Hong's mature-node profile; verify before republishing | | Quobly | Investee (silicon quantum) | ST joined the EUR 115m Series A in June to industrialise silicon-based quantum computers on ST's 300mm FD-SOI line, first commercial product by end-2026 | Better for Quobly; strategically optional for STM. Financially immaterial today. Incrementally worse for superconducting-qubit approaches if CMOS-compatible spin qubits industrialise on standard 300mm tooling. Option value, not a modelled line | | UNNAMED LEO operators — "main customer", "second largest", "other constellation" | Customers and their competitors | "very much a launcher dependent business… Other constellation have faced a bit of a snag lately, and to be fixed in the coming quarters. But directionally, nothing has changed. We see actually a strong '26 and an even stronger '27" (El-Ouazzane) | Better for STM's lead LEO customer; worse for at least one rival constellation. The most explicit competitor-of-a-customer read-through on the call. Also flags launch cadence as the gating variable for the whole LEO silicon supply chain. Identities deliberately not asserted | | UNNAMED SiC "main customer" + Chinese EV carmakers | Customer / end-market driver | ANALYST framing of Chinese EV 800V adoption plus strong deliveries at a main customer — confirmed by Cassis with numbers: SiC +low-teens Y/Y, +mid-30s Q/Q, book-to-bill well above 1, FY26 double-digit growth off booked design wins | Better for the SiC complex after a brutal downcycle, and better for Chinese 800V EV platforms and their suppliers. But P&D still lost $(99)m for a fifth straight quarter — SiC demand is recovering faster than SiC profitability | | UNNAMED non-traditional AI-server entrants from solar and battery storage | New customers | "a growing number of nontraditional AI server companies, including players coming from industries such as solar power and battery storage, where we are leveraging strong relationships to support their expansion" (Chery) | The most under-appreciated read-through on the call. Solar-inverter and BESS companies entering AI-server power already buy STM high-voltage power/analog, so STM inherits the relationship for free. Worse for traditional datacenter power incumbents facing entrants with existing 800V conversion competence. Widens the AI power TAM beyond the known hyperscaler/ODM set | | OSATs (assembly & test) | Suppliers | "we have some pocket of capacity limitation time to time with OSAT, but okay, we manage it" (Chery) | Better for OSATs, mild negative for STM. Back-end capacity is a live constraint. Note the datacenter raise was explicitly enabled by back-end capability — assembly/test, not wafers, is the swing factor | | Memory manufacturers (implied) | Suppliers to STM's customers | "the market of a smartphone that is decreasing because on the low end… there is an impact because of the memory price" (Chery) | A cross-industry margin transfer from logic/analog to memory. Rising DRAM/NAND pricing is squeezing entry-tier smartphone BOMs and destroying unit volume — the sole stated cause of STM's PE going Y/Y negative in H2. Better for Micron / Samsung / SK Hynix pricing; worse for every analog, sensor and PMIC supplier levered to entry-tier handsets | | Materials suppliers and manufacturing contractors (implied) | Suppliers | "there are different materials or maybe, contractor activity that are increasing prices" (Grandi) | Better for semiconductor materials and subcontract/equipment-services suppliers; cost headwind for IDMs. The supply-side counterpart to STM's own selective price increases | | Omdia | Third-party researcher | Q1 ST ranked #1 vendor worldwide for general-purpose microcontrollers for the fifth consecutive year | Independent confirmation of category leadership — directly relevant to the "leaders remain leaders" test, and GP MCU is precisely the line now capacity-tight and driving both the industrial recovery and the optical-transceiver control plane | | Crolles (own capacity, competitive context) | Own asset | "Crolles will reach 15,000 wafer per week and will go above… we have a lot of scalability… we are not right now gated by capacity expansion" | Worse for competing silicon-photonics suppliers. STM claims it is not capacity-gated in silicon photonics while the market accelerates to 800G/1.6T — an unusual position in a tight cycle. Pure-play or constrained photonics competitors face a supplier that can take share without a capex race |

The strongest cross-company signal: 800G / 1.6T optical adoption

"we see clearly an acceleration in the adoption of 800 gig and 1.6 terabit per second pluggable optics… a triple effect… a fairly large market share when it comes to the microcontroller taking care of the control plane, a growing share when it comes to the electronic IC driven by our BiCMOS technology. And starting from next year, but really accelerating next year, is actually our growing revenue in silicon photonics." — El-Ouazzane

Confirmed by the hardest number on the call — the datacenter ambition doubled at both horizons in a single quarter, and it is contracted rather than hoped for: "Backlog for this year, 100% coverage. And next year, engagement are covering our expectation." Better for the entire AI optical-interconnect chain: transceiver module makers, silicon-photonics suppliers, and the MCU/driver-IC content layer inside pluggables. Worse for suppliers whose datacenter content is confined to copper/electrical interconnect, and MCU vendors without an established transceiver control-plane socket.

Macro axes conspicuously not discussed — absence is information

• Interest rates / central-bank policy — never raised, by management or by any analyst.
• Consumer credit, employment, general consumer confidence — not raised. The only consumer commentary is supply-side (memory prices raising low-end smartphone BOM cost), not demand-side.
• European industrial policy / Chips Act funding — not raised, despite ~27.5% French plus Italian government ownership. Capital-grant advances appear only in a balance-sheet footnote (adjusted net financial position +$1.70B versus reported +$2.01B).
• China macro — mentioned only as a capacity strategy ("China for China") and a fab start-up cost, never as a demand variable.

The thesis. STM passes the trajectory test decisively — 26% growth and accelerating, off a five-quarter base of monotonic improvement, with margin Y/Y confirming the revenue signal exactly one quarter later and a book-to-bill near 2.0 saying it continues. It passes the leadership test (#1 in general-purpose MCUs for five straight years, in the product line that is now both capacity-tight and the control plane for 800G/1.6T optics). And it is a textbook sentiment-inversion setup on the revenue side: a management team with a 5-for-5 guidance record has doubled the same forward ambition twice in three months, with 100% backlog coverage and a named, checkable mechanism, while the sell-side response was that ST had "not done enough" on gross margin.

The discipline. Revenue credibility is rising; margin credibility is not. A 2027 revenue re-rating should not be accompanied by a 2027 margin re-rating. The 40% gross-margin milestone did not slip quietly — it was rewritten, and asserted as having always been that way, which is exactly the marker investing-principles flags for a weaker management team. Underwrite the revenue catalysts, which are backlog-verified; treat the margin catalyst as an end-2027 event; and do not underwrite EPS surprise on this name, because tax and restructuring — not operations — drive it.