Tower’s Q3 beat is not the story; the SiPho capacity call is

The clean separation is important because the quarter itself did not force an estimate reset on the usual beat-versus-consensus math. What was priced in was essentially $395.0 million of revenue and $0.54 of EPS; what Tower delivered on the street-comparison basis was $395.7 million of revenue and $0.55 of EPS, surprises of +0.2% and +1.3%. A portfolio manager buying only that beat is buying the wrong thing. The investable surprise sits in the bridge from Q3 FY2025 revenue of $395.7 million and gross margin of 23.5% to Q4 FY2025 revenue of $440.2 million and gross margin of 26.7%, and in the explicit silicon photonics roadmap behind that step. The market may still be treating Tower as a specialty foundry with a low-20s gross-margin ceiling after Q1 FY2025 gross margin of 20.4% and Q2 FY2025 gross margin of 21.5%; the print argues that ceiling is being reset by capacity tied to named end markets rather than by generic utilization.

The financial trajectory supports that interpretation because the Q3 result is the third sequential revenue increase of FY2025 and the gross-margin recovery is now visible in reported history, not just guidance language. Revenue moved from $358.2 million in Q1 FY2025 to $372.1 million in Q2 FY2025 to $395.7 million in Q3 FY2025, with revenue QoQ of +3.9% and +6.3% in the last two reported quarters. Gross margin followed from 20.4% to 21.5% to 23.5%, while diluted EPS moved from $0.35 to $0.41 to $0.47. That matters because Tower’s FY2024 revenue recovery had not brought margin with it: Q4 FY2024 revenue was $387.2 million, but gross margin was only 22.4%, below Q3 FY2024 gross margin of 25.1%. Q3 FY2025 is the first print in this sequence where revenue, gross margin, and diluted EPS are all moving together from the FY2025 trough, and the Q4 FY2025 line in the quarterly history extends that pattern to $440.2 million of revenue, 26.7% gross margin, and $0.70 diluted EPS.

That margin trajectory explains why the Q4 guide carries more information than the Q3 beat. Russell Ellwanger did not merely point to a seasonal uptick; he framed Q4 as a commitment against a prior-year plan, saying Tower guides “our fourth quarter to be a revenue record of $440 million, plus/minus 5%, fulfilling our beginning of year target of quarter-over-quarter growth throughout the year with strong acceleration in the second half.” The wording matters because it ties the $440 million to a year-long sequence rather than a one-quarter pull-in. It is also consistent with the income-statement history: Q4 FY2025 revenue of $440.2 million is listed at +11.3% QoQ and +13.7% YoY, a sharper slope than Q3 FY2025 at +6.3% QoQ and +6.8% YoY. The variant perception is that the market may underwrite the +11.3% QoQ revenue step as merely operating leverage from utilization, while the more durable value is the business mix driving that utilization.

The mix evidence is unusually concrete for a specialty foundry call. Ellwanger said the RF infrastructure business increased its contribution from $67 million or 18% of corporate revenue in the third quarter of last year to $107 million or 27% for the third quarter of this year. That is not a vague AI-adjacent claim; it identifies a corporate mix shift inside a revenue base that was $370.5 million in Q3 FY2024 and $395.7 million in Q3 FY2025. Silicon photonics is the sharper lever: Tower’s silicon photonics business grew in Q3 to $52 million, “approximately 70% growth as compared to the third quarter of 2024,” and management targets 2025 silicon photonics revenue above $220 million, up from $105 million in 2024, with a Q4 ’25 annualized revenue run rate exceeding $320 million. This means the Q4 revenue record is not only a utilization story; it is an adoption story concentrated in a platform where capacity decisions and customer qualifications can create multi-quarter visibility.

That is why the incremental capacity announcement should be valued differently from ordinary foundry capex. Ellwanger said Tower has “begun an additional investment of $300 million for further substantial SiPho capacity expansion, and next-generation capabilities in Fab 3, Fab 9, Fab 2 and Fab 7,” targeted to achieve full volume in wafer starts in the second half of 2026. CFO Oren Shirazi put that in the broader capital framework: Tower had previously announced plans to invest $350 million to expand capacity in its 8-inch fabs in Israel and Texas and its 12-inch Uozu fab in Japan, 50% of that amount has been paid to date, and the remaining 50% is expected to be paid in the coming year. He then stated the total SiPho and SiGe capacity and capabilities plan would be $650 million. The distinction is that Tower is not asking investors to wait for unspecified optionality; it is matching $650 million of named SiPho and SiGe capacity and capability spend against a model targeting $2.7 billion in annual revenues at full loading of existing fab and qualified capacity, $560 million in annual operating profit, and $500 million in annual net profit.

The risk to the thesis is not that Tower missed the quarter, because it did not; the risk is whether investors should capitalize those targets before the mix proves itself through gross margin. The data conflict is straightforward. Reported Q3 FY2025 gross margin was 23.5%, still below Q3 FY2024 gross margin of 25.1% and below several peer gross-margin prints, yet Q4 FY2025 gross margin in the history is 26.7% and Q1 FY2026 gross margin is 26.8%. That is the key debate: if Q3 is the right run-rate margin, then the SiPho story is still early and the stock should not get much credit for the $2.7 billion long-term revenue model; if Q4 FY2025 and Q1 FY2026 are the right margin direction, then the Q3 print is an inflection quarter hidden by a +0.2% top-line surprise. I favor the second interpretation because revenue QoQ acceleration, gross-margin expansion, and EPS expansion are aligned in Q2 FY2025 and Q3 FY2025, and the next two quarters in the history show the margin line at 26.7% and 26.8%, not a fallback toward 20.4%.

The customer and supplier read-throughs follow directly from that mix shift, and the magnitudes are not trivial. For NXP, Tower’s role as a specialty foundry for analog/RF/power is levered to RF infrastructure rising from $67 million or 18% of Tower revenue to $107 million or 27% in one year, plus power targeting year-over-year growth of 15% with higher growth for advanced 300-millimeter platforms. For ON Semiconductor, the read-through is more selective: Tower’s sensor and displays technologies represented 14% of Q3 corporate revenue and are expected to show mid-teens full year-over-year growth, while power is also tied to data center power and handset envelope tracker volume expected to continue through the next multiyears. For STMicroelectronics, the TPSCo JV angle matters because Tower’s expansion plan explicitly includes the 12-inch Uozu fab in Japan, and the total SiPho and SiGe capacity and capabilities related CapEx plan is $650 million. For Soitec, the implication is cleaner still: Tower’s PH18 SiPho platform supplier exposure sits behind silicon photonics revenue targeted above $220 million in 2025, up from $105 million in 2024, and a Q4 ’25 annualized run rate exceeding $320 million.

The peer context reinforces the idea that Tower should not be compared only on revenue growth screens. In the latest reported quarter peer table, Tower shows revenue YoY of +15.5% and gross margin of 26.8%, while GFS shows revenue YoY of +3.1% and gross margin of 27.6%. UMC shows revenue YoY of +5.5% and gross margin of 29.2%, and 5347.TWO shows revenue YoY of +4.9% and gross margin of 29.3%. Tower is not yet the gross-margin leader in this peer cut, but its growth rate is materially above GFS, UMC, and 5347.TWO while its gross margin is already close enough to GFS that mix improvement matters more than blanket foundry cycle comparisons. That is the core portfolio point: if investors screen Tower as a lower-margin analog foundry, they miss that the reported revenue base is rotating toward RF infrastructure at $107 million, silicon photonics at $52 million in Q3, and power growth targeted at 15%.

The call delivery was consistent with that interpretation, although not uniformly clean. The relevant tone history shows Q3 FY2025 sentiment at 0.44, barely below Q2 FY2025 at 0.45, but guidance_tone remained elevated at 0.73 after Q2 FY2025’s 0.83. The notable shift is in the composition of tone: Q3 FY2025 prepared_sentiment was 0.70, while qa_sentiment was 0.21, and uncertainty dropped to 39.1 from Q2 FY2025’s 55.9. That combination says the prepared narrative was much more forceful than the Q&A, but the call was less uncertain than the prior quarter on the measured index. I would not overread the sentiment dip from 0.45 to 0.44; the more useful data point is that Q3 FY2025 qa_evasiveness was 11.2, far below Q4 FY2024’s 51.4 and Q1 FY2025’s 46.5, which fits a management team giving more direct answers around capacity, run rates, and long-term loading. The hedge is that Q4 FY2025 later shows uncertainty of 55.0 and qa_evasiveness of 29.9, so Q3’s unusually low uncertainty may have reflected the simplicity of the Q3 setup rather than a permanently cleaner disclosure cadence.

That delivery nuance matters because management made large claims, and the stock should be paid only for the portions with measurable milestones. The most important management language was Ellwanger’s silicon photonics bridge: Tower is targeting “over $220 million for the full year with a $320 million plus run rate in Q4, but bringing that to well above $900 million by target.” The phrase “well above $900 million” is not in the same category as reported Q3 silicon photonics revenue of $52 million or the Q4 ’25 annualized run rate exceeding $320 million; it is a longer-term aspiration tied to capacity and customer adoption. The right underwriting approach is therefore staged. Give credit now for the move from $105 million in 2024 to above $220 million in 2025 if Q4 confirms the run rate, give partial credit for the $650 million SiPho and SiGe capacity plan as payments and wafer starts progress, and reserve the “well above $900 million” target until full volume in wafer starts in the second half of 2026 becomes visible.

The handset and data-center power comments add a second vector, but they are less bankable than the SiPho numbers. Power is targeting year-over-year growth of 15%, with disproportionately higher growth for advanced 300-millimeter platforms, and management cited 16-volt operating voltage devices with more than 40% lower Rdson than prior technology, plus new elements to the 1.2, 3.3 volt 65-nanometer BCD flow. These numbers matter because they align Tower with power conversion efficiency rather than only legacy specialty foundry loadings. But unlike silicon photonics, the call does not give Q3 power revenue dollars or a Q4 annualized power run rate. For that reason, power should be treated as a margin and customer-stickiness enhancer, not the primary reason to re-rate the shares after this print. The same applies to sensors and displays: 14% of Q3 corporate revenue and mid-teens full year-over-year growth are useful, but the revenue acceleration into Q4 is more tightly anchored to RF infrastructure and SiPho.

The debate into the next quarter is therefore simple and testable. The thesis is confirmed if Q4 revenue lands around the guided $440 million, plus/minus 5%, if the history’s $440.2 million revenue and 26.7% gross margin profile is sustained by reported accounts, and if silicon photonics exits Q4 at an annualized revenue run rate exceeding $320 million while full-year 2025 silicon photonics revenue is above $220 million. It is strengthened further if RF infrastructure remains near the Q3 scale of $107 million or 27% of corporate revenue, and if power stays on track for 15% year-over-year growth. It breaks if Q4 revenue fails to show the promised acceleration from Q3 FY2025’s $395.7 million, if gross margin falls back toward the Q3 FY2025 level of 23.5% rather than progressing toward 26.7%, or if management backs away from the $650 million SiPho and SiGe capacity plan, the second half of 2026 full-volume wafer-start target, or the $2.7 billion annual revenue model at full loading. Until those numbers fail, the right read is that Q3’s near-inline beat hid a more important event: Tower is turning specialty foundry mix into a capacity-backed silicon photonics growth model.