Earnings Highlights: Tower Semiconductor (TSEM) Q3 FY2026 — One Quarter of Silicon Photonics Revenue Tops All of Last Year, Start Capacity Reaches a $2.8B Run Rate by Q3 2027, Contracts Reach 2029
1. Three Signals to Remember First
First, growth is highly concentrated in silicon photonics. Q2 SiPho ran at a $680M annualized rate, or about $170M a quarter. Q3 came in at about $240M, up roughly 41% sequentially and adding about $70M. Total company revenue rose only $66M over the same period, so everything else combined was roughly flat (about $290M to about $286M). One engine is pulling the whole vehicle, which is both why Tower has pricing power and why competition gets so much scrutiny.
Second, profit is climbing faster than revenue, but the slope should not be extrapolated. Each additional $1 of revenue dropped $0.62 to gross profit this quarter. Part of that ratio comes from delayed depreciation, and the CFO himself pulled expectations back to 55–58%. Next quarter's gross margin is the real test.
Third, the market's response was a decline. The stock fell about 6% on the day of the call (Yahoo Finance quote). Three analysts raised competition or share in the Q&A, and management conceded "there will be competition." The market's question is no longer whether demand is sufficient. It is how long this share holds.
2. Revenue and Financials
All figures are GAAP in US dollars, taken from the 6-K earnings release and cross-checked against the call transcript.
Metric | Q3 2026 | Q2 2026 | Q3 2025 | YoY |
Revenue | $526.2M | $460.1M | $395.7M | +33.0% |
Gross profit | $178.8M | $137.8M | $93.0M | +92.1% |
Gross margin | 34.0% | 29.9% | 23.5% | +10.5 pp |
Operating profit | $127.1M | $90.3M | $50.6M | +151.4% |
Operating margin | 24.2% | 19.6% | 12.8% | +11.4 pp |
Net profit (attributable) | $120.3M | $90.8M | $53.6M | +124.2% |
Diluted EPS | $1.05 | $0.79 | $0.47 | +123.4% |
▲ Tower Semiconductor FY2026 Q3 core financials: revenue $526.2M, gross margin 34.0%, operating margin 24.2%, net income $120.3M, diluted EPS $1.05, approximate segment revenue, Q4 guidance $600M (Image source: Simple Tech Trend original; data from SEC 6-K earnings release and earnings slides)
The revenue mix has flipped. Converted from the revenue shares in the earnings slides (approximate):
Segment | Share Q3 2026 | Share Q3 2025 | Implied revenue |
RF Infrastructure (incl. SiPho, SiGe) | 56% | 27% | ~$295M |
Power | 14% | 17% | ~$74M |
RF Mobile | 10% | 26% | ~$53M |
Sensors & Displays | 9% | 14% | ~$47M |
Discrete | 7% | 10% | ~$37M |
MS/CMOS/Other | 4% | 6% | ~$21M |
Subtract $240M of SiPho from $295M and silicon germanium (SiGe) plus other items in RF Infrastructure come to about $55M, roughly the same as Q2. Management calls SiGe demand "unprecedented," but it has yet to show up in revenue. That capacity is waiting on new tools.
Balance sheet and cash flow. Cash and short-term deposits total $1.47B against $154M of debt, for net cash of about $1.32B. Customer advances (current plus long-term, with the current line including deferred revenue) stand at $341M, versus just $27.5M at the end of last year. Q3 operating cash flow was $185M and capex $191M, for free cash flow of about -$6M, or about -$28M excluding the $23M of new advances this quarter. "All expansion self-funded" holds for now, but the company is already at break-even on cash, with 35% of the $920M still to pay between Q4 and H1 2027 and Track-1 tool payments scheduled for 2027 and 2028.
3. Capacity Breakdown: How the $2.8B Was Derived
This table drew the most questions on the call. One analyst even got it wrong live (reading it as one more multiple on top, or $960M a quarter), and the CFO corrected him personally: 3x means total capacity at three times Q3 shipments.
Timing | Relative to Q3 shipments | Implied quarterly revenue | Annualized |
Q3 2026 (actual) | 1x | ~$240M | ~$0.96B |
Q4 2026 (company expected shipments) | — | above $275M | above $1.1B |
End of 2026 (start capacity) | 2x | ships in Q2 2027, ~$480M | ~$1.9B |
Q3 2027 (start capacity, incl. Track-1) | 3x | ~$720M | "a bit above $2.8B" per company |
Track-1 in the table is the first phase of the Japan 300mm expansion (repurposing the Arai facility and maximizing Fab 7 output, supported by Japan's METI).
Four things to keep in mind when reading this table:
It is wafer starts, not shipments. The CEO's example: wafers started from December through February ship in Q2. On that rhythm, 3x would reach shipments around late 2027 to early 2028. That is our estimate. The company gave no such date.
It is a multiple of raw capacity. The CEO declined to give a 2028 revenue target and offered only multiples of Q3 revenue, which assumes constant mix and pricing. A rising NPO share would push pricing up, while utilization below full would pull it down.
It is not the full picture. Integrated lasers are excluded, and the CEO said another expansion tranche, mainly at two 200mm sites, is likely to be announced shortly. Neither the capex nor the capacity has been disclosed.
Track-2 is on another scale. Fab 7 runs at about 10,000 300mm wafers a month today, and Track-2 takes it to about 40,000. The CEO added that a 300mm wafer has 2.25x the area of a 200mm wafer, so if it were all SiPho, that approaches 100,000 200mm wafers. Building starts in 2028, with the contribution mainly in 2029 and possibly a quarter earlier at best.
Does the $3.6B model add up? An analyst asked directly whether $3.6B is too conservative. The CEO replied "no comment," then added that it is an aggressive target. We tried a rough reconciliation: $2.88B times the model's 85% utilization is about $2.45B, and adding the current annualized non-SiPho revenue of about $1.14B lands at about $3.6B. The company has never broken it down this way, and this is just one calculation that happens to fit. What it suggests: $3.6B roughly equals capacity running at model assumptions with no growth elsewhere, while integrated lasers and the new 200mm tranche would be additional.
How to read the one-year payback. The company says the $920M investment achieves "a full ROI within one year" but does not define how. Going from 1x to 2x adds about $960M of revenue a year, which matches numerically. Using the CFO's 55–58% incremental gross margin, that is about $530–560M of gross profit a year, or a payback of about 1.7 years. SiPho margins may run above the company average, and the $920M also supports SiGe, so the real figure should be faster. Still, when you read "one-year ROI," the basis of the calculation remains unclear.
4. Technology and Business Highlights
NPO is the main lever for expanding the served market. Tower is prototyping near-packaged optics (NPO) with more than 10 top customers, all of them existing pluggable customers. NPO helps Tower in three ways: the SiPho die can integrate both transmit and receive (pluggables usually carry only transmit), more lanes mean a larger die (current tape-outs are mostly 16 and 32 lanes, largely at 300mm), and a new platform starts at a higher price. Vijay Rakesh asked whether 3.2T and 400G per lane carry a higher mix and a higher dollar value, and the CEO answered "yes" to both, without elaboration. We analyzed the buyer structure of NPO in OCP Global Summit 1/6: NPO Wins Purchasing Power, Not Shipments, and the fact that all of these customers are pluggable incumbents fits that thread.
SiGe grows alongside NPO. Most NPO implementations remove the DSP in favor of SiGe TIAs and drivers, and moving equalization into those parts also increases their footprint. Tower announced a substantial capacity agreement with Semtech this quarter. SiGe is now served from three 200mm fabs and one 300mm fab, and the next-generation H6 process is already in production. For the demand backdrop, see Earnings Highlights: Semtech (SMTC) Q2 FY2027.
Heterogeneous integration pushes Tower toward the laser market. Tower announced initial production for NewPhotonics: indium phosphide (InP) epi chiplets are attached die-to-wafer onto SiPho wafers, then processed in the fab into lasers. The same approach is being applied to thin-film lithium niobate (TFLN) and InP modulators for next-generation 400G per lane products. On how much of the market integrated lasers will take, the CEO said plainly that he has no clear answer and that analysts are split, while stressing that customer forecasts are strong and ASPs appreciably higher. On how integrated lasers change light-source purchasing, see CPO Lasers Don't Have to Chase 400 mW: NPO Integrated Lasers Bring the Bar Back to Taiwan's Home Turf.
Other new fronts. SiPho solid-state optical circuit switches (OCS) challenge MEMS with nanosecond switching and CMOS manufacturing scale, and management calls Tower the go-to solution. That market remains highly concentrated, though, as covered in ECOC 2026: 95% of OCS Shipments Are One Company's Business. Coherent SiPho with Marvell is being evaluated for scale-across inside the data center. Further out are wide-and-slow approaches using micro LEDs or VCSELs, with tier-one engagements under way.
The non-optical businesses are busy too. BCD grew about 10% year over year, several strategic accounts signed long-term agreements, and blended 2027 pricing is forecast 20% above 2026, a rare sign of strong pricing outside optics. In RF-SOI, Agrate hit a quarterly record and Fab 10 set a tape-out record. RF Mobile's share fell, which the CEO attributed to moving Fab 7 toward optics and requalifying customers at Fab 10. He expects growth to resume in the first half of 2027, with Q2 a record.
Utilization keeps rising. 200mm fabs ran at 85–90% (Q2 was 80–85%, with Japan's Fab 5 at only 75%), and 300mm is full.
5. Management Outlook
Q4 guidance: revenue of $600M (±5%, a $570–630M range), up about 36% YoY and about 14% QoQ. Full-year revenue would be about $2.0B (2025 was $1.57B), with a $2.4B annualized run rate.
No Q1 seasonality. Management targets a Q1 2027 above the $2.4B annualized level, more than 45% above the Q1 2026 starting point, with sequential growth expected from there.
The Q4 SiPho target moved from "crossing $1B" to "exceeding $1.1B." In our Tower 2026 Q2 earnings highlights the figure was $1B. If Q4 SiPho is about $275M, everything else is about $325M, up about $39M from Q3. That suggests Q4 growth no longer relies on SiPho alone and that new SiGe capacity may start contributing.
The next step for gross margin. If Q4's incremental $74M of revenue drops through at 55–58%, Q4 gross margin would be about 37%. That is our estimate, not company guidance.
The long-term model stays at the version raised in August: revenue $3.6B, gross margin 45%, operating margin 38%, net profit $1.2B (33%), assuming 85% utilization across all fabs. This quarter's annualized revenue is about 58% of the model, with gross margin still 11 points short.
The language has distinct levels of certainty:
"Confirmed": Q3 numbers, Q4 guidance, the $920M program ahead of schedule, all Track-1 tools ordered. The company says Track-1 is ahead of schedule, with full production starts targeted for Q3 2027 (the Q2 call gave Q4 2027 for volume production).
"Target": Q4 SiPho at $1.1B, no Q1 seasonality, the $3.6B model. The press release uses the word target for these.
"Conditional": NPO. The CEO expects an "appreciable amount" of shipments in H2 2027, then immediately adds that it depends on market adoption. The president added the most useful line: if it is not NPO, it is pluggable, and both are Tower wafers.
Note this one: capacity being "spoken for" rests on "signed agreements and contracts in advanced stages of negotiation," so not everything is signed.
6. Supply Chain and Customer Clues
The competitive playbook got named. Asked about pricing, the CEO said new entrants often come in by deep discounting "because they don't compete technically," and that long-term agreements include pricing targets that set a starting bar. He also left an opening: if the market truly demands lower prices, Tower would be reasonable. In Q2 we flagged wafer pricing as the model's most sensitive assumption. This answer does not remove that risk, and he would not give a number for NPO starting prices.
No market share figure. Management claims a majority share of the pluggable scale-out market, but the CEO called any share forecast beyond 2027 prognostication. The argument is that customers are signing through 2029 because of Gen+2 and Gen+3 co-development, and that today's shipping products will not be a big portion of 2029 volume.
Risk disclosure is blunter than the call. The press release's forward-looking statements list several items worth reading: over-demand could cause bottlenecks and delivery delays, leading to compensation or prepayment repayments; the GlobalFoundries patent infringement claims; the Fab 3 landlord's alleged material breach claim; the dispute over the Intel capacity corridor; and the 15% global minimum tax. The first item and Fab 3 (Newport Beach, one of the SiPho sites) are the most directly relevant.
300mm is a bet on yield leadership. The CEO stressed that 300mm is shipping at high yields with clearly leading insertion loss, and called its DEMUX capability the best in the world. 300mm SiPho then triples from its Q1 2027 level by Q3. That ramp is the execution risk most worth watching.
Customer signals elsewhere. OLED-on-silicon display customers expect a steep ramp in 2027, and production has started on a high-speed global shutter sensor for a leading camera maker.
Summary
Tower made one thing clear this quarter: the SiPho question has shifted from "is there demand" to "when does capacity arrive, and can pricing hold." That is a good problem for a foundry, and also why the market has become picky.
Three things to track next:
Gross margin in Q4 and Q1 2027. Once new tools begin depreciating, can incremental gross margin hold at 55–58% or better.
Whether Q2 2027 SiPho revenue approaches $480M. This is the first time 2x start capacity fully converts into shipments, and the first real checkpoint on the $2.8B line.
The first NPO customer press releases and the new 200mm tranche. The company says these are coming, and they will determine whether $3.6B is a ceiling or a floor.
This article is for technology and industry trend analysis only and does not constitute investment advice.
Comments