Tower Semiconductor Ltd. (TSEM) | The Buildout — AI Infrastructure
The Verdict
Tower Semiconductor runs a specialty analog foundry. It does not chase leading-edge digital transistors; it sells differentiated specialty analog process technologies, and its most important one is silicon photonics. These are the photonic ICs, plus the SiGe analog silicon, that sit inside optical transceivers moving data between AI processors. Management describes high-bandwidth, low-latency, energy-efficient optical connects as a critical enabler of AI infrastructure. Tower sells at the optical interconnect layer — the photonic IC and the driver silicon inside someone else's transceiver — not the finished module.
| Market Cap | — |
| Revenue (TTM) | $1.7B |
| Revenue Growth | +14.9% |
| EBITDA Margin (TTM) | 34.8% |
| Net Cash | $1.3B |
| Earnings Beats | 7 of 7 |
| P/E (TTM) | — |
| EV/EBITDA (TTM) | — |
What We Like
- Q2 2026 revenue was a record $460M, up 24% year-over-year, with record gross profit of $138M at a 30% gross margin.
- Silicon photonics revenue grew more than 60% quarter-over-quarter and more than 270% year-over-year, reaching a >$680M annualized run rate.
- Customers have contracted roughly $1.3B of silicon photonics revenue for 2027, backed by $290M of prepayments received in Q1 2026.
- The 2028 model was raised to $3.6B revenue, 45% gross margin and $1.2B net profit, from $2.8B and $750M.
- The company beat analyst earnings estimates in 7 of 7 tracked quarters and met or beat its last two quarterly revenue guides.
What We’re Watching
- The $1.3B 2027 SiPho booking was not updated on the Q2 call even as the model rose; management distinguishes 'spoken for' from 'booked.'
- The 2028 model assumes all fabs at 85% utilization; Fab 5 is at 75% and the 200mm fabs at 80-85%.
- Near-package optics timing is contested: Tower guides to 'tens of the percentage' of 2027 shipments, while component supplier MACOM points to 2028.
- Japan Track 1 full production readiness is targeted for Q4 2027, with Track 2 installed and functioning by Q4 2028.
The operating thesis is strengthening: records across revenue, gross margin, net margin and earnings, margins expanding, and a 7-of-7 beat record. The raised 2028 model and the multi-year contract visibility point the same direction. The tension is that the contracted book did not move when the model rose, and the company is adding capacity into an industry that is adding capacity too. The open question is whether signed volume catches up to the raised plan, or whether demand intent is carrying it.
Earnings Beat
Q2 2026 revenue was a record $460M, up 11% sequentially from $414M and 24% year-over-year from $372M. Gross profit was a record $138M at a 30% gross margin, up 72% from $80M a year earlier. Operating profit was $90M, 2.26x the year-ago quarter, and net profit was a record $91M at a 20% net margin, up 95% year-over-year.
| Metric | Q2 FY2026 | Q1 FY2026 | Q2 FY2025 | YoY |
|---|---|---|---|---|
| Revenue | $460M | $414M | $372M | +23.6% |
| Gross margin | 29.9% | 26.8% | 21.5% | +840bps |
| EBITDA | $174M | $148M | $114M | +51.9% |
| EPS | $0.79 | $0.57 | $0.41 | +93.0% |
| SiPho annualized run rate | >$680M | n/a | n/a | >270% YoY |
The second quarter was quite significant, setting substantial company records across all key metrics … the profitability gains we delivered in the second quarter are not a one-time achievement, but rather just an initial step towards profitability expansion and cash generation.— Russell Ellwanger, CEO, 2026-08-04
Management tone: Management raised the long-term model rather than reaffirming it, and described Q2 profitability as an initial step rather than a peak. The competitive stance softened: in Q1 the CEO described Tower as having by far the leading silicon photonics market share; on the Q2 call, asked about peer capacity additions, he said supply is increasing and that he did not have a good feel for how much capacity is coming. Management conceded that near-package optics may not raise content per transceiver, confirmed the 300mm RFSOI decline, and said the indium phosphide constraint was addressed with the IQE contract.
Management Guidance
Management guided Q3 2026 revenue to $520M midrange, which it framed as an annualized revenue run rate above $2B. The 2028 model targets $3.6B revenue, $1.63B gross profit at 45% gross margin, $1.38B operating profit at 38%, and $1.2B net profit at 33%. The model assumes all fabs at 85% utilization after the $920M program, forward wafer selling prices, cost assumptions, and qualification timing. The CFO named wafer selling price as the largest swing factor. The Japan Track 2 expansion is not included in any model.
Trajectory
Revenue has stepped up through 2026: $372M in Q2 2025, $396M in Q3, $440M in Q4, $414M in Q1 2026 and a record in Q2 2026. Gross margin on a reported basis has expanded from 21.5% in Q2 2025 to 30% in Q2 2026 as the richer silicon photonics mix grows. Management attributes the gains to product mix and operational execution, and reported incremental gross margins of 58% on the sequential revenue increase. The non-AI base is a drag: 300mm RFSOI revenue fell 14% year-over-year and full-year 2026 RF mobile is expected down against the prior year.
The Model
The model projects FY+1 revenue of $1,973.7M with EBITDA of $772M, a 39.1% EBITDA margin. For FY+2 it projects revenue of $2,930M with EBITDA of $1,351M, a 46.1% EBITDA margin. The near term is anchored by the silicon photonics capacity ramp and contracted 2027 revenue. FY+2 depends on the Japan 300mm expansion, the shift toward higher-margin optical products, and the utilization assumptions management has stated.
| Metric | FY2025 | Next FY (E) | Following FY (E) |
|---|---|---|---|
| Revenue | $1.6B | $2.0B | $2.9B |
| YoY Growth | — | +26.0% | +48.5% |
| EBITDA | $496M | $772M | $1.4B |
| EBITDA Margin | 31.7% | 39.1% | 46.1% |
Projections are the median of 5 independent model runs. The model’s revenue sits 15.8% above analyst consensus.
Management guided Q3 2026 revenue to $520M midrange, which it framed as an annualized revenue run rate above $2B. The 2028 model targets $3.6B revenue, $1.63B gross profit at 45% gross margin, $1.38B operating profit at 38%, and $1.2B net profit at 33%. The model assumes all fabs at 85% utilization after the $920M program, forward wafer selling prices, cost assumptions, and qualification timing. The CFO named wafer selling price as the largest swing factor. The Japan Track 2 expansion is not included in any model.
What Could Go Right — and Wrong
- Silicon photonics reaches the $1B annualized run rate target in Q4 2026, converting added capacity into revenue.
- The contracted 2027 SiPho book is raised above $1.3B, closing the gap with the raised 2028 model.
- Japan Track 1 reaches full production readiness in Q4 2027 and adds 300mm capacity.
- RFSOI turns with the 3x 300mm wafer-start step-up by mid-2027.
- Wafer pricing holds firm as capacity comes online, supporting the 45% gross margin target.
- The contracted SiPho book stays flat while the model rises, leaving the raise resting on demand intent.
- Industry capacity additions pressure wafer pricing and the gross margin target.
- Near-package optics slips to the 2028 timeline that component suppliers describe, rather than 2027.
- The Japan Track 2 build or tool schedule slips, or requires external capital beyond the internal funding plan.
- The Intel Fab 11X mediation stays unresolved, with no update and a committed investment in the facility.
Looking Ahead
Over the next year the milestones are dated and checkable. Q3 2026 revenue guidance is the next print, the silicon photonics run-rate target is Q4 2026, and wafer-start capacity is expected to fully ramp in Q4 2026 with full financial effect in Q2 2027. Japan Track 1 targets full production readiness in Q4 2027, with Track 2 installed and functioning by Q4 2028. Management has said it will update the long-term financial model in the first quarters as it gets into 2027 once the Track 2 build-out is known. RFSOI is expected to step up 3x in 300mm wafer starts by mid-2027.
- Q3 2026Q3 revenue print — Tests the Q3 midrange guide and the delivery streak.
- Q4 2026SiPho run-rate milestone — Silicon photonics annualized run-rate target.
- Q4 2026Wafer-start ramp — Full wafer-start ramp; financial effect expected Q2 2027.
- Early 2027Long-term model update — Updated model once the Track 2 build-out is known.
- Mid-2027RFSOI 3x step-up — 3x 300mm RFSOI wafer starts versus Q2 2026.
- Q4 2027Japan Track 1 ready — Full production readiness for the Arai 300mm repurpose.
Financials
Annual Summary
| Metric | FY2024 | FY2025 | TTM | YoY |
|---|---|---|---|---|
| Revenue | $1.4B | $1.6B | $1.7B | +9.1% |
| Gross Margin | 23.6% | 23.0% | 26.8% | 60bps |
| EBITDA | $466M | $496M | $596M | +6.4% |
| EBITDA Margin | 32.5% | 31.7% | 34.8% | 79bps |
| Net Income | $208M | $220M | $290M | +6.1% |
| Free Cash Flow | $12M | −$45M | $305M | — |
| Net Cash | — | — | — | — |
Key Ratios (Trailing)
- P/E TTM—
- EV/EBITDA TTM—
- EV/Revenue TTM—
- Price/FCF TTM—
- Gross Margin (TTM)26.8%
- EBITDA Margin (TTM)34.8%
- Net Margin (TTM)16.9%
- ROIC12.5%
- FCF Conversion51.1%
- SBC / Revenue0.0%
The Company
Tower Semiconductor operates as a specialty analog foundry. It does not compete on leading-edge transistor density; it builds differentiated specialty analog process technologies, and management says a portion of its wafers sell at higher prices than commoditized standard CMOS. Its most important technology is silicon photonics (SiPho), the photonic ICs used in the optical transceivers that connect AI processors. Management says high-bandwidth, low-latency, energy-efficient optical connects have become a critical enabler of AI infrastructure.
Tower runs its own fab network rather than outsourcing production. It has a 200mm facility in Israel, two 200mm facilities in the U.S., two facilities in Japan through the 51%-held TPSCo (one 200mm, one 300mm), and shares a 300mm clean room and capacity with ST in Italy. Named fabs include Fab 2 in Migdal Haemek, Israel; Fab 3 in Newport Beach, California; Fab 5/Tonami and Fab 7/Uozu in Japan; Fab 9 in San Antonio, Texas; and Fab 10 in Agrate, Italy. The company reports as a single segment — its analog foundry operations — and does not allocate resources among segments.
Business Segments
Competitive Landscape
Tower competes in a foundry industry where the largest players are all adding photonics capacity. Management described itself in Q1 2026 as having by far the leading silicon photonics market share; on the Q2 2026 call, asked directly about peer capacity additions, the CEO said supply is increasing and that he did not have a good feel for how much is coming. The company's stated defense is lead-customer qualification, multi-generation co-development, exclusivity agreements, and contracts through 2028.
- GlobalFoundriesNamed as a competitor, 'mainly in the RF space.' Neighbor reads: comms infra and data center up 62% YoY, more than 40 SiPho customers, SiGe oversubscribed through 2027. Filed three patent lawsuits against Tower in March 2026, which the company disputes.
- TSMCNamed among pure-play advanced-node providers that also provide specialty. Neighbor read: the COUPE optical-interconnect platform is in production.
- Named among pure-play advanced-node providers. Neighbor read: first mass production of a 12-inch photonics IC, with a TFLN modulator in production.
- SMICNamed in filings; not discussed.
- Named in filings; not discussed.
Supply Chain
Tower sits upstream in the optical interconnect chain, supplying photonic ICs and analog silicon to transceiver and module makers. Its key inputs are III-V epitaxial material and wafer-fab equipment. Neighbor companies including Lightwave Logic have cited Tower by name.
Related companies
See all Chip Making companies → · How this layer works: Chapter 1, The Chip →
More on TSEM: Earnings recap