Capex cycle — fabs, Tower-ST Agrate, Tower-Intel Fab 11X, $920M SiPh envelope
As of: 2026-04-29 (data through Q4 2025 6-K release filed 2026-02-11; FY 2025 20-F expected April-May 2026 will provide additional capex breakouts)
Reference: Capex figures reconcile to each year’s 6-K cash-flow statement / 20-F MD&A “Investing Activities” line. Site-level capex is not disclosed publicly by Tower; allocation by fab is analyst-inferred from program announcements.
Confidence legend: ✓ verified-primary 6-K / 20-F · ◐ aggregator / press release · ⚠ inferred / estimate
1. Capex history — six-year trail
| Year | Capex ($M) | YoY $Δ | YoY %Δ | Capex / Revenue | Notes |
|---|---|---|---|---|---|
| FY 2020 | 74 ⚠ | — | — | 6% | Pre-Agrate baseline; minimal expansion spend |
| FY 2021 | 251 ⚠ | +177 | +239% | 17% | Tower-ST Agrate JV announced June 2021; initial cleanroom equipment Phase 1 spend begins |
| FY 2022 | 377 ⚠ | +126 | +50% | 22% | Agrate Phase 2 cleanroom equipment + Newport Beach maintenance + TPSCo Japan |
| FY 2023 | 471 ⚠ | +94 | +25% | 33% | Agrate continued; early Tower-Intel Fab 11X equipment scoping (Sep 2023 announcement) |
| FY 2024 | 432 ✓ | −39 | −8% | 30% | Agrate productized 2024-12-23; Tower-Intel Fab 11X equipment ramp |
| FY 2025 | 437 ✓ | +5 | +1% | 28% | Stable run-rate; SiPh envelope 28% paid through Q4 2025 |
| FY 2026e ⚠ | 700-900 | +260-470 | +60-105% | ~40-45% | Peak — $920M SiPh/SiGe envelope deployment + maintenance |
| FY 2027e ⚠ | 500-700 | −200 | −15-30% | 25-30% | Tail of SiPh capacity build; FY 2028 production-ready |
| FY 2028e ⚠ | 300-400 | −200-300 | −40-45% | 12-14% | Reverts toward maintenance levels |
Sources: FY 2024 + FY 2025 capex per Q4 2024 + Q4 2025 6-K releases (Tower Q4 2025 release ✓ and Tower Q4 2024 release ✓). FY 2020-2023 values via StockAnalysis ◐ aggregator; precise primary-source extraction queued from FY 2022 + FY 2023 20-F filings.
The FY 2020 → FY 2025 cumulative capex is approximately $2.0B — equating to ~16% of cumulative revenue ($12.5B over 6 years) and almost equal to current PP&E gross book value before depreciation. This is structurally heavy capex spend typical of a specialty foundry going through capacity-extension cycles. The FY 2026 peak of ~$700-900M would push cumulative six-year capex to ~$2.7-2.9B — a doubling of the existing PP&E base on a gross-spend basis.
The FY 2026 capex guide gap. Management has not publicly disclosed a specific FY 2026 capex dollar number on the Q4 2025 call. The $700-900M range is analyst-derived from (a) the $920M SiPh/SiGe envelope of which
28% ($258M) is already paid through Q4 2025, leaving ~$662M to deploy across 2026-2028, plus (b) maintenance capex run-rate of ~$200-300M baseline, plus (c) the Tower-Intel Fab 11X equipment ramp continuation. The precise FY 2026 capex range is the single most important open data point for forward FCF / leverage modeling — flagged for Q1 2026 6-K extraction (TODO ⚠).
2. Capex composition — site-by-site allocation
Tower discloses capex on a consolidated cash-flow line only. Site-level capex is not publicly disclosed. The following is analyst allocation based on cumulative program announcements.
2.1 Newport Beach, CA — Fab 3 (200mm, legacy Jazz) + Fab 9 (200mm, legacy Maxim)
Strategic role: The 200mm specialty-process backbone — the PH18 silicon-photonics process is fabricated in Fab 3 (per Tower-Jazz May 2008 merger heritage). Fab 9 was acquired from Maxim Integrated in May 2016 for ~$70M, providing an additional 200mm BCD power / SiGe BiCMOS line on the same campus.
Capex composition:
- Maintenance + tooling: ~$50-80M annually for sustaining capex on existing 200mm equipment
- PH18 SiPh capacity expansion: the majority of the $920M SiPh/SiGe envelope is allocated to Fab 3 capacity extension — adding 200mm SiPh wafer-starts capacity to the December 2026 target of >5× Q4 2025 actual
- CPO foundry technology platform (announced Nov 12 2025) — Fab 3 supplemented by hybrid 200mm + 300mm capacity flows for co-packaged optics (GlobeNewswire ✓)
- Q4 2025 $105M Fab 3 lease prepayment — disclosed at Q4 2025 release; one-time facility cost prepayment that depressed Q4 2025 OCF (normalized ex-prepayment OCF ~$145M for the quarter)
Why is Newport Beach the load-bearing capex site? Because PH18 is the unique merchant 200mm SiPh process that anchors the bull thesis. GlobalFoundries Fotonix is a 300mm 45CLO process; Intel’s Mod-9 is internal/non-merchant; TSMC SiPh is internal/limited merchant. Tower’s PH18 captures the only 200mm merchant SiPh capacity at scale — and the $920M envelope is the capacity step that turns this into a structural duopoly with GFS Fotonix. A material portion of the FY 2026 capex peak flows here.
2.2 Migdal Haemek, Israel — Fab 1 (150mm) + Fab 2 (200mm)
Strategic role: Tower’s headquarters site and historical specialty-CMOS / image-sensor base. Fab 1 (150mm) is being run down — management has explicitly discontinued legacy 150mm flows, transferring volume to 200mm Fab 2 to support gross-margin expansion. Fab 2 (200mm) is the primary RF Mobile / image sensor / power management workhorse.
Capex composition:
- Maintenance + tooling: ~$40-60M annually
- Fab 2 selective capacity expansion: envelope-tracker ramp for the “tier-one handset” customer (widely understood to be Apple iPhone PMIC, ⚠ inferred) is a key forward growth driver
- No greenfield expansion at Migdal Haemek — strategic constraint due to Israeli geopolitical-risk overlay; Tower has prioritized geographic diversification (Newport Beach, Agrate, Tower-Intel Fab 11X, TPSCo Japan) over Israel-only build
2.3 Agrate Brianza, Italy — Tower-ST 300mm joint venture
Strategic role: The 300mm capacity-extension partnership with STMicroelectronics, announced June 24, 2021. Tower owns a minority share of the joint venture; ST owns the majority. The facility is a greenfield 300mm specialty-process line in Lombardy, Italy. First 65nm BCD platform productized December 23, 2024 (Tower release ✓).
Capex composition (Tower’s portion):
- Initial cleanroom equipment 2021-2023: approximately $300-400M of cumulative Tower investment in the JV through 2023 ⚠ (estimate; specific figure not disclosed)
- Productization 2024: Tower began absorbing operational cost in Q4 2024 ahead of meaningful revenue contribution per Q4 2024 6-K commentary
- 2025-2026 ramp: continued capex absorption + utilization fill-out drives FY 2025-2026 P&L impact
- Forward maintenance + extensions: modest ($30-50M annually) as Agrate enters full operational phase
Why Agrate matters for capex modeling: The Q4 2024 6-K explicitly cited “the company took on for the first time its portion of incremental costs of the greenfield Agrate facility” (GlobeNewswire ✓). This is the load-bearing FY 2025 GM-compression driver — Agrate startup costs absorb gross profit while utilization is sub-optimal. The offset is sequential GM expansion as utilization climbs through 2026-2027; the FY 2025 quarterly progression Q1 (20.4%) → Q4 (26.8%) confirms the trajectory.
2.4 Tower-Intel Fab 11X (Rio Rancho, NM) — capacity-services arrangement
Strategic role: Announced September 5, 2023 (Tower release ✓) — Tower deploys $300M of equipment capex into Intel’s existing Fab 11X (Rio Rancho, NM) in exchange for a 600,000+ photo-layer-per-month 300mm corridor for Tower’s specialty processes. This is a capacity-services arrangement, NOT a Tower acquisition of Fab 11X.
Capex composition:
- Initial $300M Tower equipment: announced September 2023, ramp 2024-2026
- Production ramp 2024-2025: equipment installation + qualification flows
- Forward incremental: if Tower elects to expand the corridor beyond the initial 600K-photo-layer block, additional capex would be required; no public commitment beyond the original $300M
Read on Tower-Intel. This is the lowest-capital-intensity way Tower has accessed 300mm specialty-process capacity in the US — Tower pays only for equipment, while Intel provides the cleanroom + utilities + facility overhead under a capacity-services contract. The strategic value is US 300mm 65nm-class specialty supply for customers requiring domestic supply (CHIPS-aligned end markets, defense / automotive). The financial impact is modest in P&L terms but strategically de-risks Tower’s geographic concentration vs Israel + Newport Beach + Italy.
2.5 TPSCo Japan — Uozu (200mm) + Arai (300mm)
Strategic role: The Tower-Nuvoton joint venture covering Japan-based 200mm + 300mm legacy specialty CMOS. Ownership flipped to Nuvoton majority in 2020; Tower retains minority interest. Capex flows are JV-internal and not consolidated to Tower’s primary capex line.
Capex composition:
- Modest equity-method allocation — Tower’s share of any TPSCo capex flows through the equity-method investment line, not the cash-flow capex line
- Strategic role declining — TPSCo is a legacy infrastructure asset rather than a growth-capex investment for Tower
3. The $920M SiPh/SiGe capacity envelope — the forward capex spine
The most important capex-modeling data point for FY 2025-2028 is the $920M SiPh/SiGe envelope announced in two tranches:
| Tranche | Date | $M | Cumulative | Source |
|---|---|---|---|---|
| Tranche 1 | Q3 2025 release (2025-11-10) | $650M | $650M | Tower release ✓ |
| Tranche 1 expansion | November 2025 announcement | +$300M (combined with new CPO foundry tech) | n/a (overlap with Tranche 1) | GlobeNewswire ✓ |
| Tranche 2 | Q4 2025 release (2026-02-11) | +$270M | $920M | Tower release ✓ |
| Total committed envelope | $920M |
Payment trajectory:
- ~28% paid through Q4 2025 = approximately $258M already deployed (per Q4 2025 release commentary)
- Remaining ~$662M to be deployed across 2026-2028
- Approximate annual cadence: ~$300-400M FY 2026, ~$200-250M FY 2027, ~$60-100M FY 2028 ⚠ (analyst-derived from typical specialty-foundry capacity-build cadence)
Capacity targets (per TipRanks ✓ Q4 2025 release commentary):
- December 2026 SiPh wafer-starts capacity > 5× Q4 2025 actual — implies a step-function capacity expansion roughly aligned with FY 2027 production
- >70% of capacity reserved or in process of being reserved through 2028, with customer prepayments
- Customer prepayments materially fund the remaining capex — the qualitative disclosure suggests hyperscaler / module-OEM customers are providing upfront cash that partially offsets Tower’s own capex burn
Why the customer-prepayment language is structurally important. It changes the cash-flow math materially. If $660M remaining capex is partially pre-funded by customers (analyst estimate: $200-400M of customer prepayments through 2026-2028 ⚠), then Tower’s net cash deployment from balance sheet is closer to $260-460M rather than the headline $660M. This brings the FY 2026 net-cash-burn risk to a manageable level without requiring debt issuance or equity raise. The specific dollar quantum of customer prepayments is not disclosed at FY 2025 release; the FY 2025 20-F deferred-revenue line will likely give the first explicit number.
4. Forward capex profile — analyst working assumptions
| Year | Capex range ($M) ⚠ | Mid-point ($M) | Capex/Revenue (mid) | Driver detail |
|---|---|---|---|---|
| FY 2026 | 700-900 | 800 | ~44% | Peak: $920M envelope balance + maintenance + Fab 11X tail + Agrate sustaining |
| FY 2027 | 500-700 | 600 | ~25% | Tail of SiPh envelope + Agrate sustaining; production-ready by Dec 2026 |
| FY 2028 | 300-400 | 350 | ~12% | Maintenance + selective extensions; envelope deployed |
| FY 2029 | 280-350 | 315 | ~9-10% | Maintenance steady-state |
| FY 2030 | 250-330 | 290 | ~8-9% | Maintenance steady-state |
Capex / Revenue trajectory: Spikes from ~28% (FY 2025) → ~44% (FY 2026 peak) → reverts to ~10-12% maintenance by FY 2028+. The +1,600 bps capex/revenue spike in FY 2026 is the primary FCF-tension data point through the cycle.
Implied FCF trajectory (analyst working assumption ⚠):
| Year | OCF (analyst, $M) | Capex (mid, $M) | FCF (headline, $M) | FCF margin |
|---|---|---|---|---|
| FY 2025 | 395 | 437 | (42) | (3%) |
| FY 2026 | 470 | 800 | (330) | (19%) |
| FY 2027 | 580 | 600 | (20) | (1%) |
| FY 2028 | 850 | 350 | +500 | +18% |
| FY 2029 | 1,050 | 315 | +735 | +24% |
The FY 2026 FCF trough of ~$(330)M is the load-bearing balance-sheet stress point. With ~$1.0B net cash entering FY 2026, Tower can absorb a single year of $300M+ FCF burn without drawing debt. Two consecutive years of similar burn would require external financing or working-capital release. This is the central reason that customer-prepayment de-risking matters — it can pull forward $200-400M of cash inflows that smooth the trajectory.
5. Comparison to specialty-foundry peer capex
| Comp | FY 2025 Capex ($B) | FY 2025 Capex / Revenue | Forward peak |
|---|---|---|---|
| Tower (TSEM) | $0.44 | 28% | FY 2026 peak ~40-45% |
| GlobalFoundries (GFS) | $0.72 | 10.6% | FY 2026-27 ramp to ~14-16% |
| VIS | ~$0.30 ◐ | ~21% ◐ | Steady mid-cycle |
| DBHiTek | ~$0.20 ◐ | ~20% ◐ | Steady |
| UMC | ~$1.5 ◐ | ~20% ◐ | Steady mid-cycle |
| TSMC | ~$40 | ~36% | Peak cycle |
Tower’s FY 2026 capex/revenue ratio of ~40-45% would be the highest in the merchant-foundry comp set ex-TSMC — reflecting the concentrated, single-program nature of the SiPh envelope vs the diversified capex base of GFS / UMC / VIS. The cycle nature is front-loaded rather than steady — Tower spends heavily for one year then reverts to maintenance, vs GFS which has a more sustained mid-cycle level supported by CHIPS subsidies.
Tower has NO equivalent of CHIPS funding for the $920M envelope. The Israeli Investment Center and Israel Innovation Authority do provide some grant funding for fab expansion ([per the FY 2024 20-F deferred-grant-income line, TODO ⚠ for direct extraction]), but the dollar quantum is small relative to GFS’s $1.575B + $570M federal/state stack. Tower’s reinvestment is fully self-funded plus customer prepayments; this is a meaningful structural difference vs GFS’s subsidized capex base.
6. Run-rate trajectory through FY 2028 — the maintenance floor
Once the $920M envelope is fully deployed (target completion 2027 H2), Tower’s steady-state maintenance capex reverts to:
- Newport Beach Fab 3 + Fab 9: ~$80-120M annually (combined sustaining capex on PH18 + adjacent specialty processes)
- Migdal Haemek Fab 1 + Fab 2: ~$40-60M annually (Fab 2 only as Fab 1 runs down)
- Tower-ST Agrate maintenance: ~$30-50M annually (post-productization)
- Tower-Intel Fab 11X sustaining: ~$20-40M annually
- TPSCo Japan equity-method allocation: ~$10-30M annually
- Selective extensions / new-process tape-outs: ~$50-100M annually
Total maintenance run-rate: ~$300-400M annually — roughly equal to FY 2024-2025 pre-envelope levels but supporting a larger revenue base (FY 2028 model $2.84B vs current $1.57B). Capex / revenue at the FY 2028 model = ~12-14% steady state, consistent with peer specialty-foundry maintenance levels.
The “post-cycle FCF inflection” is the bull-thesis underwrite. If Tower hits the FY 2028 model ($2.84B revenue × 31.7% Op margin = ~$900M op profit + ~$500M D&A = ~$1.4B EBITDA), and capex normalizes to $300-400M, then steady-state FCF ramps to ~$700-900M annually — a >10× lift from current levels and a strong basis for capital-return initiation in FY 2028+. This is what the 13.0× current EV/Revenue multiple is implicitly underwriting.
7. Open items / backfill queue
- Specific FY 2026 capex guide — Tower has not provided a dollar number; analyst $700-900M range is implied. Q1 2026 6-K (expected 2026-05-13) and FY 2025 20-F (expected April-May 2026) are the most likely disclosure venues. TODO ⚠.
- Customer-prepayment dollar quantum — referenced qualitatively as “firmly backed with customer prepayments”; specific dollar figure not disclosed. FY 2025 20-F deferred-revenue note is the primary source. TODO ⚠.
- Site-by-site capex disclosure — Tower does not disaggregate capex below the consolidated cash-flow line. Industry-channel triangulation queued.
- Tower-ST Agrate JV cash contributions — Tower’s portion of Agrate capex flows through JV financial statements; specific Tower contribution figures are not in the consolidated 6-K. Equity-method or proportional-consolidation accounting determines the precise treatment.
- Israeli Investment Center / Innovation Authority grants outstanding — Tower has historically received Israeli grants for fab expansion. Outstanding deferred-grant-income balance + cash-receipt cadence not extracted. TODO ⚠.
- Tower-Intel Fab 11X capex extension beyond initial $300M — if Tower elects to expand the Rio Rancho corridor in 2027-2028, additional capex would be required; no public commitment to date.
Sources
- Q4 2025 6-K release (filed 2026-02-11) — Tower release ✓ and GlobeNewswire ✓. FY 2025 capex $437M, $920M SiPh/SiGe envelope expansion +$270M, Fab 3 lease prepayment.
- Q3 2025 6-K release (filed 2025-11-10) — Tower release ✓ and GlobeNewswire ✓. $650M initial SiPh/SiGe expansion announcement.
- Q4 2024 + FY 2024 6-K release (filed 2025-02-10) — GlobeNewswire ✓. FY 2024 capex $432M; Agrate first absorbing operating cost.
- Tower-Intel Fab 11X announcement (2023-09-05) — Tower release ✓. $300M Tower equipment investment in Fab 11X Rio Rancho.
- Tower-ST Agrate JV announcement (2021-06-24) — Tower release ✓. Initial JV announcement.
- Tower-ST first product announcement (2024-12-23) — Tower release ✓. 65nm BCD productization.
- CPO foundry technology platform (2025-11-12) — GlobeNewswire ✓.
- TipRanks Q4 2025 summary — TipRanks ✓. FY 2028 financial model + capacity targets.
- FY 2024 20-F (acc. 0001178913-25-001537, filed 2025-04-30) — SEC EDGAR ✓. Audited capex baseline.
- StockAnalysis aggregator — StockAnalysis ◐. FY 2020-2023 capex history.
Cross-references
- Quarterly trend — quarterly capex cadence
- Balance sheet — net cash position, $920M envelope mechanics, customer prepayments
- Margins and pricing — utilization-driven GM expansion that justifies the capex
- Capital returns — reinvestment-heavy posture; capex deployment vs returns
- Comps and valuation — peer capex/revenue benchmarking
- DCF assumptions — capex / revenue trajectory in DCF model
- Technology overview — PH18, Tower-ST Agrate 300mm, Tower-Intel Fab 11X process detail
- Bull case — pillar (iii) capacity expansion ramp underwrite
- Bear case — pillar (iii) AI-photonics revenue dilution / capex-vs-FCF tension