Business Overview

Tower Semiconductor is a specialty analog/mixed-signal semiconductor foundry. Pure-play — no proprietary chips. It manufactures wafers for fabless designers and IDMs on differentiated process platforms: silicon photonics (SiPho), silicon germanium (SiGe), RF silicon-on-insulator (RFSOI), bipolar-CMOS-DMOS (BCD), and CMOS image sensors (CIS).

The distinction that matters: Tower competes on process technology expertise, not transistor node shrink. While TSMC and Samsung race to 2nm, Tower operates at 65nm-180nm on specialty platforms where the value is in the physics — integrating photonics with electronics, handling high-frequency RF signals, managing power efficiently. These are analog problems where decades of process knowledge compound.

Revenue model is per-wafer pricing, value-based. A SiPho wafer commanding premium pricing for AI optical transceivers uses the same 200mm fab as a legacy discrete wafer, but the revenue and margin per wafer are dramatically different. This is the entire financial thesis in one sentence: replace low-value wafers with high-value wafers in the same fabs.

Revenue Composition (FY2025, $1.566B)

The composition tells the story. Three platforms growing (SiPho/SiGe, power, sensors), three shrinking (RF mobile, discrete, CMOS). Management frames this as intentional cannibalization — CEO Ellwanger on Q4 call: "proactively working with our customer partners to responsibly reduce exposure to lower margin controller offerings in favor of higher-value optical and RF mix." The alternative reading: discrete was weak industry-wide in 2025 (automotive softness) and mixed-signal CMOS competes on price with TSMC and UMC. Some shared resources (cleanroom space, engineering attention, utilities) are being reallocated to SiPho, but the declines may partly reflect soft end-market demand reframed as strategic choice. Either way, the margin impact is the same — higher-mix revenue replacing lower-mix revenue.

The SiPho Variable

Silicon photonics is the single variable that transforms Tower's financial profile. SiPho revenue trajectory:

2023:  $28M   (7% of revenue)
2024:  $106M  (7% of revenue — company-wide denominator grew)
2025:  $228M  (15% of revenue)
Q4'25: $95M   ($380M annualized run rate, ≈22% of revenue)*

*Q4 SiPho revenue includes some non-wafer NRE for Gen+1 and Gen+2 development (CEO disclosure). The production-only run rate is somewhat below $380M annualized.

SiPho is growing at 115% YoY while the rest of the business is flat to declining. The Q4 run rate (inclusive of NRE) is already 67% of full-year 2025 SiPho revenue — the acceleration in H2 2025 is steeper than most models capture.

Why SiPho matters for AI: Every GPU cluster needs optical interconnects to move data between nodes. The transition from 400G → 800G → 1.6T → 3.2T pluggable optical transceivers requires silicon photonic integrated circuits (PICs) — the silicon chips that route and modulate light. Tower claims to be "by far the majority supplier of 1.6T silicon PICs." The 1.6T node is the fastest-growing in the industry, driven by NVIDIA's networking ecosystem and hyperscaler buildout.

Tower is investing $920M in total SiPho/SiGe capex to achieve 5x Q4 2025 monthly wafer shipments. Management states over 70% of this expanded capacity is reserved through 2028 with customer commitments — though the balance sheet shows deferred revenue declining, raising questions about how "reserved" translates to contractual cash (see Forward Expectations Gap). Equipment installation targets completion by December 2026. Full revenue impact by 2028.

Fab Footprint

Fabs 2 and 9 at 60-65% utilization represent significant operating leverage as SiPho capacity qualifies. Fab 7 at >85% is a near-term constraint, particularly as it absorbs customer flows redirected from the Intel Fab 11X dispute.

Financial Profile

Four-Year P&L (Normalized)

*FY2023 normalized — GAAP includes $313.5M Intel merger termination fee **FY2023 OCF inflated by termination fee cash receipt ***FY2025 OCF depressed by $105M Fab 3 lease prepayment; adjusted OCF ≈$500M

The normalized earnings story: net profit has been flat at ≈$200-265M for four years through a full semiconductor cycle (peak in 2022, trough in 2023-2024, inflection in 2025). FY2025 is where the mix shift becomes visible — annual margins are still compressed, but the Q4 exit rate tells a different story.

Quarterly Trajectory (FY2025) — This Is the Inflection

*Q4 includes nonrecurring tax benefit (2% effective rate). Normalized Q4 net margin ≈15.5-16%.

Revenue grew $82M from Q1 to Q4, and $40M dropped to net profit — CEO-stated 48.78% incremental net margin. Important caveat: Q4 included a nonrecurring tax benefit (2% effective rate vs ≈12-15% normalized). At a normalized 15% tax rate, Q4 net profit would have been ≈$68-70M, putting the tax-adjusted incremental net margin at ≈36-37%. The realized incremental gross margin Q1→Q4 was 53.7% ($44M / $82M), unaffected by the tax distortion. The mix enrichment mechanism is real, but the 48.78% headline overstates the after-tax economics by approximately 12 percentage points.

Balance Sheet

Fortress. $1.15B in cash and deposits against $161M total debt. Net cash ≈$991M. Current ratio 6.5x. Debt/equity 6%. The $920M capex is funded entirely from operations and balance sheet. No dilution, no leverage.

The critical nuance: FCF is negative in the investment phase (FY2025: -$42M GAAP, +$63M adjusted for lease prepayment). This is the trough. Post-capex (FY2028+), management's model implies operating cash flow of ≈$1.05B (NI $750M + D&A $400M - WC $100M) against maintenance capex of ≈$250M, yielding ≈$800M annual FCF. At current market cap, that's 18x P/FCF on the model.

2028 Target Financial Model

*Using 113.6M diluted shares (FY2025 diluted count). Assumes: 85% utilization across all fabs, $920M capex deployed, modest ASP reduction on existing products offset by new product introductions, ≥15% effective tax rate (Pillar 2), EXCLUDES Fab 11X entirely.

The model's credibility rests on three pillars: (1) high incremental gross margin on SiPho mix — the model assumes 59%, FY2025 Q1→Q4 realized 53.7%, (2) operating leverage as revenue scales over a relatively fixed cost base, and (3) SiPho mix reaching 40-50% of revenue. The first pillar is directionally proven but the model requires further margin improvement beyond what FY2025 demonstrated. The second is mechanical. The third requires the $920M capex to execute on schedule and demand to persist.

Non-operating income dependency: FY2025 pre-tax profit of $240M included $46M in net financing income (19% of total) earned on the ≈$1.15B cash/deposit balance. As cash deploys into capex, this income shrinks — partially offset by higher operating profit, but creating a bridge risk in FY2026-2027 where operating profit must grow fast enough to backfill declining interest income.

Tax Structure — Pillar 2 Transition

Tower's Israeli operations have benefited from a 7.5% "Preferred Enterprise" tax rate. The OECD Pillar 2 global minimum tax (15%) kicks in starting FY2026. Tower qualified for a transitional safe harbor through December 31, 2025.

FY2025 effective tax rate:  8.8%  (full year), 2% (Q4 — nonrecurring benefit)
FY2026+ guided:             ≥15%  (all jurisdictions)

On the 2028 model's pre-tax income of ≈$882M, the difference between 8.8% and 15% is approximately $55M in additional annual tax. The $750M net profit target already incorporates the 15% rate. However, the year-over-year EPS comparison in 2026 will be ugly — Q4 2025 at 2% effective rate makes a high baseline that Q1 2026 at 15% can't match. Consensus has Q1 2026 at $0.57 EPS versus Q4 2025 actual of $0.78 — a 27% sequential decline that is entirely tax-driven, not operational.

Subsidiary tax signals: Tower US Holdings released its deferred tax valuation allowance in FY2025 — a positive signal indicating US operations are now generating sufficient taxable income. However, TSIT (Italy/Fab 10) accumulated NOLs grew to ≈$13M (from $5.5M), confirming the Italy operations remain loss-making in early ramp. The net picture: US improving, Italy still burning.

Competitive Position

SiPho Foundry Landscape — Duopoly Forming

GlobalFoundries (GFS) is the primary SiPho competitor. GFS reported >$200M in SiPho revenue for FY2025 (roughly doubled YoY), with corridors "oversubscribed" and a target of $1B SiPho run rate by end of 2028. GFS acquired AMF (Singapore) and InfiniLink to accelerate its SiPho roadmap. GFS has 300mm wafer capability (vs Tower's primarily 200mm), which provides scale advantages at volume.

UMC is targeting industry-standard SiPho PDK delivery for 12-inch wafers by 2027 — two or more years behind Tower and GFS. Not a near-term competitive threat, but confirms the SiPho foundry market is attracting new entrants.

No one else is at scale. The SiPho foundry market is a duopoly today (Tower + GFS), likely becoming a triopoly (adding UMC) by 2028.

GFS vs TSEM — The Scoreboard

Tower is ahead on revenue scale and growth rate, and far ahead on stock performance. The factor regression confirms this isn't just sector beta — GFS has -13% alpha while TSEM has +95%. Same business, 108 percentage point gap. This proves company-specific execution, not just SiPho demand.

But GFS has structural advantages: 300mm wafer capability (more dies per wafer, lower unit cost at scale), a global manufacturing footprint, and aggressive capacity expansion through acquisition (AMF). GFS also won a CPO design win on its CLO platform, positioning for the next architectural transition. The SiPho competitive moat may be narrowing.

Customer Ecosystem

Tower files as a single segment and does not name SiPho customers in audited filings. Customer identification comes from cross-referencing transcripts, partnerships, and supply chain inference:

Broadcom (AVGO) — Almost certainly the largest SiPho customer. Broadcom is the "only player" shipping 1.6T DSPs, and Tower claims "by far the majority supplier" of 1.6T PICs. Broadcom's optical networking business is growing ≈60% YoY. Hock Tan's Q1 FY2026 call: "Optical components like lasers, pin diodes, going nuts."

Coherent (COHR) — Key module maker customer. Book-to-bill >4x. Calendar 2026 booked out, 2027 filling quickly. Supply-demand imbalance on indium phosphide expected through 2027+. Coherent offers 1.6T transceivers on three platforms (SiPho, EML, VCSEL) — the architecture battle matters.

eOptoLink — China's largest optical transceiver maker, keynoted Tower's China symposium. Asia ex-Japan revenue growth from 27% to 39% of Tower's total likely reflects Chinese optical module demand.

NVIDIA — Partnership announced February 5, 2026 for 1.6T data center optical modules. Foundry-to-end-customer relationship (unusual), suggesting Tower's platform is reference-grade for the NVIDIA networking ecosystem.

Customer Diversification — Improving

From the FY2025 audited financial statements (Note 14):

Customer A:        11% (was 13% in 2024, 14% in 2023) — declining share
Customer B:         7% (was 11% in 2024, 9% in 2023) — declining share
Other (6 at 4-7%): 32% (was 16% in 2024) — EXPLODED

The "other customers" bucket doubled from 16% to 32% of revenue in one year. Six customers now each contribute 4-7% of revenue. Only one customer exceeds 10% of receivables (versus two in 2024). Customer concentration is improving rapidly.

Switching Costs and Defensibility

SiPho foundry switching costs are high: multi-month wafer qualification process, process-specific design rules, co-developed IP, and capacity reservation agreements with prepayments. Customers don't switch foundries mid-generation. However, next-generation transitions (1.6T → 3.2T) are natural re-evaluation points where customers can dual-source.

Tower is investing in structural lock-in through co-development of 3.2T material systems with lead customers and wafer-to-wafer bonding technology that integrates SiPho PICs with SiGe electrical ICs on a single bonded wafer — a capability neither GFS nor UMC has disclosed.

Counterparty Demand Validation

Every counterparty independently confirms explosive SiPho demand:

No counterparty reports demand weakness. The only constraint anyone mentions is supply, not demand. This is unanimous across the supply chain.

Management & Governance

Executive Team

CEO Russell Ellwanger — 20+ years at Tower. Built the company from near-bankruptcy to $14B market cap. Deeply involved in customer relationships ("deeply trust-rooted partnership alliances"). Controls the narrative on every earnings call. The driving force behind the SiPho bet — started development 8+ years ago with a single partner. Key man risk: no disclosed succession plan.

President Marco Racanelli, PhD — 20+ years at IBM Semiconductor. Brought leading-edge technical credibility. CFO Oren Shirazi — since 2004, 20+ years of institutional memory. COO Rafi Mor — since 2014. CTO Dr. Avi Strum — since 2019, ex-Intel. Exceptionally stable senior team with no recent turnover.

Board

Chairman Amir Elstein (ex-Intel Israel GM, ex-Teva board). Nine directors total. The standout: Carolin Seward, VP of Google Custom Silicon Sourcing — a direct line into hyperscaler demand planning. Google is one of the largest consumers of optical interconnects. This is a strategic relationship, not a decorative appointment.

Ownership

Aggregate insider ownership: 0.46% — low. No individual director or officer owns >1%. This is the main governance concern: management has limited direct skin in the game relative to a $14.2B market cap.

Major shareholders (>5%, March 2025):

Four Israeli institutional holders collectively own 26.1% — stable, long-term pension and insurance capital with low turnover. This reduces forced-selling risk. Point72 and Senvest at 12.2% combined represent sophisticated validation.

Compensation

CEO total compensation ≈$12.1M (FY2024): $970K base, $1.65M cash bonus, $9.5M equity-based (78% of total). PSUs tied to net profit and cash from operations targets. Long-term incentive PSU at $80 share price target (already exceeded at $126). Minimum shareholding requirement: CEO must hold 3x base salary (≈$2.9M) in shares.

Compensation structure is well-aligned. Heavily equity-based, tied to profit and cash metrics. Israeli Companies Law requires shareholder approval of compensation policy.

Capital Allocation Track Record

The SiPho bet was prescient. Management identified the opportunity 8+ years ago, invested through multiple cycles, and is now "by far the majority supplier." The $920M capex commitment funded from operations with zero dilution is aggressive and well-timed. Customer prepayments de-risk the investment. The Intel Fab 11X commitment ($300M in equipment at risk) is the one blemish — trusting Intel as a counterparty proved questionable.

No dividends, no buybacks, no M&A. All cash directed to organic growth. At this stage of the investment cycle, this is appropriate. The capital allocation question becomes interesting post-2028 when FCF inflects to ≈$800M/year.

Insider Transactions

Tower is a foreign private issuer. Insiders are not required to file SEC Form 4s. Insider transaction visibility is limited to Israeli regulatory disclosures not accessible via standard US data providers. This is a structural limitation — we cannot observe open-market purchases or sales by management.

As a proxy: the $920M capex commitment (6x FY2025 net income) deployed into a single technology bet is perhaps the strongest conviction signal possible. Actions over Form 4s.

Factor Profile

What Drives This Stock's Returns

Factor regressions across multiple specifications consistently show TSEM at 49-57% idiosyncratic variance — well below the 75% target.

This is half factor bet, half company bet.

The trailing annualized alpha is +61% to +95% depending on model specification — but this is a backward-looking descriptor of a stock that quadrupled, not evidence of forward mispricing. Any stock rising 338% in a year will show massive alpha in any regression. The more informative signal is the relative alpha versus peers in the same business (see GFS divergence below) and the latent factor question: "SiPho/optical AI interconnect" is an emerging return driver not captured by existing ETFs. As this factor gets recognized, some of TSEM's measured alpha will migrate to factor exposure — the thesis working is not the same as the alpha persisting.

The GFS Divergence — The Strongest Signal

The most informative finding in the factor profile: GFS has -13% alpha while TSEM has +95% alpha. Same SiPho duopoly, 108 percentage point gap. GFS explains only 5.9% of TSEM's variance. They don't co-move despite being in the same business.

If TSEM's returns were just "SiPho demand," GFS would show similar alpha. It doesn't. The divergence proves company-specific execution edge — Tower is doing something GFS isn't, and the market is rewarding it.

Portfolio Implications

At 49-57% idio and Edge% of 49-65% (depending on whether SiPho sector conviction counts), position sizing per the Paleologo framework:

For a 5% target position: effective edge-weighted exposure = 2.5-3.3%
Remaining 1.7-2.5% is unhedged factor exposure (tech sector + momentum)

To run this position clean, hedge semiconductor sector exposure (short SOXX/SMH) and monitor momentum risk. The +238% one-year return loads heavily on momentum — if momentum rotates, TSEM drops regardless of fundamentals.

Forward Expectations Gap

What Current Price Requires

At $126.34 and 26.4x forward P/E, the market implies NTM EPS of ≈$4.79 — a 147% jump from trailing $1.94. This requires:

1. Revenue ramping from $1.57B (FY2025) to ≈$2.0-2.2B in the NTM period
2. Gross margins expanding from 23.2% toward 30%+ as SiPho mix increases
3. Net margins absorbing the Pillar 2 tax step-up and still expanding
4. $920M capex executing on schedule (tools installed and qualified by Dec 2026)
5. No major customer loss or SiPho demand slowdown
6. Fab 3 lease resolved without disruption

Implied Model Probability

A scenario decomposition that matches the current price:

Beat model ($8 EPS × 25x = $200):   ≈25% probability
Hit model ($6.60 EPS × 20x = $132): ≈50% probability
Miss significantly ($3.50 × 15x):   ≈15% probability
Major setback ($1.50 × 12x):        ≈10% probability

EV = 0.25×200 + 0.50×132 + 0.15×52.5 + 0.10×18 = $125.68 ≈ current price ✓

The market prices ≈75% probability of at least hitting the $2.84B/$750M model. Street mean price target of $159 implies higher model probability — getting there requires 2-3 quarters of execution confirmation.

Six Disconnects

1. Demand is stronger than priced. Counterparty evidence is unanimous — COHR book-to-bill >4x, GFS corridors oversubscribed, Broadcom optical "going nuts," Marvell interconnect +50% YoY. No one reports demand weakness. But the market already prices strong demand. The edge isn't "demand is good" — it's "demand is THIS good and durable through 2028."

2. Fab 3 lease risk is not priced. The landlord has requested a judicial declaration of "material non-curable breach" at Tower's primary SiPho facility in Newport Beach. The 20-F explicitly states that loss of Fab 3 would require "significant amounts to acquire process equipment tools to increase capacity" at other fabs. The Fab 3 lease was extended to 2030, but the landlord's non-curable breach claim could theoretically void it. Separately, TPSCo/NTCJ commercial agreements expire March 2027 (see Key Risks). This risk is in the 20-F risk factors but absent from sell-side analysis.

3. Customer prepayment narrative contradicts the balance sheet — material credibility risk. Management states "70%+ of SiPho capacity reserved, firmly backed with customer prepayment." The balance sheet shows deferred revenue + customer advances of $27.5M — declining from $29.3M the prior year. If 70%+ of expanded capacity (implying $600M+ of the $920M investment) is cash-prepaid, the balance sheet should show it. It doesn't. The commitments are likely contractual reservation agreements (take-or-pay), not cash deposits. CEO's language is also revealing: "presently reserved or in the process of being reserved" — the second clause is not "reserved." The entire "de-risked capex" narrative hangs on this claim, and the audited financials don't support the strongest reading. This doesn't mean demand is fake — counterparty evidence unanimously confirms demand is real. But the degree to which capex risk is customer-shared may be overstated.

4. CPO timing is later than Tower implies. Broadcom (the most influential voice in data center architecture) is actively pushing against CPO adoption. Hock Tan: "Don't need to go run into bright shiny objects CPO." Broadcom's 200G SerDes extending copper's reach for scale-up, with 400G by 2028. Tower's "substantially incremental" CPO opportunity is more likely 2029-2030 than 2027-2028.

5. Platform extension is unpriced optionality. Five partnership announcements in two months — CPO, optical circuit switches, quantum computing, LiDAR, plus AR display entry and wafer-to-wafer bonding. None of this is in any model. Zero-cost options on non-transceiver SiPho applications.

6. GFS at 17.6x vs TSEM at 26.4x — 50% premium. Same duopoly, similar SiPho revenue scale. TSEM earned the premium through execution, but GFS has 20.7% short interest and is priced for skepticism. If the thesis is "SiPho demand broadly," GFS may offer better risk-adjusted entry.

D&A Compression Window

A near-term margin risk that's not in consensus models: as $920M of capex deploys through 2026, depreciation jumps ≈$61M annually before the full revenue ramp. This creates a 2-3 quarter window in mid-2026 where reported margins may compress even as the underlying mix shift continues. Revenue needs to grow faster than depreciation for margin expansion to show in GAAP results.

Key Risks

Tier 1 — Material, Non-Consensus

1. Fab 3 (Newport Beach) lease dispute. Landlord seeking judicial declaration of "material non-curable breach" at Tower's primary SiPho production facility. The lease was extended to 2030 in November 2025 with a $105M upfront prepayment — the landlord accepted this payment, suggesting the relationship is transactional rather than irreconcilable. However, the noise abatement claim and a separate third-party claim create ongoing litigation risk. If the landlord prevails on the non-curable breach claim, the lease extension could theoretically be voided. Tower has runway (4+ years to 2030) and is replicating SiPho flows at Fabs 2, 7, and 9 as contingency. Material litigation risk, not existential — but this is the primary SiPho production facility.

2. March 2027 TPSCo/NTCJ nexus. TPSCo commercial agreements with NTCJ (49% minority partner) expire March 2027 — a separate timeline from Fab 3 (which now expires 2030). NTCJ is Tower's largest named customer at 13% of revenue. TPSCo is loss-making: NCI losses more than doubled from $0.64M (FY2024) to $1.66M (FY2025) despite Fab 7 running above 85% utilization. NTCJ's 49% stake is increasingly underwater (NCI balance deeply negative). At renewal, NTCJ has leverage to push for better pricing or reduced commitment, threatening both the Japan capacity and the 13% revenue relationship. Fab 7 is also absorbing Fab 11X redirected customers — if these flows carry sub-optimal margins, TPSCo losses could accelerate.

3. Single-segment opacity. All SiPho revenue and margin claims are from earnings call commentary, not audited financial statements. Tower reports as a single segment. No independent verification of SiPho mix, margins, or growth trajectory from SEC filings.

Tier 2 — Known but Magnitude Underappreciated

4. GFS competitive ramp. GFS expects ≈$400M in SiPho revenue for FY2026 (approaching Tower's $380M Q4 run rate), with 300mm wafer scale advantages and CPO design wins. The SiPho duopoly may tilt toward GFS post-2028 when both have expanded capacity and demand growth potentially slows.

5. Pillar 2 tax transition. FY2025 effective rate of 8.8% (Q4 at 2%) stepping up to ≥15%. Creates ugly YoY comparisons throughout 2026. The 2028 model incorporates 15%, but investors focused on quarterly EPS progression may react negatively.

6. Capex execution. $920M of equipment — tools shipped in pieces, installed, qualified, ramped. CEO hedged on Q4 call: "whether it's December or January, maybe February." He mentioned "lemon tools" — equipment that doesn't work on installation. Any delay pushes revenue contribution into 2027-2028, compressing the path to the model.

Tier 3 — Known, Priced

7. AI capex cyclicality. SiPho demand is directly tied to hyperscaler data center buildout. If AI investment pauses or redistributes toward inference (lower optical bandwidth intensity), transceiver demand decelerates.

8. Intel Fab 11X. Up to $300M in equipment at risk at Intel's New Mexico facility. Intel has "expressed its intention not to perform." Mediation underway. Already excluded from the 2028 model — resolution is upside, loss is already provisioned structurally.

9. Geopolitical. Israel conflict risk (Fabs 2 and 5), Japan yen volatility (Fab 7), US-China export controls (Chinese optical module customers represent growing share of revenue). The 20-F contains unusually detailed war risk disclosure including potential boycotts, force majeure claims, credit downgrades, and Sabbath operating exemption risk.

10. Key man. CEO Ellwanger IS the company's face to customers and investors. No disclosed succession plan. All executive team members have 10-20+ year tenure — stability is a strength until it becomes a cliff.

What to Watch

Near-Term (Next 90 Days)

• Q1 2026 earnings (May 13): First quarter under Pillar 2 taxation. Consensus at $0.57 EPS — 27% below Q4's $0.78. The REAL sequential comparison (tax-adjusted) is more like -10%. Watch for revenue trajectory vs $412M guidance and any update on tool installation timing.

• Fab 3 lease proceedings: Any court filings or settlement news on the Newport Beach noise abatement dispute. This is the single most important non-earnings data point.

• GFS Q1 2026 earnings (May 5): GFS SiPho revenue trajectory. If GFS reports acceleration toward the ≈$400M annual target, competitive dynamic intensifies.

Medium-Term (6-12 Months)

• Tool installation completion (target: Dec 2026): Management committed to having all $920M of equipment installed and qualified by year-end 2026. Any slippage is visible in quarterly utilization data and margin trajectory.

• Customer diversification: Whether the "other customers" bucket (6 at 4-7%) continues to grow, reducing concentration risk.

• 3.2T material system definition: Tower is "playing a key role, partnering with lead customers to define the material systems." Whoever defines the 3.2T platform has structural advantage for the 2028-2030 cycle.

Long-Term (12-24 Months)

• March 2027 TPSCo/NTCJ renewal: Commercial agreements with 49% minority partner expire. Separate from Fab 3 (lease extended to 2030, but litigation ongoing).

• CPO commercialization: Marvell targeting CPO scale-up revenue from Celestial AI acquisition starting FY2028. Broadcom resisting. The Broadcom-vs-Marvell architecture battle determines CPO timing — and Tower's incremental TAM.

• 300mm SiPho transition: Tower released 300mm SiPho design kits. Qualification at Fab 7 (Japan) or Fab 10 (Italy) would dramatically expand output capacity and close the wafer-size gap with GFS.

• Post-capex capital allocation: If the 2028 model hits, Tower generates ≈$800M annual FCF with no disclosed plans for dividends, buybacks, or M&A. What management does with that cash will define the next chapter.

Evidence