14 March 2026

Tower Semiconductor (TSEM): The Indispensable Foundry at the Heart of the AI Optical Revolution

Nasdaq: TSEM  |  Price: US$118.54  |  Market Cap: US$13.4B  |  Rating: ACCUMULATE  |  Base Case Target: US$138  |  +16% upside

Section I — Executive Summary and Investment Thesis

Tower Semiconductor occupies a rare and structurally advantaged position in the global semiconductor supply chain: it is, by wide industry acknowledgement, the most capable open-access specialty foundry for silicon photonics (SiPho) at commercial scale. At a moment when the world’s largest hyperscalers are racing to deploy 1.6 Terabit-per-second optical interconnects across AI data centres — and when co-packaged optics (CPO) are emerging as the bandwidth solution for next-generation GPU clusters — Tower is the foundry upon which this infrastructure ultimately depends.

The market has begun to recognise this, with the stock rallying sharply since the company’s NVIDIA collaboration announcement in February 2026, but we believe the full extent of Tower’s strategic indispensability remains underpriced, particularly in the context of its 2028 financial targets. We initiate with an Accumulate recommendation and a base-case price target of US$138, implying 16% upside from the current level of US$118.54.

Thesis Snapshot

Tower Semiconductor is the specialist foundry that treats silicon photonics as a core business rather than a research sideline, giving it a process maturity and capacity dedication that general-purpose foundries cannot match. As the AI infrastructure stack migrates toward optical interconnects at 800G and 1.6T speeds, and ultimately toward co-packaged optics directly inside GPU packages, Tower stands as the critical manufacturing choke point — a picks-and-shovels play on the bandwidth arms race.

Key Financial Metrics (as at Q4 FY2025)

* FY2025 OCF of US$395M includes US$105M Fab 3 lease extension prepayment outflow; adjusted OCF approximately US$500M.

Valuation Summary

Current Price: US$118.54. The market is pricing Tower close to — but marginally below — our Base Case, suggesting the optical ramp is partially but not fully discounted.

Section II — Business Model Analysis

Tower Semiconductor is the world’s leading independent specialty foundry — a company that manufactures semiconductors to third-party customers’ designs rather than selling its own branded chips. It operates eight wafer fabrication facilities across Israel, the United States, Japan, and Italy, producing chips across a highly differentiated set of specialty process technologies.

Revenue Model and Segment Economics

Tower generates revenue on a per-wafer basis. The company serves over 300 customers across six primary end markets: consumer, mobile, power management, industrial, medical, and — increasingly and critically — AI data centre infrastructure. The business model is a long-cycle industrial one: customers qualify Tower’s process platforms over 12–24 months before volume production, creating high switching costs and multi-year revenue visibility.

The revenue mix has shifted dramatically in 2025. RF Infrastructure — dominated by silicon photonics for optical transceivers and interconnects — grew to become Tower’s largest segment at 27% of FY2025 revenue (US$423 million), up from 17% in 2024, with 75% year-over-year growth. Silicon photonics revenue specifically reached US$228 million in 2025, more than doubling from US$106 million in 2024. SiPho carries materially higher margins than legacy RF Mobile or core CMOS, providing the engine for the management’s targeted gross margin expansion from 23% in 2025 to 39% by 2028.

Margin Structure and Operating Leverage

Tower’s current margin profile understates its destination. FY2025 gross margins of 23.2% reflect a period of heavy capital investment and mixed revenue composition — Q4 2025 alone saw gross margins of approximately 27%. Management’s 2028 target of 39.4% gross margin is aggressive but credible: it is driven by SiPho mix-shift, utilisation improvement, and the operational leverage inherent in a wafer fabrication business where incremental revenue contribution margins are high once fixed costs are absorbed.

Customer Relationships, Switching Costs, and Pricing Power

Tower’s customer relationships are structurally sticky. The chip design-to-production qualification cycle takes 18–24 months and involves significant IP co-development. Over 70% of its total SiPho capacity is either currently reserved or in active reservation discussions through 2028, with customer-paid prepayments securing this capacity. This is an extraordinary level of demand visibility.

The NVIDIA partnership, announced in February 2026, is strategically significant not merely as a revenue contract but as a technology co-development signal — NVIDIA’s validation of Tower’s SiPho platform de-risks customer hesitation and accelerates design-in activity across the ecosystem.

Capital Intensity and Balance Sheet

The company is committing US$920 million to SiPho and SiGe equipment with the goal of achieving SiPho wafer capacity exceeding five times the Q4 2025 run-rate by Q4 2026. The balance sheet is healthy: US$1.15 billion in cash and equivalents provides ample coverage for the capex programme. Tower can fund the expansion largely without external financing.

Section III — Strategic Trajectory and Competitive Moat

Growth Vectors

Silicon Photonics for AI Infrastructure (Primary Driver). The bandwidth arithmetic is inexorable: training and inference clusters are scaling to 100,000+ GPU configurations; each GPU-to-GPU link requires bandwidth that copper cannot sustain beyond relatively short distances. Tower’s SiPho platform currently serves 800G transceiver production and is actively engaged in 1.6T module development with NVIDIA. The 400G-per-lane modulator demonstrated on Tower’s PH18DA platform in early 2026 signals that 3.2T capability is already in the technology roadmap. The addressable market for SiPho-enabled transceivers and CPO modules in AI data centres is estimated to reach US$8–12 billion by 2028.

Co-Packaged Optics (Emerging but Potentially Transformational). CPO integrates the optical engine directly onto the switch or GPU package, dramatically reducing energy consumption per bit. Tower announced CPO foundry technology availability in November 2025. Meaningful volumes likely 2027–2029.

Quantum Photonics (Long-Term Option). Tower’s February 2026 collaboration expansion with Xanadu extends the SiPho platform into photonic quantum hardware fabrication. Revenue contribution is minimal in the near term.

Non-AI Specialty Markets (Steady-State). Power management, CMOS image sensors, and industrial/medical analog remain steady contributors and provide operating leverage and cross-utilisation of fab capacity.

Competitive Positioning

Tower’s competitive moat in silicon photonics rests on three reinforcing pillars:

• Process Maturity. Tower has been developing its SiPho process since 2013 — earlier than most competitors. The resulting PDK maturity is the highest available in the open foundry market.
• Open Foundry Model. Tower operates as a pure-play foundry — it does not design or sell its own photonic chips, removing any competitive conflict with customers.
• Ecosystem Density. Over 300 customers across the optical interconnect ecosystem have qualified designs on Tower’s platforms. Partnerships with Lumentum, Scintil Photonics, Xanadu, and NVIDIA compound this advantage.

Competitive Threats. TSMC’s SiPho engagement is accelerating — its N3-era platform development is a legitimate medium-term threat. This is the primary competitive risk: TSMC exercising its foundry scale to commoditise SiPho at the expense of specialty players.

Special Situation: Intel Fab 11X Dispute

Intel has expressed its intention not to perform on the Fab 11X agreement and the matter is in mediation. Tower’s revised 2028 financial model explicitly excludes the Intel Fab 11X contribution and still targets US$2.84 billion in revenue. We treat the Intel situation as a neutral factor in our Base Case.

Section IV — Risk Assessment

Geopolitical and Country Risk

Tower’s headquarters and primary wafer fabrication facility are located in Israel. The ongoing security environment remains a persistent discount factor. A material escalation involving direct strikes on Israeli industrial infrastructure would be a significant production risk. This risk is partially mitigated by Tower’s geographically diversified fab network (US, Japan, Italy).

Customer Concentration and Demand Visibility

The silicon photonics business is relatively concentrated among a small number of large optical transceiver manufacturers. A hyperscaler capex digestion period — not our base case but a plausible scenario — could defer SiPho volume. The 70%+ reserved capacity through 2028 mitigates this substantially.

Execution Risk on US$920M Capex

The 5x capacity target by Q4 2026 is ambitious; slippage of six to twelve months would delay the financial leverage assumed in our 2027–2028 estimates. Equipment supply chains for specialty photonic process equipment are tighter than for standard CMOS.

TSMC Competitive Entry

The most structurally threatening competitive scenario is TSMC committing commercial-scale SiPho capacity. This is not imminent, but a 3–5 year horizon risk. Tower’s process maturity lead is meaningful but not permanent.

Valuation Risk

At US$118.54, the stock trades at approximately 61x trailing earnings and 44x forward earnings. The investment case is binary: if the SiPho ramp proceeds broadly as guided, Tower’s 2028 earnings power makes the current price a compelling entry. If execution disappoints, the valuation provides limited margin of safety. Our Bear Case implies US$83 per share — a 30% drawdown.

Section V — Multi-Scenario DCF Valuation

Scenario Narratives

Bear Case — US$83/share (–30% downside): SiPho capacity ramp encounters meaningful delays. Intel Fab 11X mediation produces no recovery. Competition from GlobalFoundries accelerates. Tower achieves approximately US$2.0 billion in revenue by 2028 with gross margins reaching only 30%.

Base Case — US$138/share (+16% upside): Tower executes the SiPho capacity expansion broadly on schedule. Revenue reaches approximately US$2.55 billion by 2028, roughly 90% of management’s target. Gross margins expand to 35–36%. The NVIDIA partnership expands from collaborative development into volume production.

Bull Case — US$193/share (+63% upside): SiPho ramp exceeds the 5x target. NVIDIA collaboration expands into CPO production volumes by mid-2027. Tower’s 2028 revenue target of US$2.84 billion is achieved approximately on schedule with gross margins reaching the targeted 39%. Xanadu quantum collaboration begins contributing commercial revenue.

Valuation Output

Comparable Company Analysis

Section VI — Investment Conclusion

Tower Semiconductor is a genuinely differentiated semiconductor business at an inflection point. The thesis is not complicated: the AI data centre buildout requires optical interconnects; optical interconnects require silicon photonics chips; silicon photonics chips require a foundry with the process maturity, platform openness, and capacity dedication to produce them at commercial scale; and Tower is that foundry.

The financial trajectory supports conviction. Tower’s 2028 targets — US$2.84 billion in revenue, 39.4% gross margin, US$750 million in net profit — represent an earnings base from which the current US$13.4 billion market capitalisation looks inexpensive. At our Base Case execution of approximately 90% of those targets, Tower trades at 17–18x 2028 earnings today — a multiple we consider undemanding for a business with this level of structural competitive advantage.

The risks are real: Israel geopolitical exposure, the Intel Fab 11X overhang, a US$920 million capex programme that must execute with precision, and the long-duration threat of TSMC entering SiPho at scale. But the risk/reward at the current entry point — 16% to Base, 63% to Bull, against 30% downside to Bear — is an attractive asymmetry for a quality business in a secular growth market.

Recommendation: Accumulate. Base Case price target of US$138. Investment time horizon: 18–24 months. Catalysts to watch: Q1 2026 earnings for SiPho revenue run-rate; Q2 2026 capacity utilisation disclosures; any CPO volume production announcements with NVIDIA; and resolution of the Intel Fab 11X mediation.

Summary

Tower Semiconductor is the world’s leading specialty chip foundry for silicon photonics — the technology that moves data using light rather than electricity inside AI data centres. As artificial intelligence drives an unprecedented expansion of computing infrastructure, the bandwidth connecting all those AI processors is becoming the critical bottleneck, and Tower makes the chips that solve it. The company has secured demand commitments covering over 70% of its expanded capacity through 2028, is actively collaborating with NVIDIA on next-generation optical modules, and is targeting revenues of US$2.84 billion and net profits of US$750 million by 2028 — versus US$1.57 billion and US$220 million in 2025. At the current price of US$118.54, Tower trades at roughly 18x its 2028 earnings target, which we consider undemanding for a business with this structural position. Our base case target of US$138 implies 16% upside over an 18–24 month horizon, with a bull case of US$193 if the ramp exceeds expectations. The primary risks are geopolitical (Israel) and execution on the US$920 million expansion programme.