Tower Semiconductor: Silicon Photonics Inflection, Buy The Drawdown

Tower Semiconductor is now generating nearly half its revenue from AI optical content and reported record Q2 2026 revenue of $460M. The company delivered 55% incremental operating leverage alongside a mix shift to higher-margin 300mm SiPho and SiGe wafers, and raised its 2028 outlook to $3.6B revenue and $1.2B net profit. Guidance is supported by strong customer prepayments and fully funded capacity expansions, which should be supportive for the stock over the next few quarters.

Analysis

The market is likely underestimating how much of Tower’s AI exposure is now de-risked by customer-funded capacity and how much operating leverage can still surface if utilization stays high. That combination matters because it turns a cyclical foundry story into something closer to a scarce infrastructure supplier: better cash conversion, less balance-sheet strain, and a higher probability that each incremental wafer contributes disproportionately to earnings rather than just revenue.

Second-order winners are the broader optical interconnect stack — COHR, LITE, MRVL, and ANET — because a faster silicon-photonics ramp should lower cost-per-bit and accelerate port/content growth in data centers. The relative losers are more commoditized foundry names and legacy interconnect/copper-heavy exposures such as GFS and UMC, which could start to look structurally slower if Tower keeps taking the highest-growth niche and investors re-rate the group by AI content rather than by end-market beta.

The key risk is duration, not direction: silicon-photonics demand can be lumpy, and hyperscaler capex pauses can hit bookings before the secular thesis breaks. What would falsify the setup is any slip in the capacity ramp, margin expansion failing to follow mix, or a material haircut to the long-range model over the next 1-2 quarters. Near term, chasing the stock after strength is lower quality than owning pullbacks; over 6-18 months the bigger question is whether the market starts valuing TSEM like a strategic AI supplier rather than a normal foundry multiple.

Contrarian view: consensus may be extrapolating a hot sub-cycle too far too fast. Customer prepayments and fully funded expansion reduce risk, but they do not guarantee sustained end-demand; if AI networking spend normalizes, the multiple can compress quickly even if revenue still grows. That makes the stock attractive only if investors can separate temporary order visibility from durable share gains in silicon photonics.

Key Decisions for Investors

• Accumulate TSEM on a 3-7% post-print pullback, with a 6-18 month horizon; thesis breaks if next-quarter margin expansion stalls or management trims the long-range model.
• Pair trade: long TSEM / short GFS as a relative-value expression of AI-photonics scarcity versus broader foundry exposure; expect outperformance if optical content remains tight over the next 1-2 quarters.
• Use an alert, not a chase, for COHR and LITE: buy dips only if hyperscaler optical budgets are reaffirmed, since they are the cleaner second-order beneficiaries of a continued silicon-photonics ramp.
• If options liquidity is workable, favor a call spread over outright stock chase for TSEM to define downside while keeping exposure to a rerating if the market starts valuing the business as strategic AI infrastructure.