$GFS

Summary

The analyst is bullish on TSMC (TSEM) due to its dominant position in silicon photonics, strong revenue growth, pre-sold capacity, and unique full-stack manufacturing capabilities, expecting significant earnings growth by 2028 despite some risks.

Reasoning

TSMC's dominant position in silicon photonics with rapid revenue growth, supplying key components to Nvidia, unique full-stack analog/photonic manufacturing capabilities leading to high switching costs, pre-sold capacity through 2028, and a strong financial model with expected 50%+ earnings CAGR. Risks include patent infringement lawsuits, declining RF mobile revenue, ambitious capex, and dependency on government subsidies.

$TSEM is one of my highest conviction trades. They've built a dominant position in SiPho. (The tech that moves data between AI chips using light instead of copper) SiPho revenue went from $28M in 2023 to $238M in 2025, with a $380M annualized run rate in Q4 alone. ...Huge structural inflection. $TSEM are supplying SiPho components for 1.6 terabit optical modules to $NVDA As AI cluster complexity scales, GPU-to-GPU bandwidth becomes the hard constraint, and $TSEM sits at that exact chokepoint. No pluggable transceiver alternative at this data rate matches the cost and efficiency of silicon photonics. $TSEM and $NVDA know this, and 8 of the 10 largest optical transceiver manufacturers have already voted with their design wins. But I think the market are underpricing some things. $TSEM is the only foundry simultaneously providing silicon photonics PICs, SiGe driver electronics, and power management ICs from a single manufacturing partner. That full-stack analog/photonic capability means customers designing 1.6T and 3.2T modules consolidate their supply chain into one foundry. Which deepens switching costs, expands wallet share, and makes $TSEM harder to displace. We also have huge demand validation signals. $920M SiPho/SiGe capacity expansion, with over 70% of new capacity already reserved / being reserved through 2028 with customer prepayments. Customers are literally pre-funding $TSEM's expansion because they cannot afford to lose allocation. The 2028 model targets $2.84B revenue at 39% gross margins and $750M net profit. A 50%+ earnings CAGR from a company with a $1.2B cash balance and no net debt. The power delivery angle adds another AI vector. $TSEM's new Gen3 LDMOS BCD platform addresses a TAM growing from $2.5B today to $4.7B by 2031. And the company has already secured 3 of the top 4 RF front-end module providers as customers with production underway. On valuation, trailing P/E above 100x looks rich in isolation, but the 2028 forward P/E compresses to approximately 29x on the financial model. Not unreasonable for a company with locked-in hyperscaler relationships, pre-sold capacity, and no credible full-stack competitor. The risks are real and deserve disclosure though. $GFS filed an 11-patent infringement suit in March 2026 at the US ITC and Western District of Texas. $GFS holds 8,000+ patents versus $TSEM's ~500. RF Mobile revenue is declining on Chinese domestic sourcing pressure. The $920M capex plan is ambitious and execution-dependent. And the Fab 7 4x expansion is contingent on Japanese government subsidy approval. The structural bull case survives all of that imo. AI cluster scaling requires exponentially more optical bandwidth. Silicon photonics is winning at 1.6T and beyond. 70%+ of $TSEM's new capacity is pre-sold, and no other open foundry replicates the full-stack capability. The question, as ever, is whether $TSEM can execute the ramp and justify a valuation that prices in a near-flawless outcome.

Return by horizon