TSMC price target raised at Citi as ’AI megacycle accelerates’ By Investing.com

Citi raised its TSMC price target to NT$2,800 (from NT$2,600), implying a 53.8% upside from the NT$1,820 close and reiterated a Buy. Citi's 2027/2028 EPS forecasts of NT$130.72/NT$165.99 are 18%/28% above Bloomberg consensus, with net profit projected to rise from NT$1.72T in 2025 to NT$3.39T in 2027 and NT$4.30T in 2028 (growth of ~41.5%, 39.5%, 27% respectively) and revenue beating consensus by ~2.2%, 12% and 17.7% in 2026-28. Citi also expects gross margins to expand to ~63.7%-65.6% by 2026-28, raises capex to ~US$67B in 2027 and ≥US$75B in 2028, forecasts N2 becoming the largest revenue contributor (29% of 2027 revenue), and sees AI-related revenue doubling year-on-year into 2027; near-term Middle East energy risks are deemed manageable.

Analysis

The AI-driven demand story is becoming a capital-allocation event as much as a product cycle: fabs and packaging capacity, not just wafer orders, will determine winners. That elevates equipment, advanced packaging and optical-interconnect suppliers while commoditized, older-node fabs and standalone memory vendors face margin compression as demand concentrates on a handful of leading-edge suppliers. Geopolitical friction around energy and shipping routes is a non-linear amplifier — short-term shocks can force deferment of ramp schedules and raise unit costs materially, while longer-term diversification of fabs increases industry structural capital intensity.

Key near-term catalysts to watch are operational cadence metrics rather than high-level guidance: wafer-starts at advanced nodes, CoWoS/Co-packaged optics throughput, tool lead times and foundry yield curves — these will reveal whether demand is translating into sustainable revenue per wafer. Time horizons differ: days/weeks for geopolitics-driven logistics shocks; quarters for customer order patterns and yield improvement; 18–36 months to validate multi-node capacity growth and return on elevated capex. Tail risks include model-efficiency breakthroughs or tighter macro capex that would remove the need for the projected capacity expansion and compress realized ASPs for bleeding-edge nodes.

Consensus may be underestimating two things: the execution risk of near-term node and packaging scale-ups, and the working-capital drag from front-loaded capex. Market multiples currently price a seamless demand-to-capex conversion; any slippage in yield or customer design wins will compress free cash flow and rerate the group quickly. The most actionable signals will be three consecutive months of accelerating CoWoS wafer starts, a sustained decline in tool lead times, or major cloud customers issuing multi-year supply commitments — each would validate the bullish thesis or, if absent, argue for rapid derisking.

Key Decisions for Investors

• TSM: Establish a core long (3–5% NAV) via a Jan-2029 call spread (buy LEAP ITM call, sell higher-strike call) to capture multi-year node/packaging upside while capping premium. Add on >10% pullbacks; trim into sequential positive operational prints (wafer-starts, packaging throughput). Expected R/R: 40–60% upside by end-2028 vs ~20–25% downside if AI capex stalls; stop-loss on a 20% realized drawdown.
• NVDA: Tactical long (1.5–2% NAV) in Jan-2027 call spreads to play first-adopter GPU demand with defined risk. Use this as a leveraged play on model and data-center accelerators; take profits on 2x move or on signs of customer order rephasing. Premium at risk; target asymmetric payoff if Feynman-class GPUs see rapid adoption.
• GOOGL/GOOG: Buy Jan-2026 calls (2–3% NAV) as a lower-volatility way to capture secular migration of AI chip demand to large cloud providers and their in-house ASIC strategies. Acts as a hedge to a scenario where nodes ramp but customer concentration shifts value to cloud integrators; expect modest upside (30–50%) with lower tail volatility than pure semiconductor longs.