JPMorgan maintains TSMC at Overweight with a NT$2400 target, with the key focus on continued upward revisions to CoWoS and SoIC advanced packaging demand

Overview

This is a JPMorgan company research report on TSMC, focused on updates to advanced back-end packaging such as CoWoS, SoIC, CoPoS, and CoWoP. The report argues that AI compute demand continues to tighten supply of TSMC's leading-edge nodes and advanced packaging, with N3/N5 utilization above 100% and N3/N2 constraints likely to persist through 2027. JPMorgan maintains an Overweight rating on TSMC and a December 2026 target price of NT$2400.

Core views

The core views are: first, TSMC's CoWoS and SoIC capacity expectations are being revised upward, with AI ASICs, GPUs, CPO, and silicon photonics forming the medium-term demand increment. Second, advanced packaging demand from customers such as NVDA, Broadcom, AWS/Alchip, and AMD remains strong, but the 2026-2027 product cycles and bottlenecks diverge. Third, CoPoS and CoWoP represent important technical evolution for large AI packaging after 2028, but in the near term they are still constrained by engineering, yield, and supply-chain maturity. Fourth, TSMC benefits from tight advanced nodes, premium pricing for expedited wafers, cross-fab optimization, and a weaker New Taiwan dollar; the report expects FY26/FY27 U.S.-dollar revenue growth of about 35%/30%, and sees potential for 1Q26/2Q26 gross margin to beat expectations at around 67%.

Analysis framework

The report mainly uses supply-chain checks, customer project breakdowns, advanced packaging capacity estimates, AI accelerator shipment forecasts, and valuation-multiple analysis. On the demand side, it breaks out projects such as NVDA, Broadcom TPU, MediaTek TPU, AWS Trainium, AMD MI450/Venice, Meta MTIA, and MAIA; on the supply side, it assesses TSMC CoWoS/SoIC, OSAT outsourcing, Intel EMIB, ASE/Amkor capacity, and CoPoS trial-production progress; on valuation, it derives the target price using about 20x 12-month forward P/E.

Methodology notes

• Valuation methods12-month forward P/E

  Using about 20x 12-month forward P/E as the basis for TSMC's target price.

  JPMorgan's NT$2400 target price is based on about 20x 12-month forward P/E, above TSMC's five-year historical average multiple, reflecting AI demand and fundamental momentum.

• supply-demand_estimationCoWoS/SoIC capacity and customer allocation

  Estimating advanced packaging tightness using wfpm and customer project demand.

  The report assesses the CoWoS and SoIC supply-demand gap based on TSMC versus non-TSMC capacity, demand from customers such as NVDA, Broadcom, AMD, and AWS, and the possibility of OSAT outsourcing.

• technology_roadmapAdvanced back-end packaging evolution

  Comparing the adoption timing and risks of CoWoS, SoIC, CoPoS, CoWoP, and EMIB packaging solutions.

  The report believes CoPoS may be used for larger AI accelerator packages after 2028, CoWoP is still in parallel development, and EMIB is more likely to see large-scale AI ASIC adoption only around 2028.

Asset mapping & comparison

Structured mapping from thesis to named assets (strengths, weaknesses, peers, risks).

• TSMC / 2330.TW

  Report subject and core beneficiary asset

  Strengths

  Tight supply in leading-edge nodes and advanced packaging, upward revisions to CoWoS/SoIC capacity, AI demand driving FY26/FY27 revenue growth, and potential gross margin upside.

  Weaknesses

  Client demand from PCs and Android remains weak, and part of future growth depends on the AI capex cycle and advanced packaging mass-production progress.

  Comparison

  Compared with OSATs and other packaging routes, TSMC still holds the key integration capability for CoWoS, SoIC, and leading-edge nodes.

  Risks

  A slowdown in the AI capex cycle, weak PC/smartphone demand, CoPoS/CoWoP engineering challenges, and delays in customer products.

• NVDA

  Key CoWoS demand customer

  Strengths

  Blackwell shipments were raised, 2027 CoWoS demand was revised upward, and Rubin/Rubin Ultra and Feynman may continue to drive advanced packaging demand.

  Weaknesses

  Rubin ramps more slowly in 2026, constrained by HBM4 availability.

  Comparison

  Compared with other AI accelerator customers, NVDA remains one of the largest drivers of CoWoS demand.

  Risks

  Insufficient supply of bottleneck components such as HBM4 and N3P wafers, and uncertainty over the maturity of Feynman adopting A16, 3D SoIC, or CoPoS.

• Broadcom / TPU supply chain

  Source of upside revisions to AI ASIC and CoWoS demand

  Strengths

  TPU v7/v8 demand is rising sharply, with 2026/2027 CoWoS forecasts raised to 250K/400K.

  Weaknesses

  TPU v9 faces competition across multiple routes, and the final high-volume solution remains unclear.

  Comparison

  The report views Broadcom's Pumafish route as lower execution risk than MediaTek's Humufish route.

  Risks

  High-speed SerDes, 3D SoIC design, packaging execution, and changes in customer demand.

• ASE / Amkor / OSAT

  Supplemental capacity for CoWoS outsourcing and advanced packaging

  Strengths

  Tight TSMC CoWoS supply is pushing OSATs to participate in projects such as AMD Venice, NVDA Vera CPU, Trn3, and TPU.

  Weaknesses

  Capabilities for high-RDL-layer CoWoS, qualification, and mass-production yield still need validation.

  Comparison

  OSATs can relieve part of the supply constraint, but in the near term they are unlikely to fully replace TSMC's core capability.

  Risks

  Qualification failures, slow mass-production ramp, and slower-than-expected customer migration.

Key data

• TSMC CoWoS capacity forecast115K wfpm by end-2026, 155K wfpm by end-2027, and 175K wfpm by end-2028Slight upward revision versus prior estimates.
• Non-TSMC CoWoS capacity forecastAbout 25K wfpm by end-2027 and about 35K wfpm by end-2028Expected to ramp slowly from OSAT and other suppliers.
• SoIC capacity forecast35K wfpm by end-2027 and 65K wfpm by end-2028Driven mainly by 2nm-class AI ASICs, silicon photonics/CPO, and future AI accelerator demand.
• NVDA 2026 CoWoS forecast675K wafers per yearDown 4% versus prior; Blackwell shipments raised to 5.8M, Rubin lowered to 2.2M, and total GPU shipments are expected at 8.3M, up 34% year over year.
• Broadcom CoWoS forecast250K in 2026 and 400K in 2027Mainly driven by upside in TPU v7/v8 demand; TPU shipments are expected to be 4.3M/6.9M in 2026/2027.
• MediaTek TPU-related CoWoS forecast18K in 2026 and 60K in 2027Corresponds to about 0.4M/1.5M units of TPU v8x Zebrafish; design changes could affect shipment timing.
• Trainium3 lifetime shipments4.4M units2026 CoWoS estimate cut by 13%, 2027 estimate raised by 19%, with demand from Anthropic and AWS internal workloads.
• AMD 2026 CoWoS forecast77KBelow the prior 90K estimate due to MI450 needing a 2nm top-die retape-out and facing HBM4 validation challenges; 2027 estimate unchanged.
• TSMC FY26/FY27 U.S.-dollar revenue growth forecastAbout 35%/about 30%Driven by N3 growth and strong advanced packaging demand.
• Target price and current priceTarget price NT$2400.0; current price NT$1950.0Target price implies about 23.1% upside.

Impact & implications

In terms of investment implications, the report reinforces TSMC's position as the core bottleneck asset in the AI compute supply chain. The upward revisions to CoWoS and SoIC suggest that advanced packaging is not just a short-term capacity shortage, but a medium-term growth driver deeply tied to GPU, AI ASIC, CPO, and silicon photonics roadmaps. If N3/N2 supply tightness persists, and expedited orders plus advanced packaging lift ASPs and gross margin, TSMC's earnings leverage and valuation premium could continue to be supported. However, customer product cycles, HBM4, N3P wafers, CoPoS engineering maturity, and weak PC/smartphone demand may still affect the realization path.

Risks

• There is debate over how long the AI capex growth cycle will last; if capex slows, demand for advanced packaging and leading-edge nodes will be affected.
• Weak PC and smartphone demand in the second half of 2026 could weigh on TSMC's fundamentals.
• Insufficient supply of HBM4, N3P wafers, and other bottleneck components may limit accelerator shipments from NVDA, AMD, and others.
• CoPoS is still in engineering and trial production; warpage, equipment compatibility, material selection, and yield issues could delay adoption into 2028.
• MediaTek's TPUv9 Humufish route involves multiple leading technologies, including high-speed SerDes, Intel EMIB, and 3D SoIC, and carries high execution risk.
• It remains uncertain whether OSAT CoWoS-R or equivalent capacity can pass customer qualification smoothly and scale production.

What to watch

• TSMC's actual CoWoS and SoIC expansion pace in 2026-2028.
• The advanced packaging configuration and supply allocation for NVDA Rubin, Rubin Ultra, Feynman, and Vera CPU.
• Demand realization for Broadcom TPU v7/v8 and the mass-production share of TPU v9 Pumafish and Humufish routes.
• AWS Trainium3's ramp after 2Q26 and whether some orders shift to ASE.
• The mass-production progress of AMD MI450, Venice CPU, and other 2.5D packaging products in 2026-2027.
• Progress on the CoPoS trial line and engineering fixes, and whether it is first adopted by products such as NVDA Feynman in 2028.
• CoWoP mass-production decisions after Rubin/Rubin Ultra small-batch trials from end-2026 to 1H27.
• TSMC N3/N5 utilization, expedited wafer demand, ASP premiums, and 1Q26/2Q26 gross margin performance.