[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]  Tech Fund [@techfund1](/creator/twitter/techfund1) on x 11.2K followers Created: 2025-07-06 17:22:16 UTC Needham (Charles Shi) - some detailed modelling on $TSM and semicap: X. Concludes that $XX billion in AI revenues for TSMC '29 looks conservative X. Sees strong semicap growth in '27 for TSMC at +47%. "Using TSMC's publicly stated AI revenue target (~$90B by 2029) and AI accelerator outlook (~50MM units by 2030), we developed our AI wafer demand model and found that TSMC doesn't need much unit volume growth (counted in package) to reach the target; significant silicon content growth from increasing number of compute dies in a package and the transition to custom HBM base dies should sustain the rapid AI revenue growth to ~$90B within four years. We see potential deceleration of TSMC AI revenue into 2026, due to AI accelerator unit volume decel and the lack of silicon content growth in Nvidia Rubin. Still, we forecast TSMC's AI revenue will at least grow at a decent XX% YoY next year. Looking ahead, we see significant growth acceleration to nearly XX% YoY in '27 and XX% YoY in '28, as silicon content growth resumes with Rubin Ultra and Feynman. We believe the strong AI revenue growth should support TSMC's total revenue to reach $130B in '26 and $150B in '27. We see Nvidia's AI GPUs alone should require a peak capacity of 60kWSPM on 5nm in 2026, 80kWSPM on 3nm in 2028, and 100kWSPM on 2nm in 2030. We see strong AI growth should fully digest TSMC's existing 5nm capacity by 2026, but TSMC may have to raise 3nm and 2nm capacity investment in 2027 to satisfy Nvidia demand. We forecast TSMC CapEx should grow from $40B in 2025 to $45B in 2026 and $50B in 2027. While CapEx growth looks moderate, we foresee a significant mix shift away from infrastructure and into equipment, which may lead to a record $29B WFE spending by TSMC in 2027. We see incremental CoWoS capacity (ex-OSAT) additions to decline in '26 and remain muted in '27, before 9.5x reticle CoWoS fully ramps up to support Rubin Ultra in '28. We forecast HBM packaging CapEx (ex-China), which is declining in '25, could be light in '26 and see a strong recovery in '27." XXXXX engagements  **Related Topics** [$90b](/topic/$90b) [coins ai](/topic/coins-ai) [$tsm](/topic/$tsm) [taiwan semiconductor](/topic/taiwan-semiconductor) [stocks technology](/topic/stocks-technology) [Post Link](https://x.com/techfund1/status/1941910581407043613)
[GUEST ACCESS MODE: Data is scrambled or limited to provide examples. Make requests using your API key to unlock full data. Check https://lunarcrush.ai/auth for authentication information.]
Tech Fund @techfund1 on x 11.2K followers
Created: 2025-07-06 17:22:16 UTC
Needham (Charles Shi) - some detailed modelling on $TSM and semicap: X. Concludes that $XX billion in AI revenues for TSMC '29 looks conservative X. Sees strong semicap growth in '27 for TSMC at +47%.
"Using TSMC's publicly stated AI revenue target ($90B by 2029) and AI accelerator outlook (50MM units by 2030), we developed our AI wafer demand model and found that TSMC doesn't need much unit volume growth (counted in package) to reach the target; significant silicon content growth from increasing number of compute dies in a package and the transition to custom HBM base dies should sustain the rapid AI revenue growth to ~$90B within four years.
We see potential deceleration of TSMC AI revenue into 2026, due to AI accelerator unit volume decel and the lack of silicon content growth in Nvidia Rubin. Still, we forecast TSMC's AI revenue will at least grow at a decent XX% YoY next year. Looking ahead, we see significant growth acceleration to nearly XX% YoY in '27 and XX% YoY in '28, as silicon content growth resumes with Rubin Ultra and Feynman. We believe the strong AI revenue growth should support TSMC's total revenue to reach $130B in '26 and $150B in '27.
We see Nvidia's AI GPUs alone should require a peak capacity of 60kWSPM on 5nm in 2026, 80kWSPM on 3nm in 2028, and 100kWSPM on 2nm in 2030. We see strong AI growth should fully digest TSMC's existing 5nm capacity by 2026, but TSMC may have to raise 3nm and 2nm capacity investment in 2027 to satisfy Nvidia demand. We forecast TSMC CapEx should grow from $40B in 2025 to $45B in 2026 and $50B in 2027. While CapEx growth looks moderate, we foresee a significant mix shift away from infrastructure and into equipment, which may lead to a record $29B WFE spending by TSMC in 2027.
We see incremental CoWoS capacity (ex-OSAT) additions to decline in '26 and remain muted in '27, before 9.5x reticle CoWoS fully ramps up to support Rubin Ultra in '28. We forecast HBM packaging CapEx (ex-China), which is declining in '25, could be light in '26 and see a strong recovery in '27."
XXXXX engagements
Related Topics $90b coins ai $tsm taiwan semiconductor stocks technology