Goldman Sachs Raises TSMC's 2028 Capex Forecast to $98 Billion, Extending AI Expansion Cycle Beyond 2032

Continuing expansion in AI demand is driving TSMC into a new cycle of larger-scale capital investment.

In its latest research report, Goldman Sachs significantly raised its capital expenditure forecasts for Taiwan Semiconductor for 2027 and 2028 to $85 billion and $98 billion, respectively, up markedly from the previous estimates of $78 billion and $82 billion. Meanwhile, Goldman Sachs expects that any potential new fab in Texas will not enter mass production until after 2032, implying that the duration of this expansion cycle far exceeds prior market expectations. The firm maintains its "Buy" rating on TSMC and raises its 12-month target price for the Taiwan-listed stock from NT$3,100 to NT$3,300, implying a 28% upside from current levels; the target price for the US-listed American Depositary Receipts (ADR) is simultaneously raised from $620 to $660, representing an upside of approximately 40%.

Goldman Sachs analysts Evelyn Yu and James Schneider pointed out in the report that the synergistic expansion in demand for AI accelerators, networking chips, and server CPUs constitutes the core driver of TSMC's performance over the next two years. Goldman Sachs forecasts that TSMC's revenue growth (in USD terms) will reach 42.0% in 2026 and 36.9% in 2027, while raising its earnings per share (EPS) forecasts for 2027 and 2028 by 6.5% and 8.2% to NT$150 and NT$195.67, respectively. The report also emphasizes that the acceleration in server CPU demand driven by Agentic AI is one of the most significant structural changes in demand over the past year.

Capital Expenditure Significantly Raised; Texas Fab Mass Production Delayed Until After 2032

Goldman Sachs maintains its 2026 capital expenditure forecast for TSMC at $64 billion, but raises the forecasts for 2027 and 2028 from $78 billion and $82 billion to $85 billion and $98 billion, respectively.

Goldman Sachs attributes this upward revision to two factors: first, cost inflation pressures from equipment suppliers, and second, initial expenditures for the potential new fab in Texas.

However, Goldman Sachs explicitly states that the Texas fab is not expected to enter mass production until after 2032 and will not make a substantial contribution to capacity in the short term. Regarding existing plans, Goldman Sachs maintains its forecasts for TSMC's N3 and N2 node capacity: by the end of 2027, monthly capacity for N3 and N2 is projected to be 200,000 wafers and 140,000 wafers, respectively, further expanding to 220,000 wafers and 200,000 wafers per month by the end of 2028.

In terms of advanced packaging, Goldman Sachs expects CoWoS annual capacity to increase from 675,000 wafers in 2025 to 2.73 million wafers in 2027 and 3.48 million wafers in 2028, with an average annual growth rate consistently exceeding 100%, providing key support for large-scale shipments of AI chips.

Strong Quarter-on-Quarter Growth in Q3; N2 Ramp-Up Temporarily Pressures Gross Margin

Goldman Sachs expects TSMC's Q3 revenue to increase by 15.3% quarter-on-quarter (in USD terms), with a further 11.0% growth in Q4, maintaining rapid expansion throughout the year supported by strong demand from AI and high-performance computing sectors.

Regarding profitability, Goldman Sachs predicts gross margins of 67.5% in Q3 and 67.3% in Q4, a slight decline from 67.7% in Q2, primarily reflecting the short-term cost dilution effect brought about by the N2 node ramp-up. Entering 2027, gross margin is expected to rebound to 67.5% as N2 yields improve, the product mix optimizes, and utilization rates remain high, further improving to 67.8% in 2028.

Goldman Sachs has simultaneously raised its earnings forecasts for 2026 through 2028, with a 1% upward revision for 2026, and 7% and 8% increases for 2027 and 2028, respectively. This is mainly due to the gross margin dilution effect from the N2 ramp-up being lower than previously expected, as well as higher utilization rates for N2/N3 nodes driven by AI and high-performance computing demand than assumed.

AI Demand Landscape Expands; Server CPUs Become New Growth Pole

The report heavily emphasizes that TSMC's AI demand drivers are extending from GPU accelerators to broader fields. Citing comments from TSMC management at the Communacopia technology conference, the report notes that the continuous decline in AI inference costs will stimulate larger-scale consumption of AI applications, while continued investment by US cloud service providers further solidifies demand visibility.

The rise of Agentic AI is accelerating demand for server CPUs, which Goldman Sachs lists as the most important structural change in demand over the past year. This means TSMC's advanced node customer base is further broadening, no longer limited to AI accelerator clients like Nvidia, but extending to CPU players such as Intel and AMD. Management also stated that despite the continuous expansion of N2 and N3 capacity, supply still struggles to meet robust demand.

Goldman Sachs stated that at the Q3 analyst meeting scheduled for October 15, the market will focus on management's latest views on AI demand prospects through 2030, the scale and timeline of US expansion plans, and whether TSMC can continue to widen its technological lead over Intel, Samsung, and emerging competitors like Terafab.