Morgan Stanley's latest research report states: As major cloud service providers (CSPs) compete to establish their own AI computing, applications, and ecosystems, Taiwan Semiconductor will be the biggest beneficiary of AI empowerment

On December 17th, Morgan Stanley's latest research report stated: As major cloud service providers (CSPs) compete to establish their own AI computing, applications, and ecosystems, we believe that Taiwan Semiconductor Manufacturing Company (TSMC) will be the biggest beneficiary of AI empowerment, as mentioned in our ASIC Global Insights report. In this report, we pointed out that major CSPs are still struggling to develop their own AI ASIC chips to compete with NVIDIA's GPUs. These efforts not only help them establish a second source for AI chips but also provide a potentially more energy-efficient and cost-effective alternative. In other words, regardless of whether ASICs or GPUs emerge as the winners in the future AI market share, we believe TSMC will still benefit from its leading process nodes and advanced packaging technologies.

At the analyst meeting in the third quarter of 2024, TSMC stated that it expects its CoWoS packaging capacity to double or more year-on-year by 2025, following a doubling in 2024. This forecast indicates that the inflection point for AI capital expenditures remains difficult to predict, but we believe supply chain data remains positive. Based on current supply chain visibility, we have raised our optimistic forecast for the market size in 2025 to $235 billion, which includes approximately $200 billion in AI GPU production value, as well as $30 billion in commercial ASICs, other general-purpose GPU solutions, and AI CPUs. In our optimistic scenario, we expect the compound annual growth rate (CAGR) of the cloud AI semiconductor market to reach 43% from 2023 to 2030.

TSMC to Benefit from Cloud AI Market, Regardless of ASIC or GPU Winning

Many investors believe that most of TSMC's AI semiconductor foundry services are aimed at NVIDIA, but we estimate that NVIDIA's AI GPUs will contribute only about 70% of TSMC's AI semiconductor revenue by 2025. Meanwhile, we believe that AI ASICs will become a more significant contributor to TSMC's AI revenue by 2027, expected to account for about 25% of TSMC's AI revenue, while the share of NVIDIA GPUs will decrease to about 65%. Our analysis is as follows:

1. Contribution of AI GPUs to TSMC's Revenue

Last year, we only estimated that NVIDIA's AI GPUs would contribute 25% to TSMC's revenue. However, this year, this proportion has increased mainly due to strong demand for Hopper and Blackwell, as well as CoWoS and testing prices (including TSMC's probe pad costs) exceeding expectations. Therefore, we expect TSMC to generate $6.8 billion in revenue from NVIDIA's AI GPUs in 2024, accounting for 8% of TSMC's total revenue.

2. Contribution of AI ASICs to TSMC's Revenue

For AI ASICs, we expect them to account for 4% of TSMC's total revenue in 2024 (including CoWoS and testing), approximately $3.5 billion, primarily from Google's TPU v5 and AWS's Trainium/Inferentia 2 projects. By 2025, this revenue is expected to grow by 70%, reaching $6.7 billion.

3. Contribution of CoWoS Packaging to TSMC

AI semiconductors heavily rely on advanced packaging technologies (such as CoWoS and SoIC, excluding InFO). In 2024, based on a capacity of 32k wafers per month for CoWoS, TSMC is expected to generate over $6.5 billion in revenue from this, and this value is expected to double by 2025.

4. AI Server CPUs

NVIDIA's Grace CPU accounts for a major share of TSMC's AI server CPU production, and it is expected to contribute even more by 2025. We anticipate that the front-end wafer revenue for 2024 and 2025 will be $8 billion and $11.4 billion, respectively. By 2026, the revenue contribution from AI CPUs is expected to double, accounting for about 1% of TSMC's AI-related revenue.

In our optimistic scenario, we expect the global cloud AI market to reach $405 billion by 2027 and grow to $498 billion by 2028, which aligns with AMD's recent forecasts during its Advancing AI event. If more companies gain competitive advantages and continue to invest heavily in AI computing to enhance productivity, we believe this optimistic scenario could be realized; we also consider the early advancement of AGI applications (such as humanoid robots).

Therefore, we believe that from 2023 to 2030, the annual compound growth rate of the cloud AI semiconductor market could reach 43%, meaning that by 2030, the market size will exceed $500 billion. In an optimistic scenario, the global semiconductor market size is also expected to expand to $1.2 trillion, with cloud AI semiconductors accounting for about 45% of the spending share, and incremental growth will also be driven by AI investments.

TSMC: Earnings Forecast Adjustment

We have raised our earnings per share (EPS) forecast for 2024 by 1%, for 2025 by 2%, and for 2026 by 5%. This adjustment mainly reflects the increase in CoWoS revenue; the adjustment for 2026 is more significant. We have also made slight adjustments to the wafer price assumptions for 2025 to match TSMC's pricing changes in 2025: an increase in prices for processes of 5nm and below, and a decrease for processes of 7nm and above

We have raised our target price from NT$1,330 to NT$1,388 due to adjustments in our earnings forecasts.

We continue to use the residual income model to derive our base case value, which is also our target price. All key assumptions of the residual income model remain unchanged—cost of equity at 9.2% (Beta of 1.2, risk-free rate of 2.0%, equity risk premium of 6.0%), mid-term growth rate of 10.0%, and terminal growth rate of 4.0%.

We have also raised the target price in the optimistic scenario to NT$1,680 (from NT$1,610) and the target price in the pessimistic scenario to NT$770 (from NT$740).

TSMC's stock is currently trading at a 17 times price-to-earnings (P/E) ratio based on the expected earnings per share (EPS) for 2025, close to its average P/E ratio over the past decade. The implied P/E ratio of our target price is 23.5 times, higher than the average P/E ratio of 19 times plus one standard deviation. We believe this implied target P/E ratio is reasonable as we continue to believe that AI will outpace the growth of non-AI industries in the coming years, and TSMC is a key driver of this AI trend, covering front-end wafer production and back-end advanced packaging technology.