Before the performance meeting, Taiwan Semiconductor's two major customers "gifted": NVIDIA increased orders by 25%, and Apple will be the first customer for 2nm

According to the financial news app Zhitong Finance, on the eve of the crucial second-quarter earnings conference on Thursday, Taiwan Semiconductor (TSM.US), known as the "king of chip manufacturing," has been receiving positive news about its performance expectations. Recently, industry insiders revealed that due to the strong demand for the Blackwell architecture AI GPU to be mass-produced by NVIDIA (NVDA.US) globally, NVIDIA has significantly increased its AI GPU manufacturing orders with chip manufacturing giant Taiwan Semiconductor by at least 25%. Insiders also pointed out that Taiwan Semiconductor's largest customer, Apple (AAPL.US), is set to receive the first wave of orders for the most advanced process technology shortly after its launch by Taiwan Semiconductor in 2025—namely, the 2nm process, which will be used for the iPhone 17. Taiwan Semiconductor's next-generation 3D advanced packaging SoIC is planned to be used for Apple's M5 chip, expected to be mass-produced in 2025.

NVIDIA's newly launched AI GPUs based on the Blackwell architecture—namely B100/B200/GB200, according to some market research reports, may take until the fourth quarter of this year for the Blackwell architecture GPUs to achieve limited shipments and begin customer testing, with significant market deployment not expected until at least the first quarter of 2025.

Reports suggest that Taiwan Semiconductor is preparing to fully produce NVIDIA's next-generation flagship AI GPU based on the Blackwell architecture, but due to the strong demand from customers, NVIDIA has significantly increased its orders for Taiwan Semiconductor's Blackwell AI GPUs by 25%. This trend of increased orders indicates that global enterprises and some government agencies have not slowed down their demand for the most core AI hardware. The performance of NVIDIA and Taiwan Semiconductor in the second half of 2024 is expected to be outstanding, and perhaps even more so in 2025.

C.C. Wei, the new "helmsman" who serves as both Chairman and CEO of Taiwan Semiconductor, recently reiterated at the company's shareholder meeting his expectation that the iterative development of artificial intelligence technology will drive a strong recovery in the chip industry in 2024, and C.C. Wei has explicitly mentioned Taiwan Semiconductor's intention to raise prices. C.C. Wei also revealed at the shareholder meeting: Currently, almost all AI chips on the market are manufactured by Taiwan Semiconductor.

NVIDIA's next-generation AI GPUs based on the Blackwell architecture will see a new leap in ultra-high performance. Tech giants such as Amazon, Dell, Google, Meta, and Microsoft will heavily incorporate Blackwell AI GPUs into their latest data center AI server systems. Wall Street analysts widely speculate that the demand for NVIDIA hardware from these tech giants will far exceed market expectations According to analysts' expectations, the average price of the NVIDIA B100 AI GPU built on the Blackwell architecture is expected to be around $30,000 to $35,000. On the other hand, the super server chip GB200, which features NVIDIA's in-house Grace CPU based on ARM architecture and the NVIDIA B200 AI GPU, could be priced between $60,000 to $70,000 or even higher.

As this news spreads, the demand for the AI GPU family based on Blackwell and when the price increase mode for advanced processes of 5nm and below as well as advanced packaging will start, it is expected to become a hot topic in TSMC's performance conference call.

In addition, sources revealed that Apple will enjoy the first wave of production capacity for TSMC's new 2nm process. Reportedly, TSMC will start the trial production mode of its 2nm process this week, with mass production expected by 2025. Rumors suggest that Apple's next-generation M5 chip and its 3D-level SoIC advanced packaging will achieve mass production in 2025. Insiders stated that in order to prepare the SoIC advanced packaging capacity for major customer Apple, TSMC will need to double the SoIC capacity next year, with the current monthly capacity at only about 4,000 pieces. Furthermore, a significantly higher increase is expected in 2026.

In June, an ASML spokesperson revealed that ASML will deliver the latest high-NA EUV lithography machine to TSMC, its largest customer for lithography machines and known as the "king of chip manufacturing," by the end of this year. This news directly drove ASML's stock price surge at that time. For TSMC, as well as Intel and Samsung Electronics developing 2nm and below node manufacturing technologies, the ASML high-NA EUV lithography machine is crucial.

Compared to the standard EUV lithography machines currently produced by ASML, the main difference lies in the use of a larger numerical aperture (NA). The high-NA EUV technology uses a 0.55 NA lens, achieving an 8nm-level resolution, while the standard EUV technology uses a 0.33 NA lens. This new NA technology can print smaller feature sizes on chips, which is crucial for the development of process technology for 2nm and below chips. Currently, NVIDIA's H100/H200 AI GPUs mainly use TSMC's 4nm process technology, while the newly introduced Blackwell architecture AI GPU will adopt TSMC's 3nm process technology. In the future, NVIDIA AI GPUs and Apple AI smartphone chips are expected to predominantly use TSMC's 2nm or even 1.6nm process technology.

Wall Street is bullish! The soaring stock price of TSMC may be far from stopping

With decades of semiconductor manufacturing technology accumulation in the chip manufacturing field, TSMC has been at the forefront of chip manufacturing technology improvement and innovation for a long time (ushering in the FinFET era, leading the 2nm GAA era). With leading advanced processes and packaging technologies among global chip manufacturers, as well as a long-standing dominance in the vast majority of global chip foundry orders, Especially for advanced chip foundry orders with 5nm and below processes, TSMC's performance report and revenue generated from 5nm and below processes will be the market focus. This data may largely reflect the strong demand for AI chips such as NVIDIA's H100/H200 and AMD Instinct MI300 series based on the 5nm process.

Currently, TSMC dominates the market for almost all high-end chip packaging orders with 5nm and below processes thanks to its leading 2.5D/3D chiplet advanced packaging. The advanced packaging capacity is far from meeting the demand, with NVIDIA's H100/H200 facing supply shortages, mainly due to TSMC's limited 2.5D CoWoS packaging capacity. According to Wall Street analysts, starting from the second half of 2024, driven by the demand for NVIDIA's GB200 and AMD's new MI325 series, 3nm and below advanced processes will continue to bring huge revenue contributions to TSMC; it is expected that from next year onwards, the comprehensive price increase of 3nm and below processes and advanced packaging foundry services such as CoWoS will help accelerate revenue growth in 2025 and the following years.

Top Wall Street investment banks have recently raised their target stock prices for TSMC, indicating that the momentum of TSMC, once valued at over a trillion dollars, is not stopping. Since the beginning of this year, TSMC's ADR stock price has surged by over 80%. The core logic behind the Wall Street's major banks raising TSMC's stock price lies in the surge in demand for AI chips brought by the artificial intelligence boom, as well as the significant increase in prices for 3nm level processes and CoWoS advanced packaging foundry services that may lead to substantial profit growth and valuation increase by 2025.

In a recent report by Macquarie Securities, it is pointed out that TSMC has learned through supply chain visits that most customers have agreed to price increases to ensure stable supply, which will further drive up TSMC's gross margin. Analysts predict that by 2025, TSMC's gross margin will climb to 55.1%, and by 2026, it is expected to approach sixty percent, reaching 59.3%.

This price increase may be based on the incredibly strong market demand, TSMC's production capacity constraints, and cost considerations. Major TSMC clients such as Apple and NVIDIA have already made large-scale bookings for TSMC's 3nm family process capacity, and the queue for customer 3nm foundry contracts is expected to continue until 2026.

According to a report from Taiwan's Commercial Times, next year, TSMC's 3nm foundry pricing may increase by more than 5%, while the annual pricing increase for chiplet advanced packaging represented by CoWoS is expected to be between 10% and 20%. Goldman Sachs analysts expect Taiwan Semiconductor's 3nm and 5nm chip manufacturing prices to rise by "single-digit percentages," and have raised their 12-month target price by 19% to 1160 New Taiwan Dollars (with the latest Taiwan stock price around 1040). Goldman analysts, including Bruce Lu, recently wrote in a report: "We now see that, with the increasing enthusiasm around artificial intelligence, the risk-return profile of Taiwan Semiconductor is very attractive." "As the scale of artificial intelligence applications continues to expand, we believe Taiwan Semiconductor is one of the most core beneficiaries." In addition, Goldman Sachs has set a target price for Taiwan Semiconductor's American Depositary Receipt (ADR) as high as $218 (with the latest closing price of Taiwan Semiconductor's ADR around $187)