Goldman Sachs: TSMC May Raise Prices Nearly 10% by 2027! AI ASICs to Surpass GPUs, PCB and ABF Could Become New Bottlenecks

The AI boom continues to drive semiconductor demand, and even though recent tech stocks face correction pressure after strong gains, the long-term outlook for AI infrastructure remains optimistic. In a recent closed-door meeting, Goldman Sachs indicated that as AI computing demand expands, TSMC (2330-TW) may again raise prices for advanced processes and advanced packaging by 2027. Custom AI chips—AI ASICs—are also expected to grow faster than GPUs, while supply chains for components such as PCBs and ABF substrates may become the next focus due to yield and capacity constraints.

The report notes that recent market sentiment has turned more conservative, reflecting two main concerns. First, investors are questioning whether cloud service providers’ massive AI capital expenditures can be sustained and whether AI commercialization will meet expectations. Second, uncertainty around U.S. interest rate policy has led investors to reduce exposure to high-risk assets.

Even when companies issue better-than-expected guidance, market reactions remain relatively muted. After TSMC recently raised its capital expenditure and operational outlook, its stock price still declined, indicating that funds are beginning to take profits and reduce holdings.

Nonetheless, Goldman Sachs maintains a positive view on TSMC’s medium- to long-term operations. The report initially estimated a 5% price increase for TSMC’s advanced processes and packaging by 2027, but recent supply chain updates suggest the hike could approach 10%.

If this pricing strategy succeeds, it could offset margin pressures from early 2nm volume production and overseas fab expansion, resulting in 2027 profitability exceeding market expectations.

Beyond price adjustments, TSMC’s capital expenditure is also expected to continue rising. Goldman Sachs notes that TSMC has raised its 2026 capex forecast to between $60 billion and $64 billion. According to supply chain surveys, 2027 capex could further increase to $75 billion–$80 billion, indicating that AI-related investments show no signs of cooling and will drive the semiconductor equipment sector into a new cycle of growth.

The report also forecasts that AI ASICs will become the biggest growth engine over the next two years, potentially surpassing GPUs. For example, Google’s (GOOGL-US) TPU shipments are projected to increase at least 2–3 times by 2027 compared to this year.

After MediaTek (2454-TW) joined Google’s TPU project, its AI ASIC shipments in 2027 could exceed 4 million units, with related business revenue accounting for nearly half of the company’s total revenue.

Moreover, MediaTek’s next-generation AI ASIC project, codenamed 'Humor Fish,' is scheduled for mass production between late 2027 and 2028. Due to expanded involvement, the product’s unit price could be 3–4 times higher than current projects, leading the market to anticipate MediaTek’s EPS in 2028 could surpass NT$400.

On investment strategy, Goldman Sachs believes the market has gradually shifted from chasing high-P/E, high-growth stocks toward more reasonably valued value stocks.

Looking at 2027 estimated P/E ratios, TSMC stands at around 16x and MediaTek at 18x, both offering relatively stable investment appeal. In contrast, some high-valuation AI概念股 have recently faced significant correction pressure.

Notably, recent price increases in certain mid-to-low-end electronic components are not due to a sudden rebound in end-demand, but rather changes in supply structure.

Due to strong demand for high-end AI products, manufacturers have shifted capacity to high-end MLCCs and T-glass, reducing supply of mid-to-low-end products and pushing prices upward.

However, Goldman Sachs warns that this price surge stems primarily from supply contraction, not increased demand. As the consumer electronics peak season ends, price increases may gradually slow.

Meanwhile, the AI server supply chain may face new bottlenecks by year-end. Goldman Sachs states that third-gen CCL, NVIDIA’s (NVDA-US) Rubin platform, and Google projects are all set to enter mass procurement phases from late Q3 to early Q4 this year. If multiple demands are released simultaneously, some components could face supply shortages, potentially affecting AI server delivery timelines.

Notably, Google’s supply chain faces higher risks than AWS or NVIDIA, warranting continued monitoring.

Beyond component supply, next-generation AI server platforms are placing higher demands on PCB manufacturing capabilities. The report indicates that yield challenges have already emerged for NVIDIA’s Rubin compute board, reflecting that current major PCB suppliers face insufficient precision in high-end HDI processes.

With Rubin adopting 224G SerDes technology and future products upgrading to 448G, PCB equipment precision and testing capabilities must improve in tandem. Suppliers are expected to accelerate investments in laser drilling, back-drilling, and advanced inspection equipment to improve yields. In the future, PCBs could become a new bottleneck in the AI server supply chain.

Regarding ABF substrates, Goldman Sachs points out that high-end capacity is nearly fully utilized. Unabsorbed orders are gradually shifting to lower-tier capacity, but since low-end lines produce high-end products with low yields, this significantly increases capacity consumption, keeping low-end capacity tight as well.

The report expects that if high-end demand remains strong, spot price increases for low-end ABF substrates could even exceed those of long-contract-priced high-end products.

Overall, Goldman Sachs sees no clear signs of slowing AI investment. TSMC’s continued capex increases reflect sustained strong demand for AI chips from large data centers.

Although short-term markets are affected by valuation corrections and macroeconomic factors, from AI chips and advanced packaging to PCBs and ABF substrates, supply chain capacity remains extremely tight, and there are no clear signs of cooling in AI infrastructure demand in the near term.