BEIJING, September 15 (TMTPost)— Thanks to artificial intelligence (AI) frenzy, Taiwan Semiconductor Manufacturing Co. (TSMC), the world’s leading chip maker for other companies, offset some of headwinds in the current semiconductor down cycle.

Credit：TMTPost

TSMC posted net revenue of NT$188.69 billion (US$5.89 billion) in August, with a 13.5% year-over-year (YoY) decline of 13.5%, though the revenue still rose 6.2% from the previous month.  Revenue for January to August 2023 totaled NT$1.35578 trillion, decreasing 5.2 % from the same period last year. As of August, the Taiwan-based manufacturer had a consecutive sixth month of drop in revenue, highlighting the weakening demand for electronics amid macroeconomic headwinds.

TSMC recorded net income of NT$181.72 billion in the quarter ended June 30 with a 23.4% YoY decrease, its first quarterly decline in profit since the second quarter of 2019. The net revenue that quarter fell 10% YoY to NT$480.84billion with a 10% YoY decrease that quarter, following a 15% YoY plummet in revenue in March, the first monthly revenue fall for the chipmaker since May 2019.

Surprisingly, TSMC lowered its 2023 revenue forecast when releasing the second quarter results in July. The company expected revenue for the year to drop about 10%, while the previous projection is a low-to-mid single digit decrease. Management also warned investor not to set too high expectations for demand for AI products.

Analysts believe that TSMC’s less-than-estimated slide in both the top and bottom line in the June quarter was mainly driven by the AI boom. TSMC is a major contract manufacturer for Nvidia graphics cards. Nvidia is the semiconductor designer that dominates the market for AI chips, which empower AI systems including the large language model behind ChatGPT.

TSMC is expected to gain traction from both Nvidia and Apple to meet its revenue guidance of the third quarter. Nvidia’s best-selling AI GPU H100 is made on TSMC’s 4-nanometer (nm) chip-making process. As one of the first customers that adopt TSMC’s 3nm process technology, Apple is set to deliver part of its flagship iPhone series empowered by the A17 bionic chip fabricated on 3nm process.

While AI boosts demand for TSMC’s advanced chips, other side of the same coin flags the potential hit. TSMC CEO C.C. Wei said in July that his company is unable to fulfill customer demand buoyed by AI boom. TSMC Chairman Mark Liu recent warned that lacking of advanced packaging capacity used to stitch the silicon together is holding up production. Demand for chip on wafer on substrate (CoWoS) package for advanced chips has tripled in the past year, Liu noted. Therefore, TSMC fails to meet 100% of customers’ CoWoS needs at present, but will try its best to meet about 80 %, Liu said. CoWoS is used in some of the most advanced chips on the market today, and almost all the applicable high-bandwidth memory (HBM) which is ideal for AI workloads relies on the packaging technology.

Liu expects insufficient CoWoS capacity is a temporary bottleneck, and it will take about one and a half year to ease the shortage as TSMC expands capacity. The supply laggard suggests shortage of Nvidia’s data center GPUs could last through 2025.

TSMC CEO C.C. Wei said at an earnings call in July that the company plans to roughly double its capacity for advanced packaging. Calling CoWoS capacity “very tight ”, Wei vowed TSMC to increase capacity as soon as possible and forecasted the tightening to be eased next year, probably towards the end of next year. The chipmaker later that month confirmed it is investing nearly NT$90 billion (US$2.87 billion) in a CoWoS packaging fab for AI chips at the Tongluo Science Park in Miaoli County. The facility in northern Taiwan will cover seven hectares and is expected to complete construction by the end of 2026, with mass production ready in the third quarter of 2027.