Currently, the global semiconductor industry is entering a new period of transformation. With the rapid development of AI, big data, cloud computing, and other technologies, the demand for high-performance computing chips, optical communication chips, and advanced packaging has surged, and recent reports suggest that prices for these types of chips are increasing.

Advanced Process and Advanced Packaging Products May See Price Increases

According to a recent report by Morgan Stanley, TSMC is considering raising prices for its 3nm process and CoWoS advanced packaging technology in response to soaring market demand. TSMC plans to implement these price increases in 2025, with the cost of its 3nm process potentially rising by up to 5%.

Industry analysts point out that on the demand side, major AI chip manufacturers such as NVIDIA and AMD heavily rely on TSMC’s 3nm process, and the explosive growth in AI technology has driven continuous demand for these chips, contributing to the price increase.

On the supply side, the high research and production costs associated with advanced process technology—including equipment investment, material costs, and R&D personnel—add significant pressure to the supply chain. Multiple factors have led to a tight supply of such chips, further driving up prices.

Additionally, TSMC’s 5nm and 4nm process quotes have increased more than previously anticipated by 4%, with some price hikes reaching as much as 10%.

Reports indicate that TSMC also plans to raise prices for its CoWoS advanced packaging technology, with potential increases between 10% and 20%. High demand for CoWoS from major companies like NVIDIA, AMD, Microsoft, Amazon, and Google has resulted in a shortage of CoWoS packaging capacity, which has driven up prices.

According to TrendForce research, NVIDIA is the primary driver of demand for CoWoS, and with the upcoming launch of its Blackwell series, demand for CoWoS is expected to increase by more than 10 percentage points annually by 2025.

Optical Communication Chip Sector Begins Price Increases

Demand for high-speed, high-bandwidth, and low-latency optical communication is rising, particularly in data centers, enterprise networks, and telecommunications, driving demand in the optical communication chip market. Recently, media reports revealed that Marvell, a major optical communication chip manufacturer, has issued a price increase notice, with its entire product line set to see price hikes starting January 1, 2025. According to TrendForce, Marvell ranked sixth in the global IC design market in 2023.

Industry forecasts predict that, driven by ongoing advances in optical communication technology and expanding applications, the global optical communication chip market will grow rapidly in the coming years.

The development and application of technologies such as silicon photonics, optoelectronic hybrid integration, and high-performance photonic chip materials are expected to bring new growth points and opportunities to the optical communication chip market.

(Photo credit: Marvell)

According to Liberty Times, citing Wccftech, while initial expectations suggested that Huawei’s new Mate 70 smartphone would feature a next-generation flagship SoC, like the Kirin 9100, recent rumors indicate that Huawei may instead equip the standard Mate 70 model with a different Kirin chip. Only the high-end models, such as the Pro, Pro+, and RS Ultimate, are likely to receive the advanced Kirin 9100 chip.

Wccftech reports that this decision highlights Huawei’s ongoing challenges with chip yield rates, which have limited its ability to produce the new high-performance chips at scale. Consequently, Huawei has been compelled to reserve these advanced chips for higher-priced models.

Leaked information previously indicated that the Kirin 9100 is manufactured by SMIC using a 6nm process, according to Wccftech. However, since SMIC is limited to using older DUV equipment, production costs remain high, and yields are still constrained, preventing Huawei from producing these chipsets in large quantities. As a result, Huawei can only use this chip in a limited number of high-end models, as the report noted.

As for the standard version of the Mate 70, the report indicated that it may feature a different Kirin chip that’s slightly less powerful than the high-end processor. Some speculate that the standard model might adopt the previous generation Kirin 9000s or 9010, and at this stage it is still uncertain whether the Kirin 9100 will appear in the top-tier versions of the Mate 70 series, as the report from Liberty Times pointed out.

Read more

• [News] Huawei Expected to Unveil Mate70 and Foldable Device Following Apple’s iPhone 16 Release
• [News] Huawei’s Mate 70 Will Reportedly Use a Kirin SoC Made with SMIC’s 7nm Node instead of 5nm

(Photo credit: Huawei)

Amid the passion of the wild election, U.S. chip giant Qualcomm reported an upbeat first-quarter (ended December 31) sales forecast, with revenue rising between USD 10.5 billion and USD 11.3 billion, eyeing for an over 10% quarter-over-quarter growth at most. The strong momentum, according to a report by Reuters, could be attributed to the smartphone market recovery led by China.

Boosted by New Smartphone Launches led by Xiaomi, Oppo

It is worth noting that China remains Qualcomm’s largest market, and the momentum is driven by new smartphone releases from brands like Xiaomi, Oppo, and Vivo, according to the report. Qualcomm has derived 46% of its revenue in its most recent fiscal year from customers with headquarters in China, Reuters says.

A previous report by the South China Morning Post suggests that China’s smartphone maker Xiaomi will be the first to equip Qualcomm’s newly-released Snapdragon 8 Elite with its Xiaomi 15 series at the end of October, followed by other local smartphone brands such as Honor, Oppo’s OnePlus and Realme.

As the smartphone market starts to rebound following a challenging 2023, Qualcomm’s positive outlook is said to be driven by consumers upgrading devices for AI applications like chatbots and image generation tools, Reuters notes.

Business Expansion beyond Apple Remains Key  

On the other hand, Qualcomm is working hard to diversify its revenue streams in anticipation of the eventual end of its profitable partnership with Apple, which is developing its own modem chips to replace Qualcomm’s. According to Reuters, though the agreement to supply chips to Apple lasts until at least 2026, attention is on whether Qualcomm’s expansion into laptops and AI-driven data centers will grow swiftly enough to balance any future reductions in Apple-related revenue.

Regarding the potential impact if President-elect Donald Trump does impose broad tariffs of 10% to 20% on nearly all imports, with potential tariffs exceeding 60% on Chinese goods, Reuters notes that if higher tariffs were applied to chips from Taiwan, though rather unlikely, could incentivize Qualcomm to shift manufacturing to the U.S.

For the fourth quarter (ended September 30), Qualcomm posted a net income of USD 2.92 billion, or USD 2.59 per share, marking a significant increase from last year’s USD 1.49 billion, or USD 1.23 per share. The company’s total revenue for fiscal 2024 reached USD 38.9 billion, a 9% rise compared to 2023, according to its press release.

Read more

• [News] Qualcomm Rumored to Decide Its Next Move Regarding Potential Intel Acquisition after US Election

(Photo credit: Qualcomm)

According to a report from Tom’s Hardware, while major tech companies are investing heavily in datacenter GPUs, the lifespan of these GPUs may only be 1 to 3 years, depending on their utilization rates.

The report, citing a general architect at Alphabet, noted that because GPUs are under heavy workload of AI training and inference, they tend to wear out more quickly than other components.

According to the report, in datacenters operated by cloud service providers (CSPs), the utilization rate of GPUs for AI workloads ranges from approximately 60% to 70%.

The report indicated that, citing the words from the general architect at Alphabet, at this utilization rate, a GPU can typically survive for 1 to 2 years, or up to 3 years. While the report stated that this claim cannot be considered 100% accurate and requires further confirmation, it highlighted that modern datacenter GPUs for AI and HPC applications consume and dissipate 700W of power or more, which is significant stress for chips.

One way to extend the life of the GPUs is to reduce the utilization rate, according to the report. However, to reduce the utilization rate implies that the GPUs will lose value more gradually and it will take longer to return their capital, which isn’t ideal for business. Therefore, the report pointed out that most cloud service providers will use their GPUs at a high utilization rate.

The report also references a study conducted by Meta, which describes training its Llama 3 405B model on a cluster powered by 16,384 NVIDIA H100 80GB GPUs. According to the report, in that study, the model flop utilization (MFU) rate of the cluster was about 38% (using BF16), while during a 54-day pre-training snapshot, out of 419 unforeseen disruptions, 148 (30.1%) were caused by GPU failures (including NVLink fails) and 72 (17.2%) were due to HBM3 memory failures.

This result carried out by Meta, according to the report, is quite favorable for NVIDIA’s H100 GPUs. If GPUs and their memory fail at Meta’s rate, the annualized failure rate will be about 9%, and in 3 years, it will be about 27%. However, GPUs will likely fail more frequently after a year of heavy use, as the report pointed out.

(Photo credit: NVIDIA)

AMD posted third-quarter results on October 29th, with a quarterly revenue of USD 6.8 billion and net income to USD 771 million, while data center revenue surged 122% year-over-year. With new products such as MI300X hitting the market, the world’s second largest data center GPU provider also raises its AI chip sales forecast for this year to USD 5 billion, up from an earlier estimate of USD 4.5 billion, according to a report by CNBC.

For the fourth quarter of 2024, according to the company’s press release, AMD expects revenue to be approximately USD 7.5 billion, plus or minus USD 300 million. At the mid-point of the revenue range, this represents year-over-year growth of approximately 22% and sequential growth of approximately 10%. Non-GAAP gross margin is expected to be approximately 54%.

Fourth Quarter Forecast Falls to Impress; Concerns Raised on Capacity Constraints

However, the fourth-quarter forecast is slightly below market expectations, which raises concerns about whether the growth of the AI sector might be slowing down. According to Bloomberg, analysts had an average estimate of USD 7.55 billion.

AMD CEO Lisa Su reiterated that the company still sees robust momentum in AI, as interest from customers and partners in the MI325X is strong, a report by CNBC notes. AMD plans to begin production shipments of the MI325X this quarter, according to Su.

In October, AMD introduced the MI325X, and projected that the AI GPU market could reach USD 500 billion by 2028.

Nonetheless, Su also said that the environment will “continue to be tight”, but AMD has also planned for significant growth going into 2025, according to Bloomberg. She stated that the company feels good “about our overall supply-chain capability,” Bloomberg indicates.

AMD’s major foundry partner, TSMC, indicated in July that constraints on AI chip production will persist into 2025, which may imply a significant hurdle for clients like AMD, as it not only has to compete with NVIDIA on product performance, but also on the race of securing capacity.

Strong Data Center Revenue with 122% YoY Increase, while Gaming/ Embedded on the Decline

For the third quarter of 2024, AMD delivered a quarterly revenue of USD 6.8 billion, gross margin of 50%, operating income of USD 724 million, net income of USD 771 million and diluted earnings per share of USD 0.47. On a non-GAAP basis, gross margin was 54%, operating income was USD 1.7 billion, net income was USD 1.5 billion and diluted earnings per share was USD 0.92.

AMD’s AI chips are included in its data center segment, which saw annual sales more than double, reaching USD 3.5 billion. Overall, data center revenue rose 122% year-over-year. Su attributed the strong results to higher sales of EPYC and Instinct data center products and robust demand for the Ryzen PC processors, according to AMD’s press release.

The company also sees robust growth in its client segment, as revenue was USD 1.9 billion, up 29% year-over-year and 26% sequentially primarily driven by strong demand for “Zen 5” AMD Ryzen processors.

However, the gaming segment revenue was USD 462 million, down 69% year-over-year and 29% sequentially primarily due to a decrease in semi-custom revenue. According to CNBC, this could be attributed to reduced “semi-custom revenue” from custom chips used in consoles like the Sony PlayStation 5.

Embedded segment revenue was also declining, down 25% year-over-year to USD 927 million, as customers normalized their inventory levels. On a sequential basis, revenue increased 8% as demand improved in several end markets.

Read more

• [News] AMD Unveils its Latest AI Accelerators MI325X and MI355X, with TSMC Emerges as the Major Beneficiary
• [News] Intel Shifts to TSMC for Chip Production: A Looming Threat for AMD?

(Photo credit: AMD)