As TSMC stands at the core of the global semiconductor industry, NVIDIA founder Jensen Huang stirred up an AI frenzy when he visited Taiwan in June. According to a report from Commercial Times, during his visit, he met with TSMC founder Morris Chang and then-President C.C. Wei, and personally visited TSMC’s headquarters.

According to a report by Mirror Media on July 23rd, sources revealed that Huang requested TSMC to set up a dedicated CoWoS production line for NVIDIA at an external facility. However, this request was met with skepticism from TSMC’s senior management, resulting in a tense situation.

Notably, TSMC Chairman C.C. Wei recently pointed out that despite strong AI demand, the company has yet to achieve a supply-demand balance and continues to increase production capacity. Many customers are eager for advanced process capacity, and TSMC is striving to balance pricing and capacity. Wei also revealed that the profit margins for advanced packaging CoWoS are being adjusted to align more closely with the company’s average profit levels.

C.C. Wei emphasized TSMC’s commitment to meeting customer demands with three “Whatever” statements. Regarding timelines, he revealed that supply will remain tight through 2025, with hopes for improvement by 2026.

Sources cited by the report further estimates that current utilization rates for 5nm and 3nm processes are at full capacity. To address demand, production for 3nm is set to increase gradually from 100,000 wafers per month to approximately 125,000 wafers per month by the second half of the year.

The 2nm process is expected to start mass production in the fourth quarter of 2025, with a target monthly output of 30,000 wafers. With future expansion at the Kaohsiung plant, the combined monthly capacity of the Hsinchu Science Park and Kaohsiung facilities is projected to reach 120,000 to 130,000 wafers.

Despite escalating geopolitical risks, Wei stated that there will be no changes to TSMC’s expansion strategy, with projects in Arizona, Japan’s Kumamoto, and future European facilities proceeding as planned.

Currently, TSMC’s investment in its U.S. facilities has reached USD 65 billion. The Arizona site is expected to have three plants, with the first two set to start production in 2025 and 2028, focusing on 4/3nm and 3/2nm nodes, respectively.

TSMC’s advanced packaging plans in Taiwan include facilities in Hsinchu Science Park, Miaoli Zhunan, Taichung Central Taiwan Science Park, Tainan Southern Taiwan Science Park (taking over Longtan InFO), and Taoyuan Longtan (significantly expanding CoWoS). Construction is also scheduled to begin in 2024 at Miaoli Tongluo and Chiayi Science Park.

Read more

• [News] Decipher TSMC in Key Figures: Some Facts Needed to be Known
• [News] TSMC Reportedly Sees Strong 4nm Demand, with NVIDIA’s Order Up by 25%

(Photo credit: TSMC)

Recent comments from former U.S. President Donald Trump on protection fees have hit semiconductor stocks hard and brought TSMC back into the spotlight for investors.

As per a report from Barron’s, it’s suggested that America’s stance towards Taiwan is not the best diplomatic strategy. The following key figures highlight TSMC’s importance to both the U.S. and the global economy.

• 700 Billion

After Bloomberg published an exclusive interview with Trump, the Nasdaq Composite Index plummeted 2.8% on July 17th. Large semiconductor stocks, including TSMC and the seven major U.S. tech giants, collectively lost about USD 700 billion in market value overnight.

The report from Barron’s suggested that this phenomenon indicates that the market views TSMC not just as a foundry but also as a crucial supplier of key components for America’s largest and most important enterprises.

• 92%

According to the Semiconductor Industry Association, TSMC manufactures 92% of the world’s advanced semiconductors, while South Korea produces the remaining 8%.

Barron’s noted that though the U.S. aims to increase domestic production of advanced chips, targeting 20% of advanced chips to be produced locally by 2030,. However this will take some time to achieve.

• 33%

TSMC taks pride in itself that it is the world’s first dedicated semiconductor foundry. Namely, all chips produced by TSMC are supplied to semiconductor companies that do not manufacture their own chips. For companies like NVIDIA, AMD, Broadcom, Qualcomm, and others, TSMC is a major supplier.

Barron’s noted that approximately one-third of their chip manufacturing expenses go to TSMC. This has further underscored the significance of the company.

• 45%

NVIDIA and other companies purchase chips from TSMC and then resell them to other companies. According to Barron’s, Microsoft, Meta, Alphabet, Amazon, and Tesla contribute about 45% of NVIDIA’s sales. Although Apple currently doesn’t purchase many chips from NVIDIA, about 27% of Qualcomm’s sales and 17% of Broadcom’s sales come from Apple.

In theory, without TSMC, there would be no iPhone, AI servers, or other electronic products people rely on. For this reason, Tae Kim, a technology journalist at Barron’s, refers to the true risk of disruptions in TSMC’s operations as a “Global Depression.”

Read more

• [News] TSMC Dominates High-End Packaging Market, Potentially Impacting Opportunities for OSAT
• [News] TSMC Introduces “Foundry 2.0” to Include Packaging, Testing, Mask making and Others

(Photo credit: TSMC)

According to sources cited in a report from Reuters, NVIDIA is said to be planning to design a new flagship AI chip tailored for the Chinese market, which will still comply with current U.S. export control regulations.

NVIDIA, the global AI chip giant, unveiled its Blackwell chip series in March this year, with mass production expected to start later this year. The B200 chip in this series boasts powerful performance, capable of completing chatbot response tasks at speeds up to 30 times faster than the previous generation.

The sources cited by Reuters further point out that NVIDIA will collaborate with China’s Inspur to launch and sell this chip, tentatively codenamed B20. Inspur is one of NVIDIA’s primary distribution partners in China.

Currently, NVIDIA’s spokesperson has declined to comment on this news, and Inspur has also not issued any statements.

The U.S. government, citing national security concerns, began strictly tightening controls on the export of advanced semiconductors to China in 2023. Since then, NVIDIA has released three chips specifically for the Chinese market.

Per a previous report from TechNews citing industry sources, it is also believed that the US will significantly escalate the trade war after the presidential election, intensifying export restrictions on China.

It is noteworthy that the US government previously announced the imposition or increase of tariffs on Chinese electric vehicles, semiconductors, lithium batteries, and other products, with the semiconductor tariff rate set to rise from 25% to 50% by 2025.  Meanwhile, for the future direction of the US, it can be inferred that chips manufactured in Taiwan and South Korea may also face tariffs.

Read more

• [News] U.S. Reportedly Enhancing Chip Export Restrictions to China, while NVIDIA’s H20 Might Get Banned
• [News] China’s Third Big Fund Reportedly Set to Inject USD 27 Billion to Counter US Chip Restrictions

(Photo credit: NVIDIA)

According to a report from Tom’s Hardware, the U.S. is considering implementing new trade sanctions on China, looking to limit China’s access to advanced AI chip technology. This could result in a ban on NVIDIA’s HGX-H20 AI GPUs to China. If implemented, NVIDIA could potentially lose roughly USD 12 billion in revenue.

To comply with U.S. export regulations, NVIDIA introduced the HGX H20 GPU specifically for the Chinese market. Although it has reduced performance, it still offers powerful AI capabilities.

As per the report, the HGX H20 GPU features 296 INT8 TOPS/FP8 TFLOPS computational performance, 96 GB of HBM3 memory, and 4.0 TB/s memory bandwidth, making it competitive with the current entry-level AI chips on the market. Despite its downgraded performance, the HGX H20 outperforms Huawei’s self-developed Ascend 920 series AI chips in practical applications due to its better memory performance.

However, during the U.S. semiconductor export policy review in October, NVIDIA’s HGX H20 GPU might face a sales ban. The anticipated restrictions could take various forms, including product-specific bans, reduced computational power, or limited memory capacity.

Most Chinese AI companies have built their application ecosystems on NVIDIA’s CUDA computing platform, which makes switching to other platforms, like Huawei’s Ascend chips, both costly and time-consuming. Although the HGX H20 GPU’s computational performance is significantly lower than the H100, its full compatibility with NVIDIA’s CUDA computing platform makes it the preferred choice for many Chinese companies and applications, the report noted.

However, it is worth noting that despite the current export controls on China, Chinese companies still manage to acquire advanced NVIDIA GPU computing power for AI and high-performance computing through intermediaries and by renting cloud service servers from companies like Google and Microsoft. This is a primary reason which prompts the U.S. to tighten restrictions.

Additionally, the U.S. might extend export restrictions to other Asian countries, such as Malaysia, Indonesia, Thailand, and potentially overseas Chinese companies. However, due to the complexity of these measures, effective implementation poses significant challenges, according to the report.

TrendForce notes in April that the extension of export controls now includes not only the previously restricted AI chips from NVIDIA and AMD, such as the NVIDIA A100/H100, AMD MI250/300 series, NVIDIA A800, H800, L40, L40S, and RTX4090, but also their next-generation successors like NVIDIA’s H200, B100, B200, GB200, and AMD’s MI350 series.

In response, HPC manufacturers have quickly developed products that comply with the new TPP and PD standards, such as NVIDIA’s adjusted H20/L20/L2, which remain eligible for export.

Read more

• [News] NVIDIA Reportedly Facing Price Cut Pressure of H20 Chip in China Amid Competition with Huawei
• [News] NVIDIA’s China-Exclusive H20 to Begin Pre-sales Next Month

(Photo credit: NVIDIA)

The heightened tensions between China and the U.S. are expected to escalate further. According to a report from Commercial Times, this has been reflected in the past quarter’s increased demand for AI servers among Chinese Cloud Service Providers (CSPs) and enterprises.

Following NVIDIA’s recent rush orders, Taiwanese server manufacturers such as Inventec, Wistron, and Foxconn are also experiencing increased demand for H20-related orders from Chinese customers. Reportedly, these orders are expected to ramp up in the third and fourth quarters.

Industry sources cited by the same report further indicate that NVIDIA’s recent rush order, with a scale of 100,000 H20 units, is expected to be fulfilled by mid-fourth quarter. While Wistron, a major substrate supplier, started to ramp up shipments from late second quarter.

Inventec, which benefited from strong H20 demand from Chinese customers in the first half of the year, is said to be seeing a shift in AI server shipments to predominantly U.S. customers in the second half. However, the demand from Chinese customers has not disappeared and is expected to continue providing momentum for Inventec’s performance.

SuperMicro, which is reportedly expanding its server sales business in the Chinese market through the channel resources of Taiwanese graphics card manufacturer Leadtek, is also expected to bring positive benefits to Leadtek’s sales operations in China and Northeast Asia. Additionally, with the recovery of its own AI workstation business and the expected demand for H20 from Chinese customers, Leadtek is likely to become one of the major beneficiaries of the surge in orders from Chinese clients in the second half of the year.

On the other hand, due to the impact of U.S. chip restrictions, it’s hinted that some Chinese customers are increasingly leasing high-end AI computing power services from global companies to meet the current AI model training needs.

This trend is indirectly boosting the order pull for high-end AI server products from Tier 2 and Tier 3 data center operators in Europe and the Asia-Pacific region, simultaneously providing shipment momentum for related Taiwanese manufacturers.

According to sources cited by Commercial Times, the demand for H20 orders from Chinese customers has been quite unstable this year, with sudden spikes of urgent orders.

However, as the confrontation between China and the U.S. might intensify due to the upcoming U.S. presidential election, Chinese companies may be prompted to increase their demand for H20 in the coming quarters. This could significantly boost the AI server business for related Taiwanese manufacturers in the second half of the year.

Regarding the need for H20, TrendForce previously mentioned that Chinese companies would continue to buy existing AI chips in the short term. NVIDIA’s GPU AI accelerator chips remain a top priority—including H20, L20, and L2—designed specifically for the Chinese market following the ban.

Read more

• [News] NVIDIA Reportedly Facing Price Cut Pressure of H20 Chip in China Amid Competition with Huawei
• [News] NVIDIA’s China-Exclusive H20 to Begin Pre-sales Next Month

(Photo credit: NVIDIA)