Taiwan Semiconductor Manufacturing Co (TSMC) is facing significant challenges in bringing its most advanced chip technology to its new Arizona plant, the company’s CEO, C.C. Wei, said. Complex regulatory hurdles, labour shortages, and supply chain gaps have slowed progress, making it unlikely for the US factory to match Taiwan’s production timeline for cutting-edge chips. Wei noted that the Arizona project has already taken twice as long as similar facilities in Taiwan.
TSMC is investing $65 billion in three massive factories in Arizona, with support from the US government, including a $6.6 billion grant. However, Wei highlighted the high costs of compliance, including $35 million spent on establishing regulatory guidelines, as well as the logistical strain of shipping essential chemicals like sulfuric acid from Taiwan. Labour shortages have further complicated the project, requiring the relocation of workers from Texas and driving up costs.
Despite the obstacles, Wei expressed confidence in the factory’s ability to deliver high-quality chips, pointing to recent progress in producing advanced 4-nanometer chips for US clients. While most of TSMC’s cutting-edge manufacturing will remain in Taiwan, the Arizona plant marks a critical step in the US’s effort to diversify its semiconductor supply chain and reduce dependence on Asia.
Washington has blacklisted over two dozen Chinese entities, including Zhipu AI and Sophgo, for alleged links to restricted chip technology in Huawei processors. The Commerce Department has also tightened export controls on chips that could be diverted to Huawei.
Zhipu AI, backed by Tencent and Alibaba, was accused of contributing to China’s military modernisation through advanced AI research. Sophgo faced scrutiny after a chip found in Huawei’s Ascend 910B AI system matched one it ordered from Taiwan Semiconductor Manufacturing Co (TSMC).
The measures impose stricter licensing requirements for chip exports, targeting semiconductors at 14 or 16-nanometre nodes used in AI applications. New restrictions also affect DRAM memory, crucial for high-bandwidth AI processing, which could impact Chinese chipmaker CXMT.
Zhipu AI denied the claims, while Sophgo stated it had no direct or indirect ties to Huawei. Huawei and TSMC declined to comment on the latest sanctions, which build on previous curbs against Huawei and its network of suppliers.
China’s Commerce Ministry announced plans to investigate US government subsidies to its semiconductor sector following requests from China’s mature node chip industry. The ministry stated on Thursday that these subsidies, introduced under the Biden administration, allegedly provide American companies with an unfair competitive advantage in the global market.
According to the Chinese government, US firms have exported mature node chip products to China at reduced prices, causing harm to the interests of China’s domestic semiconductor industry. Beijing views these practices as a threat to its industry’s rights and competitive balance.
The investigation reflects rising tensions between the two nations over technology and trade, particularly as both seek to bolster their semiconductor sectors amid growing geopolitical competition.
Several chip and manufacturing industry groups have criticised the Biden administration’s new export controls, arguing they were implemented without sufficient consultation. A private letter sent to President Joe Biden on January 13 expressed concerns that the rules could harm US companies and shift market share to global competitors.
The Semiconductor Industry Association (SIA) and SEMI, representing chipmakers and manufacturing equipment firms, objected to the new licensing requirements for AI chip exports, including advanced high-bandwidth memory. They argued the lack of public input ignored the regulations’ economic and international consequences.
High-bandwidth memory, essential for AI chip production, is primarily manufactured by US and South Korean companies. The new rules could restrict its sale to China, further tightening controls on advanced technology exports.
A separate source suggested the restrictions may also affect companies like Lam Research, which previously benefited from a rule interpretation allowing expanded sales in China. Neither the SIA, SEMI, nor Lam Research commented immediately.
Taiwan’s government has expressed confidence in its management of AI chip exports following the island’s exemption from new US restrictions. Officials stated the exclusion should reassure partners about the country’s legal compliance and export controls.
The US announced stricter AI chip export measures this week, limiting shipments to most countries while continuing a ban on China, Russia, Iran, and North Korea. Taiwan remains a ‘tier one’ partner, enjoying unrestricted access to US AI technology. Taiwan’s Economy Ministry emphasised efforts to help local businesses stay informed about evolving export regulations.
Taiwan is home to TSMC, the world’s largest contract chipmaker and a key supplier for Nvidia. The government, alert to Beijing’s claims over the island, maintains strict export controls to China and has vowed to enforce US restrictions. TSMC suspended shipments to a Chinese company last year after a chip was illegally integrated into Huawei technology.
Huawei has faced US trade bans since 2019 due to national security concerns, with companies restricted from selling technology to the Chinese firm without a special licence.
President Joe Biden has signed an executive order to support the rapid expansion of AI data centres by providing federal land and resources. The initiative will allow AI facilities to be built on sites owned by the Defence and Energy departments, addressing the growing demand for computing power while promoting clean energy development. Companies using federal land for AI data centres will also be required to purchase a portion of American-made semiconductors, reinforcing the administration’s push for domestic chip production.
The order aims to ensure that the most advanced AI models are developed and stored within the United States, strengthening national security and economic competitiveness. The White House stressed the importance of securing energy supplies and transmission infrastructure to sustain AI growth, with experts predicting that by 2028, leading developers could need up to five gigawatts of capacity to train their AI models. Agencies have been directed to fast-track grid interconnections, permitting, and infrastructure development to meet these demands.
Efforts to keep AI technology within the United States align with broader national security concerns. The Commerce Department has announced new restrictions on AI chip exports to prevent China from accessing advanced computing power. White House technology adviser Tarun Chhabra highlighted the potential risks posed by AI, including its ability to aid in developing chemical, biological, and cyber warfare capabilities. Ensuring that AI data centres remain under US control will help safeguard military and national security interests.
Nvidia faces significant revenue pressure following new US restrictions on artificial intelligence chip exports. The regulations, aimed at limiting the distribution of advanced processors to most countries outside close US allies, are intended to prevent China from accessing technology that could strengthen its military capabilities. Analysts warn these curbs could hinder Nvidia’s growth, as nearly half of its chip sales could be affected.
A substantial portion of Nvidia’s revenue comes from global markets, with 56% of sales outside the US and China accounting for around 17%. The company’s stock dropped by 2% following the announcement. Nvidia Vice President of Government Affairs Ned Finkle argued the restrictions could stifle global innovation and weaken US leadership in AI development by imposing bureaucratic control over chip design and marketing.
Industry experts believe major cloud providers such as Microsoft, Google, and Amazon could benefit from the new rules. These firms may receive exemptions for data centre development, allowing them to expand market share as AI leaders. Their financial strength and advanced infrastructure make them well-positioned to thrive under the revised framework.
Uncertainty remains over the long-term impact of the export curbs, with the rules set to take effect in 120 days. Some analysts suggest the incoming Trump administration might revise the policy, potentially negotiating new deals with individual firms and countries while maintaining a focus on national security.
AI chip startup Blaize has announced plans to go public through a SPAC deal, which will see the company listed on Nasdaq with a valuation of $1.2 billion. Founded in 2011 by former Intel engineers, Blaize specialises in AI chips for edge devices such as drones, security cameras, and industrial robots. Unlike traditional data centre chips, its products are designed for real-world applications that prioritise low latency, power efficiency, and privacy.
The company has raised $335 million from prominent investors, including Samsung and Mercedes-Benz, and claims to have $400 million worth of deals in the pipeline. CEO Dinakar Munagala, who spent over a decade at Intel, emphasised that Blaize’s approach focuses on practical AI solutions for physical environments, differentiating the company from competitors like Nvidia, which primarily targets large-scale data centres.
Despite facing financial challenges, including a loss of $87.5 million in 2023, Blaize is betting on a future where AI chips are embedded into everyday devices. The startup is also involved in defence-related contracts, with one major deal involving AI systems capable of identifying troops and detecting drones, further highlighting its niche in edge computing.
Blaize’s IPO marks a significant shift in the AI chip industry, signalling investor interest in decentralised AI technologies that extend beyond traditional data centre applications.
Arm Holdings, a key supplier to the semiconductor industry, is planning significant price hikes and has considered entering the chip design market. The British company, which licenses technology to major firms like Apple and Qualcomm, has historically focused on royalties from intellectual property rather than manufacturing its own chips.
CEO Rene Haas and SoftBank’s Masayoshi Son are reportedly pushing for a more aggressive revenue strategy. Plans revealed during a recent trial against Qualcomm disclosed the ‘Picasso’ initiative, aiming to increase smartphone revenue by $1 billion over a decade. The approach involves raising royalty fees by as much as 300% for the latest chip designs.
Documents from 2019 showed Arm executives discussing these increases, but customers like Apple and Qualcomm, capable of designing their own chips, may avoid the higher fees. Arm has also explored making complete chips or chiplets, a strategy Haas described as long-term speculation rather than a confirmed plan.
Meetings with Samsung executives in 2022 further highlighted Arm’s strategy shift. Concerns over licensing agreements with Qualcomm led Samsung to shorten a supply deal with the chipmaker. Arm has not publicly confirmed any immediate plans for chip production or pricing adjustments.
Nvidia’s latest Blackwell AI chip racks are facing overheating issues, prompting major customers to delay or reduce their orders. Early shipments have reportedly suffered from technical glitches, affecting how the chips connect within data centre racks. As a result, companies including Microsoft, Amazon, Google, and Meta have scaled back their purchases, some opting for older Nvidia AI chips instead.
The hyperscalers had initially placed orders worth at least $10 billion each but are now reconsidering their plans. Microsoft, for example, intended to install 50,000 Blackwell chips in a Phoenix facility but has since switched to Nvidia’s previous-generation Hopper chips at OpenAI’s request. Nvidia has not commented on the situation, while Microsoft, Google, and Meta have yet to respond to inquiries.
Concerns over Nvidia’s sales extend beyond the overheating issues, as the US government has announced tighter restrictions on AI chip exports, potentially impacting revenue. Despite the setbacks, CEO Jensen Huang remains optimistic, stating that the company is still on track to generate billions in revenue from Blackwell chips this quarter. Huang has also denied earlier reports of overheating in liquid-cooled Blackwell servers.
