Nvidia formalized a deal worth over $500 billion, equivalent to about R$ 2.5 trillion, with SK Hynix, a South Korean company. The objective of this long-term partnership is to secure the supply of high-bandwidth memory (HBM) and accelerate the development of artificial intelligence (AI) infrastructure.
This pact was signed on Friday, the 24th, during a technology event held in San Francisco. The initiative aims to protect the manufacturer against the global shortage of essential components for its Graphics Processing Units (GPUs). The contract stipulates the priority supply of advanced semiconductors and the joint development of new memory technologies for the next generation, in addition to the creation of large data centers.
The new computing facilities, scheduled to become operational in 2027, will be dedicated to providing cloud processing capacity for governments, corporations, and AI model developers. At the core of this project, SK Telecom, a subsidiary of the same business group, will build an AI data center with a projected energy capacity of 2 gigawatts. This structure will utilize Nvidia's Vera Rubin processor architecture combined with SK Hynix's HBM4 memories, requiring hundreds of thousands of GPUs to support heavy generative AI applications and autonomous systems.
The collaboration establishes a guarantee of stability for Nvidia, especially since HBM memory has become the main production bottleneck in the sector. Due to the need for extremely high data transfer speeds to power deep learning chips, these components face demand far exceeding current manufacturing capacity. Through the engineering collaboration outlined in the agreement, both companies plan to synchronize the memory development cycle with future Nvidia hardware releases.
In addition to the main contract, Nvidia confirmed an investment of $1 billion (R$ 5 billion) in Naver, a South Korean cloud company, aiming to expand computing complexes in the Asian country. This announcement was made in the presence of representatives from the South Korean government. This move occurs amid a broad restructuring of the semiconductor market to meet the growing demand for intensive computing. In response to this competition, Samsung Electronics also closed a $200 billion (R$ 1 trillion) deal with Broadcom to expand cooperation in chip and memory foundry services.
Nvidia has begun shipping a limited quantity of its H200 artificial intelligence chips to China after receiving governmental authorizations from the United States. This news was disclosed during a Congressional hearing in Washington by an American trade official this Tuesday (14th).
This shipment involves one of the most sophisticated processors from the American company and occurs within the context of the technological dispute between the United States and China. This rivalry is marked by export restrictions on semiconductors classified as strategic for military purposes.
According to the U.S. Department of Commerce, only a very small volume of units has been released so far. Although Chinese companies have obtained specific permits to acquire these components, the extent of future sales still awaits security assessments and complementary approvals.
Jeffrey Kessler, U.S. Deputy Under Secretary of Commerce for Industry and Security, confirmed the first shipments of the H200 during a session of the House Foreign Affairs Committee. He stated that exports made to date have been minimal.
The American Department of Commerce reported that approximately ten Chinese companies had already received authorization to purchase the chips in May, but no deliveries had been finalized at that time. Among the approved companies were Alibaba, Tencent, and ByteDance (responsible for TikTok).
Additionally, a unit of the Chinese telecommunications equipment manufacturer ZTE and two other Chinese companies were recently included in the list of companies authorized to acquire the advanced AI chips produced by Nvidia and AMD.
The U.S. government maintains rigorous controls over the sale of sensitive technologies to Chinese entities. Companies listed under restrictions must obtain specific licenses before receiving any product, software, or equipment originating from the United States.
The commercialization of the H200 has become a central point of contention in the technological dispute between the world's two largest economies. Washington alleges that high-performance semiconductors can be used in Chinese military projects, seeking to limit Beijing's access to such technologies.
During the hearing, Congressman Gregory Meeks, the Democratic leader of the House Foreign Affairs Committee, criticized the management of American export policies. He argued that the control measures lost effectiveness when used as a bargaining tool in relations with China.
The legislator suggested that President Donald Trump turned export controls into a bargaining chip in broader negotiations with Beijing, questioning the decisions that granted licenses for the shipment of advanced AI chips.
Kessler defended the actions of the Department of Commerce, emphasizing that the priority is ensuring compliance with current restrictions. He also indicated that new regulations related to artificial intelligence and semiconductors may be announced in the future.
In addition to questions about authorized exports, lawmakers raised uncertainties about potential issues involving even more advanced Nvidia chips. Republican Congressman Bill Huizenga questioned the Department of Commerce guidelines regarding possible irregular acquisitions of Blackwell chips by Chinese subsidiaries located outside the country.
Kessler responded that companies receiving components without authorization must voluntarily report any violations. However, Huizenga insisted that the government should establish a more defined stance on retaining this equipment.
Historically, Nvidia has shown great interest in the Chinese artificial intelligence market, viewed as one of the largest centers of technological development. However, the company has faced commercial limitations imposed by the United States, which restricted some of its products destined for the Asian country.
The H200 belongs to Nvidia's Hopper generation and is considered an older model than the Blackwell chips, which offer superior performance. The permission to sell to China represents a chance to partially reverse the manufacturer's revenue in this market.
It is not yet determined whether China will permit large-scale purchases of these processors. If access to American chips remains restricted, Chinese companies may opt for domestically developed solutions, which are considered less advanced for training artificial intelligence systems.
