AI computing power has become an unavoidable part of "great power competition": The United States is concerned about chip outflow, and the UAE's massive NVIDIA orders have been put on hold

According to media reports citing informed sources, some core officials of the Trump administration in the United States are delaying a purchase agreement that would allow UAE technology companies and government-supported giants to spend billions of dollars on NVIDIA's (NVDA.US) most advanced Blackwell architecture AI chips due to national security concerns.

It is understood that U.S. President Donald Trump announced this preliminary purchase agreement during his visit to the Middle East in May this year, and both parties originally hoped to quickly finalize the details. However, as of now, the two countries have not reached a final agreement, partly because some U.S. officials are concerned that competitors like China may take the opportunity to acquire these advanced American high-performance chips.

Media reports citing informed sources indicate that unless the UAE agrees to modify the terms to address the national security concerns of U.S. officials, it is difficult to resolve the differences.

One of the options being discussed by officials is to revoke the permission for the Abu Dhabi-based AI startup G42 to directly acquire or use NVIDIA's AI chip clusters. Under the original plan, G42 was expected to receive about 20% of NVIDIA's AI chip allocation.

Currently, the U.S. Department of Commerce does not intend to approve the supply of any AI chips to G42, although there is still a possibility of future approval, the media reported.

NVIDIA, the U.S. Department of Commerce, G42, OpenAI, and Microsoft have not yet responded to any requests for comments.

A spokesperson for U.S. Secretary of Commerce Gina Raimondo stated that Raimondo "is confident that the implementation plan for the agreement signed with the UAE will proceed smoothly."

The UAE side is also optimistic about the negotiations related to trade and the AI chip purchase agreement with the U.S. government. UAE Ambassador to the U.S. Yousef Al Otaiba stated in a statement that this large-scale purchase agreement "will bring significant benefits to both countries."

On Tuesday, Raimondo met with UAE representatives at a technology and energy conference in Pittsburgh to discuss how to advance the agreement.

However, the slow progress of negotiations has led to significant divisions within the U.S. government and has frustrated some tech industry executives who hope to expand AI-related businesses overseas.

Reports indicate that NVIDIA CEO Jensen Huang emphasized the importance of the agreement to President Trump during a recent meeting.

Some government officials are also concerned that certain Chinese tech companies are trying to sell AI chips in the Middle East and may seize the market during the delay of the agreement.

The core debate of the agreement revolves around whether to directly acquire NVIDIA's most advanced Blackwell architecture AI chips and the NVL server clusters based on that architecture, which will provide the strongest computing power support for large-scale AI data centers globally.

Last month, media reports also indicated that the construction progress of the "Stargate UAE" AI infrastructure project, dubbed the Middle Eastern version of Stargate, which involves billions of dollars in AI data centers, has been delayed indefinitely due to ongoing national security concerns. This project is being jointly developed by NVIDIA, Oracle, OpenAI, Cisco, SoftBank Group, and G42 from the UAE, and the large-scale project was announced after Trump's visit to the UAE OpenAI and G42 are both supported by Microsoft's cloud computing services and cloud AI computing resources.

According to preliminary terms, UAE companies will receive hundreds of thousands of NVIDIA chips over the years and assist NVIDIA and American cloud computing giants in building more and larger AI data centers. According to the agreement, most AI chips will be used in data centers in the Middle East operated by American technology companies.

OpenAI and American tech giants like Microsoft are expected to operate multiple large AI data centers in the UAE. Insiders pointed out that exporting NVIDIA AI chips to the UAE, whether delivered to UAE companies or American tech giants, requires export licenses approved by multiple departments, which may intensify national security scrutiny.

In addition to receiving about 20% of NVIDIA AI chips, UAE AI startup G42 may also be deeply involved in the construction of data centers operated by American cloud computing giants in the UAE.

Some American officials are concerned that countries like China may gain access to advanced AI chips through G42 or other entities based in the UAE, thereby obtaining the critical technologies needed to develop and iterate AI large models.

Directly canceling the supply to G42 could anger the UAE government; the UAE views this agreement as key to maintaining a leading position in the global AI race and believes G42's involvement is central to the framework agreement reached in May.

David Sachs, head of AI at the White House, stated at the Pittsburgh summit: "If we don't provide this technology, global competitors will." He believes concerns about chip transfers are "greatly exaggerated."

Sachs is one of the principal architects of the US-UAE AI chip agreement, while Lutnik facilitated the signing of the preliminary agreement. Reports add that after the agreement was signed, key officials from both countries responsible for advancing the details have met very little.

Global AI Computing Power Demand Continues to Surge

As the US government delays NVIDIA AI chip exports to the UAE due to national security concerns, the demand for AI computing power continues to show explosive growth. Pennsylvania, the most typical "swing state" in the US, will receive $92 billion in investments to build a "super AI center." Meta, the parent company of Facebook, plans to invest hundreds of billions of dollars to build several large-scale AI data centers to support its AI technology updates, aiming to achieve artificial general intelligence (AGI), with the first data center expected to be operational next year.

Global cloud computing giant Amazon AWS announced a scaled-up cooperation agreement with South Korea's SK Group, under which AWS will build the largest data center in Korea with a capacity of up to 100 megawatts (MW). Additionally, this week, "AI chip leader" NVIDIA announced the resumption of sales of its H20 AI GPU in the Chinese market.

Meta CEO Mark Zuckerberg stated on Monday that the tech giant will invest hundreds of billions of dollars to build several large data centers to support its AI development, aiming to achieve artificial general intelligence (AGI), with the first super AI data center expected to be operational next year. Wall Street analysts generally interpret Zuckerberg's statement as a strong signal of confidence in significant mid-term cash flow growth and a strong indication that the demand for AI computing power centered around NVIDIA AI GPUs remains incredibly strong and that the AI capital expenditure cycle is far from over The generative AI leader Anthropic, known as the "OpenAI rival," predicts that by 2027, AI large models will have the capability to automate almost all white-collar jobs. Therefore, the demand for AI computing power brought by inference is described as "starry sea," and is expected to drive the artificial intelligence computing infrastructure market to continue showing exponential growth. The "AI inference system" is also considered by Jensen Huang to be the largest source of future revenue for NVIDIA.

TSMC's latest performance report released on Thursday shows a surge in AI computing power demand, driving TSMC's net profit in Q2 to increase by 61%. TSMC expects that sales in 2025, calculated in US dollars, will grow by about 30%, higher than the previous expectation of "close to 20% in the mid-range," mainly due to the continuous surge in AI chip orders based on 3nm and 5nm advanced process technologies. As the demand for AI computing power remains incredibly strong, TSMC is actively expanding its backend capacity to enhance the actual output of CoWoS advanced packaging, primarily used for NVIDIA's AI GPU capacity. This also indicates the company's confidence that the strong demand for AI chips will continue until 2026