Tencent Uses Offshore Data Centers to Access Nvidia Blackwell GPUs Despite US Export Bans
Tencent, one of China’s largest tech firms, has secured access to Nvidia’s latest Blackwell (B200) GPUs through a series of offshore arrangements, circumventing US restrictions on high-performance chip exports.
The company has been operating via Japanese and Australian data centres to continue training advanced AI models that require massive computational power.
How Tencent Circumvents Export Restrictions
The Financial Times reported that a data centre near Osaka, operated by Japanese marketing solutions and AI infrastructure firm Datasection, effectively serves Tencent.
The facility houses around 15,000 Nvidia Blackwell GPUs, secured through a $1.2 billion contract via a third-party entity.
Datasection is also planning to expand its operations to Sydney, Australia, with over 100,000 GPUs, primarily for Tencent’s use.
While the US permits the export of Nvidia’s older H200 GPUs to China, Blackwell chips remain off-limits.
President Donald Trump previously stated,
“No one outside the US will possess the most advanced chips (Blackwell).”
However, Chinese firms are exploiting loopholes, particularly the absence of restrictions on cloud-based usage of GPUs installed overseas.
Alibaba, ByteDance, and other AI developers are also using data centres in Southeast Asia to train large language models.
Beyond rentals, some firms have resorted to physically disassembling GPU servers abroad and transporting components back to China.
DeepSeek, known for its low-cost AI systems, is one such example, moving chips from Southeast Asian facilities for domestic AI development.
Why Offshore AI Training Remains Essential
The Chinese government maintains tight restrictions on Nvidia GPUs within domestic data centres, even limiting imports of the H200 series to encourage local chip development.
However, industry sources acknowledge that domestically made chips still lag behind Nvidia’s offerings in computational power.
One source noted,
“While China is fiercely pursuing semiconductor self-reliance, it is undeniable that Nvidia chips remain unmatched for AI training requiring massive computational power.”
By leveraging offshore facilities, Chinese firms gain access to high-performance GPUs without violating government directives.
Datasection’s Rapid Expansion and Strategic Role
Datasection, which pivoted from marketing solutions to AI infrastructure in 2023, has become a crucial partner for Tencent.
Their Osaka data centre, built through a contract mediated by Tokyo-based NowNaw, reportedly contains 15,000 Blackwell GPUs and additional acquisitions of 5,000 advanced chips costing $272 million under a $406 million contract with a major cloud firm.
CEO Norihiko Ishihara emphasised the pace of growth:
“Less than half a year ago…5,000 B200 chips were sufficient to support AI models, but that the floor has doubled and 10,000 should be the minimum requirement.”
The company is also constructing a Sydney data centre under an $800 million, three-year deal, aiming to deploy tens of thousands of Nvidia B300 chips.
Ishihara described it as a hyperscale AI cluster capable of handling enormous computational workloads.
Geopolitical Chessboard and Legal Loopholes
The US has tightened export restrictions citing national security concerns, particularly the potential military or surveillance use of advanced AI.
Lower-tier Nvidia chips have been allowed, but top-tier Blackwell GPUs remain restricted.
Tencent maintains that its offshore cloud-based GPU usage is “both transparent and legal.”
Analysts suggest that renting compute power abroad has become the preferred method for Chinese firms seeking access to high-performance chips.
Bernstein Research analyst Lin Qingyuan said,
“Renting compute abroad may be the more attractive choice for Chinese tech groups.”
Financial and Regulatory Risks for Datasection
Datasection faces challenges despite its rapid growth.
Its stock surged 185% in 2023 before tumbling due to overextension concerns, short seller attacks, and complex financial arrangements with Singapore-based First Plus Financial Holdings, owned by a Chinese national.
The company is raising ¥50 billion through warrants to First Plus, potentially diluting existing shareholders by up to 200%.
In addition, future US policy changes could disrupt offshore operations.
Ishihara downplayed potential interruptions, stating,
“In a worst-case scenario, we may have to stop the operation for, let’s say, one week… [the asset is] very sexy.”
Even short downtimes could have significant consequences in the high-speed AI arms race.
Implications for AI and Crypto Industries
The same GPUs powering Tencent’s AI ambitions are crucial for cryptocurrency mining and blockchain infrastructure.
High-performance GPUs accelerate Bitcoin mining and enable faster transaction processing for networks like Ethereum.
They also have potential for AI-driven smart contracts and other decentralised applications.
However, their use in centralised AI systems raises ethical concerns, as they could reinforce surveillance and authoritarian control.
While Tencent’s offshore GPU strategy allows it to stay competitive despite US restrictions, it does not represent true decentralisation.
The approach relies on foreign infrastructure and is vulnerable to geopolitical pressures.
For the crypto community, this story highlights the tension between technological sovereignty and centralised control, reflecting challenges that also affect decentralised networks.
Two-Track Strategy Signals China’s AI Ambitions
Chinese tech firms are simultaneously pursuing self-reliance in domestic chip development and exploiting overseas compute resources to maintain AI leadership.
This “two-track strategy” allows them to continue advancing AI capabilities while working within government restrictions, illustrating the complex interplay between technology, geopolitics, and strategic innovation.