Congress targets AI chip 'cloud loophole' in China export fight

The gist
Congress is racing to slam shut a billion-dollar 'cloud loophole' as Chinese tech giants sidestep U.S. AI chip controls by remotely tapping Nvidia GPUs overseas.
What to know
- Chinese firms like ByteDance and Tencent are accessing advanced Nvidia AI chips via Southeast Asian data centers, dodging U.S. export restrictions that don't cover remote cloud access.
- The Remote Access Security Act (RASA) aims to empower regulators to police remote access and force strict customer verification, winning strong bipartisan backing in Congress.
- Despite U.S. controls, an estimated 140,000 advanced chips worth $5–7 billion are expected to slip through global supply chains in 2024, fueling both enforcement headaches and China's domestic chip ambitions.
AI Loophole Powers China's Surge
Chinese tech giants are supercharging their AI ambitions by remotely tapping high-end Nvidia chips in Southeast Asian data centers, exploiting regulatory blind spots and fueling a 60% jump in compute access despite U.S. controls.
Chinese firms have adeptly exploited a critical loophole in U.S. AI chip export controls by remotely accessing Nvidia GPUs hosted in data centers across Southeast Asia, including countries like Thailand, Malaysia, Japan, and Singapore. This circumvention leverages the fact that U.S. export regulations, such as the Export Administration Regulations (EAR), focus on the physical export of semiconductor hardware and do not explicitly restrict remote access to these chips, enabling companies like ByteDance, Alibaba, and Tencent to harness advanced Nvidia GB300 and Blackwell B200 GPUs without violating existing laws. As Cassia King explains, "The U.S. export control regime controls physical AI chips. It does not cover remote access to those chips," highlighting a regulatory gap that has allowed offshore compute-rental arrangements to boost China's effective access to U.S. AI compute resources by an estimated 60% in 2026 alone.
The rapid expansion of AI data center infrastructure in Southeast Asia exacerbates this loophole, with JLL projecting global data center capacity could double to 200GW by 2030 and 31 planned mega data centers (100MW+) in Malaysia, Indonesia, and Thailand alone—up from just two today. This burgeoning infrastructure provides the physical foothold enabling Chinese firms to remotely rent cutting-edge Nvidia chips, effectively sidestepping U.S. export controls designed for a previous technological era. Michelle Nie warns that this gap "is threatening U.S. national security" because it undermines efforts to deny China the ability to train frontier AI models using advanced U.S. chips.
Beyond hardware access, Chinese military research units have further circumvented U.S. AI export controls through model distillation, a technique that creates smaller, portable AI models derived from advanced U.S. frontier AI systems like GPTs without requiring the original high-powered chips. This method effectively bypasses computational power constraints imposed by chip export restrictions, enabling China to advance military AI capabilities despite U.S. efforts. Experts emphasize that relying on private companies alone to police these transfers is insufficient, underscoring the need for stronger government intervention to address this enforcement gap.
In response to these evolving tactics, legislative efforts such as the Remote Access Security Act (RASA) seek to close the remote access loophole by granting regulators authority to oversee cloud-based compute rentals and enforce robust know-your-customer protocols. However, as Michelle Nie and Cassia King note, passing RASA is only the first step; effective enforcement will require detailed rulemaking to define the scope of covered compute, identify prohibited users, and implement practical compliance frameworks. Meanwhile, the U.S. Bureau of Industry and Security (BIS) is conducting a systematic review of legal cloud compute agreements, signaling a shift from reactive enforcement toward proactively adapting export controls to the realities of modern AI infrastructure and the intensifying U.S.-China AI competition.
Congress Closes the Cloud Gap
New bipartisan legislation like RASA and the MATCH Act aims to give regulators sweeping powers over remote chip access and third-country manufacturing, signaling a crackdown on the evolving tactics that have outpaced traditional export bans.
The Remote Access Security Act (RASA) represents a pivotal legislative effort to close the so-called 'cloud loophole' in U.S. export controls by empowering the Bureau of Industry and Security (BIS) to regulate remote access to controlled semiconductor items. Unlike current rules that do not restrict remote cloud access when no physical export occurs, RASA would extend BIS’s authority under the Export Control Reform Act to define the scope and specifics of remote access controls, potentially covering a broad range of items and end uses critical to national security. This regulatory expansion aims to address the growing trend of Chinese firms circumventing direct export bans by remotely accessing advanced AI chips hosted in third-country data centers, such as those in Southeast Asia.
By early 2026, RASA had garnered strong bipartisan support, passing the House of Representatives with a companion Senate bill sponsored by a bipartisan group of senators awaiting a vote. This momentum reflects congressional recognition of the urgent need to tighten export controls amid evolving tactics in the US-China AI competition. However, industry voices like Michelle Nie caution that cloud providers could face significant compliance burdens due to mandated know-your-customer (KYC) and customer verification requirements, highlighting the challenge of balancing national security with operational feasibility.
Beyond RASA, broader congressional initiatives such as the MATCH Act, slated for inclusion in the National Defense Authorization Act (NDAA), seek to strengthen export controls by targeting semiconductor manufacturing equipment—including deep-ultraviolet immersion lithography tools critical for AI chip production—and enhancing enforcement mechanisms. These legislative moves aim to close gaps exploited through third-country manufacturing and remote access, with the U.S. government poised to negotiate with allied nations or impose unilateral restrictions to sustain AI compute leadership over China. The Bureau of Industry and Security, backed by White House support, could swiftly implement new rules post-RASA enactment, underscoring the administration’s commitment to robust enforcement.
Export Controls Face Global Evasion
U.S. efforts to restrict AI chip flows are stymied by legal loopholes, international smuggling, and industry pushback—while Nvidia warns that these controls may ultimately accelerate China’s homegrown chip industry.
The U.S. export control strategy on AI chips walks a tightrope between maintaining American technological leadership and managing market access, exemplified by allowing Nvidia to sell mid-tier H200 chips to China while blocking the latest Blackwell generation. This calibrated approach aims to slow China's AI progress without ceding the entire $50 billion Chinese market, prompting Nvidia and TSMC to ramp up production capacity in anticipation of regulatory approvals, underscoring the high stakes and complex negotiations shaping U.S. policy.
Enforcement of AI chip export controls faces significant hurdles due to porous supply chains and legal gray zones, as evidenced by Taiwan’s inability to prosecute direct chip exports to China and the estimated smuggling of 140,000 advanced chips worth $5 to $7 billion in 2024 alone. Smugglers increasingly reroute shipments through Southeast Asia and Japan, while cases like the Supermicro indictment reveal how U.S.-listed companies exploit shell firms to circumvent controls, highlighting the urgent need for stronger international cooperation and regulatory frameworks.
Nvidia’s CEO Jensen Huang openly criticizes U.S. AI export controls as a strategic failure that inadvertently accelerates China’s indigenous chip development and pushes Chinese buyers outside the American hardware ecosystem, reflecting the compliance challenges faced by companies caught between regulatory demands and market realities. Nvidia acknowledges its limited ability to police third-party circumvention, illustrating enforcement gaps that complicate the broader U.S. effort to contain China’s AI ambitions.
While U.S. export controls have slowed China’s access to cutting-edge AI chips and forced AI training to occur in international data centers, this has simultaneously incentivized China’s domestic semiconductor innovation and strategic investments in low-cost energy infrastructure to support large-scale compute. This dynamic creates a paradox where restrictions delay immediate progress but catalyze long-term self-reliance, complicating the geopolitical landscape and underscoring the multifaceted nature of the U.S.-China AI competition.








