- What: ByteDance is accessing a massive AI cluster of 36,000 Nvidia B200 Blackwell GPUs.
- Where: The cluster is physically located in Malaysia, operated by Aolani Cloud.
- Scale: The deployment includes 500 Nvidia GB200 NVL72 rack-scale systems valued at approximately $2.5 billion.
- Compliance: Nvidia confirmed the deal does not violate U.S. export controls as the hardware remains outside of mainland China.
ByteDance, the parent company of TikTok, has secured access to a colossal artificial intelligence cluster in Malaysia featuring 36,000 of Nvidia’s flagship Blackwell B200 GPUs. This $2.5 billion infrastructure play allows the Chinese tech giant to leverage the world's most advanced AI hardware for research and development while remaining in strict compliance with U.S. export regulations that currently bar such chips from entering mainland China.
The deal, first reported by the Wall Street Journal and confirmed by industry sources, represents one of the most significant infrastructure expansions in Southeast Asia to date. By leasing compute power through a third-party provider in a non-restricted jurisdiction, ByteDance is effectively sidestepping the geographic limitations of the U.S. Department of Commerce’s Bureau of Industry and Security (BIS) while maintaining its trajectory in the global AI arms race.
The $2.5 Billion Blackwell Architecture
The heart of the new cluster consists of 500 Nvidia GB200 NVL72 rack-scale systems. Each NVL72 system is a liquid-cooled rack that connects Grace CPUs and Blackwell GPUs to act as a single, massive GPU. With 36,000 B200 GPUs in total, the cluster provides ByteDance with a level of computational "oomph" that rivals the most powerful data centers in the United States.
The hardware is being supplied through Aivres, a specialized server manufacturer that builds systems based on Nvidia’s high-end accelerators. While ByteDance is the primary user, the hardware is formally owned and operated by Aolani Cloud, a Malaysia-based provider established in late 2023. Aolani operates under a Cayman Islands holding structure and has been designated a Tier-1 Nvidia cloud partner, granting it priority access to the latest Blackwell and Hopper architectures.
The scale of this expansion is unprecedented for Aolani Cloud. According to a company spokesperson, the provider currently operates with roughly $100 million worth of hardware. The jump to a $2.5 billion Blackwell cluster represents a 25-fold increase in infrastructure value, though it remains unclear who is providing the primary funding for the massive capital expenditure.
Navigating U.S. Export Controls
The legality of the arrangement hinges on a critical distinction in U.S. trade law: the regulations focus on the physical destination of the hardware rather than the nationality of the entity leasing the compute power in the cloud. Because the B200 GPUs are being shipped to and housed in Malaysia—a country that typically does not require a BIS export license—the transaction is deemed legal under the current framework.
"By design, the export rules allow clouds to be built and operated outside controlled countries," an Nvidia spokesperson told Tom's Hardware. "All Nvidia cloud partners are evaluated and cleared by Nvidia’s field operations, finance, and compliance teams before they can receive our products."
Nvidia has emphasized that ByteDance is not currently on the U.S. government’s Entity List or Military End Use (MEU) list. Therefore, providing services to the company in a neutral territory like Malaysia does not trigger the "red flags" that would otherwise stop a shipment to a mainland Chinese data center.
Strategic Shift to Southeast Asia
ByteDance’s reliance on Malaysian infrastructure is not a new experiment. The company has reportedly been leasing AI servers equipped with Nvidia H100 (Hopper) GPUs from Aolani since February 2025. This earlier Hopper-based cluster likely served as a technical proof-of-concept, ensuring that ByteDance’s engineering teams could successfully train models on remote clusters without regulatory interference.
The Blackwell deployment is part of a broader Southeast Asian strategy for ByteDance. In addition to the 36,000-GPU cluster in Malaysia, reports indicate the company is considering a second deployment of over 7,000 B200 GPUs at a data center in Indonesia. Initial payments for the Blackwell hardware in Malaysia have reportedly already been processed.
Industry Impact and Competitive Context
This move signals a shift in how global AI players navigate geopolitical tensions. For developers and the wider AI industry, it proves that "compute sovereignty" is becoming increasingly fluid. If a company has the capital, it can access state-of-the-art silicon by moving its workloads to jurisdictions where the hardware is permitted.
For Nvidia, the deal is a vital revenue driver that keeps it competitive in the Asian market. The company warned that overly restrictive controls could cede ground to foreign competitors. As the Nvidia spokesperson noted: "Winning the business of those clouds will bring tens of billions of dollars and high paying jobs home. America cannot afford to make the same mistake across all of Asia."
For ByteDance, this access is critical for the development of next-generation large language models (LLMs) and recommendation algorithms that power its global apps. By securing 36,000 Blackwell GPUs, ByteDance ensures it is not left behind by U.S.-based rivals like Meta or Google, who are also deploying Blackwell at scale.
What’s Next
The deployment of the 500 NVL72 racks is expected to proceed in phases, with ByteDance likely moving its most intensive R&D tasks to the Malaysian site as the hardware comes online. However, the deal may draw increased scrutiny from U.S. lawmakers who have expressed unease regarding Chinese firms accessing high-end American AI compute through cloud "loopholes."
While the current framework permits this activity, the Department of Commerce has the authority to update rules at any time. For now, Malaysia is positioning itself as a primary beneficiary of the tech cold war, emerging as a critical hub for global AI infrastructure.