TikTok’s owner is looking for AI power where chip bans are harder to police
ByteDance Looks for Blackwell Compute Through Malaysia📷 AI-generated image / TECH&SPACE
- ★ByteDance reportedly plans to use about 500 Nvidia Blackwell systems in Malaysia with roughly 36,000 B200 chips.
- ★US export controls still block direct Blackwell GPU access in China, according to the available report.
- ★Malaysia is becoming a sensitive AI infrastructure node linking data centers, cloud access, and regulatory scrutiny.
ByteDance is not getting Nvidia’s newest AI hardware into China directly. According to The Decoder’s report, TikTok’s parent company plans to use a Malaysia-based cluster built around roughly 500 Nvidia Blackwell systems and about 36,000 B200 chips.
That is the practical significance here: the restriction is on direct access inside China, not necessarily every cloud or data center arrangement outside it. The reported setup would cost more than $2.5 billion, putting it in the same strategic category as the infrastructure bets now separating serious AI players from companies merely renting burst capacity.
US export controls still prohibit Nvidia Blackwell GPUs from being shipped to China, and recent Trump administration relaxations reportedly do not include these chips. ByteDance’s position, as cited in the report, is that it “adheres fully to all applicable export control regulations,” which is exactly where the dispute becomes more interesting than a simple smuggling story.
The reported 36,000-chip Nvidia B200 cluster shows how the AI race is shifting into cloud geography
A closer operational view of secured GPU racks, access-control terminals and fiber paths, emphasizing cloud access rather than chip shipment.📷 AI-generated image / TECH&SPACE
The source material also shows that for users, the immediate effect is indirect but real. More high-end compute gives ByteDance more room to train recommendation models, generative tools, ad systems, and internal AI products, even if the company has not detailed the cluster’s exact workload. It appears aimed at AI research and development rather than ordinary web hosting.
Malaysia matters because it sits at the intersection of cheap power ambitions, data center growth, and geopolitical scrutiny. The country introduced a licensing requirement for US high-performance chips in 2025, a signal that local authorities understand they are now part of the enforcement chain, not just landlords with fiber and cooling.
The uncomfortable truth is that export policy increasingly has to police access, not just boxes crossing borders. If companies can legally build controlled-country-adjacent compute offshore, then the competitive map shifts from chip sales to cloud geography. The real signal here is that AI capability is becoming a question of where compute is reachable, not merely where a company is headquartered.
