Malaysia is one of the few countries in the world running DeepSeek as part of a national AI service. So when the Chinese company signals it wants to stop leaning on other people's silicon and build its own chips, that is not distant news for Kuala Lumpur. It goes to the heart of what the country's sovereign AI stack will run on.
Editor
Kai T chevron_right
Table of Contents
What DeepSeek is actually planning
DeepSeek, the Chinese startup whose models compete with those from OpenAI and Anthropic, is moving into the chip business. According to a Reuters report picked up by Ars Technica, the company has spent around a year meeting hardware and silicon partners and hiring engineers for the effort. The focus is data-centre chips for inference, the stage where a trained model answers user prompts, rather than the far harder job of training new models from scratch.
The goal, per the reporting, is to cut its reliance on both Nvidia and Huawei. DeepSeek trained its R1 model on Nvidia's H800 processors, chips that were later banned from export to China. US export controls now block Chinese firms from buying Nvidia's most advanced accelerators, which makes a home-grown inference chip less a vanity project than a supply-chain insurance policy. DeepSeek is not alone: Alibaba and Baidu are chasing the same goal, Huawei already holds roughly half of China's data-centre chip market, and even in the US, OpenAI and Broadcom recently unveiled an in-house inference chip of their own.
Why Malaysia is caught in the middle
Malaysia has an unusually direct stake here. The Communications Ministry has stood up a Strategic Artificial Intelligence Infrastructure, a sovereign full-stack system that industry reports describe as the first national deployment of DeepSeek's model outside China, running on Huawei GPUs. At the same time, YTL Power switched on Malaysia's first Nvidia-powered AI data centre in Johor, built around Nvidia's Grace Blackwell hardware, part of a tie-up with Nvidia worth around 2.3 billion US dollars.
That leaves the country with a foot in each camp: American accelerators in Johor's commercial cloud, and a Chinese model on Chinese chips inside its national stack. If DeepSeek ships a cheap inference chip built to run its own models, the economics of that sovereign stack could shift again, and Malaysia would be one of the first markets outside China in a position to feel it.
The chokepoint problem
None of this happens in a vacuum. Malaysia introduced permit requirements for exports of high-end Nvidia AI chips in mid-2025, a move widely read as an attempt to shake off US scrutiny of the country as a transshipment route into China. Washington then tightened the screws further: on 31 May 2026 the US Commerce Department extended its controls to cover companies whose ultimate parent sits in China or Macau, aimed squarely at the Singapore and Malaysia subsidiaries that had been buying restricted Nvidia chips.
For a government that earmarked 5.9 billion ringgit for AI under Budget 2026 and wants to be Southeast Asia's data-centre hub, the chip layer is turning into a sovereignty question rather than a procurement one. A credible DeepSeek chip would give China's AI champions a route around US controls, and give countries like Malaysia a third option that is neither Nvidia nor Huawei.
The takeaway
DeepSeek making its own silicon is still a plan, not a product, and inference chips are hard to get right. But few countries have as much riding on how it turns out as Malaysia, which already runs the model and hosts both of the chip stacks it is trying to escape.
Images courtesy of Brian Kostiuk and Winston Chen on Unsplash.