In May 2025 Malaysia's deputy communications minister announced a national AI system running on 3,000 Huawei Ascend chips, and her office pulled the remarks back almost immediately without explaining why [20]. That retraction is the more useful precedent here. The decision has been live long enough to be blurted out once and buried.
Run the retracted figure against the current budget and the shape of the programme appears. Two billion ringgit spread across 3,000 accelerators is about 667,000 ringgit apiece [22], which no accelerator costs. The chip count for the current programme has not been disclosed [3] and the 3,000 belongs to a different announcement, so treat the division as a sanity check rather than a forecast. It still tells you what the money is: a facility with power, building, integration and staff in it, of which silicon is one line. Malaysia already has a large American-built data centre estate [9], and this programme adds a vault to that estate rather than more capacity.
That is the part US vendors will struggle to bid against. "I do not consider that law reasonable, but how does one argue with President Trump?" Anwar Ibrahim told university students in April, in remarks Bloomberg reported [12]. A region setting in a console does not answer a subpoena served on a vendor's parent in a US court [11], and neither does a discount. Washington's public reply stays on supplier trust: a State Department spokesperson told Bloomberg that secure technology is the foundation of a prosperous economy and that untrusted suppliers can be manipulated or controlled by authoritarian regimes, without naming China or Huawei [15]. The private objection is narrower, and US officials have told their Malaysian counterparts they are worried about Huawei equipment handling intelligence material, including material shared by foreign governments [16].
What the reporting does not show is a purchase. Bloomberg's account rests on people familiar with the matter who were not authorised to speak publicly [2], nothing is signed [1], and Telekom Malaysia, Huawei, the Prime Minister's Office and the Ministry of Digital all stayed silent, as did Nvidia and AMD [14][19]. The Next Web, which carried the report, has not independently verified it [4]. The part named is Huawei's marquee 910C, not the newer Ascend 950 line still in limited production [5]. The extraterritorial guidance has yet to be tested, though the opportunity to do so is now on the table [7].
For anyone choosing where a workload lives, this reduces to two questions, and only one pair of answers points at the slower chip: whether the data carries an obligation that survives a subpoena to a vendor's parent, and whether the workload can tolerate accelerators that still trail Nvidia's by a wide margin [8]. Obligated data plus tolerant throughput is the single quadrant where Huawei wins on merit; the other three fail on speed. Telekom Malaysia is already living on both sides of that line, running Huawei technology in its cloud infrastructure while the AI computing service it sells today uses Nvidia chips [13].
If the second tender process lands, the likely outcome is a split estate rather than a single winner [18]: classified work on Ascend, commercial work on Nvidia, and two of every operational practice for whoever has to keep both racks patched.
