Malaysia AI, Compute / AI news for Malaysia
UTM launches Blackwell GPU supercomputer it calls a first for Malaysian public universities
Universiti Teknologi Malaysia says its new GPU facility gives students and researchers industry-grade computing. The figures describe peak speed, the cost is not itemised, and rules for outside access are not yet published.
In brief
- Universiti Teknologi Malaysia (UTM) launched a high-performance GPU computing facility at its Faculty of Artificial Intelligence on 2 October 2026 and describes it as a first among Malaysian public universities.[1][2]
- UTM reports up to 72 PFLOPS for training and 144 PFLOPS for inference on NVIDIA Blackwell hardware, plus 120 AI workstations for students and researchers.[2][1]
- The facility is part of a RM20 million Ministry of Higher Education allocation under Budget 2024. UTM says other institutions can benefit, but publishes no access rules.[2][1]
What UTM launched
Universiti Teknologi Malaysia has launched a GPU computing facility at its Faculty of Artificial Intelligence in Kuala Lumpur. The launch took place on 2 October 2026 during the Vice-Chancellor's joint assembly at Dewan Azman Hashim, according to UTM's Malay-language release.[1]
UTM calls it the first ultra-high-speed GPU supercomputing facility among Malaysian universities. That is the university's own claim, and neither of its two releases cites an independent ranking.[1][2]
What the facility contains
UTM's English release describes a central system built on NVIDIA Blackwell GPUs with 1,440 GB of HBM3e memory, two Intel Xeon processors with more than 100 cores, 2 TB of DDR5 memory and InfiniBand networking. It adds 30 TB of NVMe storage and 100 TB of shared storage.[2]
A second tier is the UTM AI Innovation Lab. It holds 120 desktop workstations, each with an Intel Core Ultra 7 processor, a GeForce RTX 5070 GPU and 1 TB of storage. UTM says 120 students and researchers can build and test models at the same time, then send larger jobs to the central system.[2]
How to read the performance figures
UTM quotes peak speeds of up to 72 PFLOPS for training and 144 PFLOPS for inference, and says the system can run models of up to one trillion parameters. A PFLOPS figure is a ceiling that depends on the number format used, not the speed a real research job will reach.[2][1][3]
Our reading is that the central system is one NVIDIA DGX B200. NVIDIA's page for that product lists eight Blackwell GPUs, 1,440 GB of GPU memory, and 144 PFLOPS at FP4 and 72 PFLOPS at FP8 precision, which match the two numbers UTM quotes. UTM's English release also captions a photograph Nvidia DGX B200.[3][2]
Who gets access, and what is missing
UTM calls the facility a shared national resource. Its English release says researchers from UTM and other higher education institutions can benefit, and its Malay release points to wider collaboration with industry and government agencies. Neither release says how outside users apply, what they pay or how jobs are scheduled.[2][1]
Those rules decide the real value. A single eight-GPU system can serve only a limited number of large jobs at once. Our reading is that serving a trillion-parameter model is a capacity claim, while training one from scratch usually needs far more hardware than one server.[2][3]
Funding and Malaysia's AI talent push
UTM says the facility is part of a RM20 million investment by the Ministry of Higher Education under Budget 2024. The money was allocated to establish the Faculty of Artificial Intelligence, which UTM describes as Malaysia's first AI faculty. UTM does not say how much of the RM20 million paid for this hardware.[2][1]
The Malay release lists health care, smart manufacturing and systems built on local languages as target uses. It also says the aim is to ease the processing-power limit and produce AI graduates ready for industry. Those are goals. No projects or results have been announced.[1]
Why Malaysia should care
UTM itself cites processing power as a limit on training large AI models at Malaysian universities. This facility adds local capacity and a student lab, but access terms, usage figures and costs are unpublished.
UTM students and researchers
Hands-on access to current Blackwell hardware, per UTM.[2]
Practical move: Ask the Faculty of AI how course projects get GPU time.
Other universities and research groups
UTM calls it shared, but gives no application route.[2][1]
Practical move: Ask UTM about eligibility, queue rules and fees first.
Malaysian employers hiring AI talent
Graduates may train on current hardware, but no outcome data exists.[1]
Practical move: Treat it as a pipeline signal and test skills yourself.
What Malaysians can do now
- Researchers at other institutions: email UTM's Faculty of Artificial Intelligence and ask whether outside access exists, who approves it and what it costs.
- Employers: treat the facility as a talent-pipeline signal. Ask candidates what they trained or ran, not only which hardware they used.
- Read PFLOPS as a peak ceiling. When comparing GPU offers, ask for the number format, memory, queue time and price per hour.
What we still do not know
What we still do not know
- How many systems UTM installed, and what this hardware cost inside the RM20 million.
- Who outside UTM may use it, how jobs are queued and whether fees apply.
- Which research projects will run first, and when UTM will publish usage or results.
Sources
- 1.UTM universiti awam pertama miliki infrastruktur HPC GPU termaju, perkukuh ekosistem AI nasional Universiti Teknologi Malaysia, 2 October 2026
- 2.UTM establishes Malaysia's first ultra-high-speed university GPU supercomputer Universiti Teknologi Malaysia, 5 October 2026
- 3.NVIDIA DGX B200: The Foundation for Your AI Factory NVIDIA
- 4.File:Kuala Lumpur Malaysia UTM-Campus-Kuala-Lumpur-01.jpg Wikimedia Commons, 6 June 2014
- 5.File:Nvidia DGX-B200-HGX.jpg Wikimedia Commons, 7 September 2025
- 6.File:Racks Amravati Data Center.jpg Wikimedia Commons, 17 August 2021
