Who in Southeast Asia Actually Buys a Supercomputer?
You can find an HPE Cray supercomputer listed on a Singapore reseller's website, including ours. That raises a fair question. Who, in this region, actually buys one?
The answer is a short list. It is worth knowing who is on it, because for nearly everyone else the smarter move is to use a supercomputer, not own one.
The buyers, in one table
| Buyer type | Why they buy | Regional example |
|---|---|---|
| 🏛️ National supercomputing centres | Shared research capacity for a whole country | NSCC Singapore (ASPIRE 2A, 2A+, 2B); Thailand's ThaiSC (LANTA) |
| 🎓 Universities | Research computing under their own control | National University of Singapore, whose system entered the TOP500 |
| 🌦️ Weather and climate agencies | Forecasting and climate projection | Singapore's climate research centre, as a user of national systems |
| ☁️ AI cloud and telecom providers | Selling GPU capacity to others | Commercial AI clouds now appearing in regional rankings |
| 🏭 Large industry | Simulation for energy, chips, pharmaceuticals, finance | Usually smaller clusters, rarely publicised |
Case study: what Singapore's newest supercomputer is made of
On 8 June 2026, Singapore launched ASPIRE 2B, its newest national research supercomputer. The National Supercomputing Centre says it ranked 44th on the TOP500 list and delivers about four times the combined capacity of its predecessors, ASPIRE 2A and 2A+.
| ASPIRE 2B part | Built from (per TOP500) | Job |
|---|---|---|
| 🧠 GPU partition | HPE ProLiant Compute XD685 servers, AMD EPYC processors, NVIDIA H200 GPUs, Slingshot interconnect | AI training and inference |
| ⚙️ CPU partition | HPE Cray EX, AMD EPYC processors, Slingshot interconnect | Simulation and modelling |
Sources: NSCC Singapore: 10 years with NSCC, TOP500: National Supercomputing Centre Singapore
Two details stand out for anyone shopping in this market.
The national AI machine uses servers you can buy
The GPU partition is not some unobtainable custom design. It is built from the HPE ProLiant Compute XD685, an 8-GPU server that appears in HPE's ordinary catalogue. What makes it a supercomputer is how many there are, the interconnect between them, the storage behind them and the team running it.
It was built on the previous GPU generation
ASPIRE 2B uses NVIDIA H200 GPUs, a generation behind the Blackwell family that dominates AI headlines. A national programme with a large budget chose proven hardware it could deploy and support, not the newest part on the roadmap. That is a useful counterweight for anyone who feels they must have the latest chip.
What these machines are used for
| Use | Example |
|---|---|
| 🌏 Climate | Singapore's national climate study used NSCC systems to sharpen global climate models from 100 km resolution to 2 km over Singapore |
| 🗣️ Regional AI models | Southeast Asia-focused language and multimodal models, such as A*STAR's MERaLiON project |
| 🧬 Science at large | ASPIRE 2A and 2A+ supported more than 1,500 projects |
Sources: HPCwire on Singapore's national supercomputing investment, The Straits Times via Asia News Network on ASPIRE 2B
It is not only Singapore
Thailand's national supercomputer, LANTA, is an HPE Cray EX system with 704 NVIDIA A100 GPUs, operated by the NSTDA Supercomputer Center. Across the region the pattern repeats: a national body buys, and researchers and companies apply for time.
Source: NSTDA Supercomputer Center presentation, ISGC 2026
Three uncomfortable observations
1. The buyers are almost all public bodies
Supercomputers are, overwhelmingly, bought with public money for shared use. Singapore committed S$270 million to its national supercomputing infrastructure in 2024. That scale of funding and the mandate to serve a whole research community are what justify the purchase. A single company rarely has either.
2. "Sovereign AI" is partly about pride
Every country in the region now wants its own AI compute. Some of that is sound: local language models, sensitive data and resilience are real needs. Some of it is prestige, and a TOP500 ranking is a very visible trophy. The test is the same as for any company: is the machine full of useful work?
3. For most organisations, access beats ownership
If your simulation or AI project needs supercomputer-class capacity for weeks, not years, the rational route is to apply for time on a national system or rent from a commercial provider. Check your national centre's access schemes before budgeting for hardware. Owning makes sense only when the demand is continuous.
Should you buy one?
| Your situation | Sensible route |
|---|---|
| 🏛️ National or institutional mandate, many users, multi-year funding | ✅ Buy: Cray Supercomputing class |
| 🎓 University department or research institute | 🟡 A cluster of 8-GPU or HPC servers, plus national-centre time for peaks |
| 🏭 Company with continuous simulation workload | 🟡 A right-sized cluster |
| 🧪 Occasional large jobs | ❌ Do not buy. Apply for time or rent. |
The takeaway from ASPIRE 2B is encouraging for smaller buyers. The building blocks of a national supercomputer are ordinary catalogue servers. You can start with two of them and grow, using the same technology the national centre does.




