Global AI, Malaysia Angle / AI news for Malaysia
AWS and NVIDIA plan two million more AI GPUs
The two companies plan to add two million NVIDIA GPUs across AWS during 2027 and 2028, after a one million GPU commitment that demand has already outrun. Malaysia's cloud customers are watching the capacity, pricing and sovereign options.
In brief
- AWS and NVIDIA plan to add two million NVIDIA Blackwell Ultra, Rubin and Rubin Ultra GPUs across AWS infrastructure in 2027 and 2028, building on a more than one million GPU commitment made at GTC 2026.[1]
- The partnership adds NVIDIA Vera CPUs, NVLink Fusion with custom NVHBM memory designed with Amazon's Annapurna Labs, and faster EC2 instances built on the Nitro System and Elastic Fabric Adapter.[1][2]
- NVIDIA's Nemotron open models come to Amazon Bedrock and SageMaker, while AWS data, search and robotics services adopt NVIDIA acceleration.[1]
AWS and NVIDIA agreed a deeper AI infrastructure partnership
AWS and NVIDIA say demand for AI compute is running ahead of every forecast. On 26 August 2026 the two companies announced a deeper partnership that plans two million additional NVIDIA GPUs across AWS global infrastructure during 2027 and 2028.[1]
The plan builds on a commitment of more than one million NVIDIA GPUs made earlier at GTC 2026. AWS chief executive Matt Garman said customers want the freedom to choose the best tools for their AI workloads, while NVIDIA chief executive Jensen Huang described the pairing as one of the great growth engines of the AI era.[1]
The announcement is not only about GPUs. It pairs NVIDIA Vera CPUs, NVLink Fusion networking and custom NVHBM high-bandwidth memory with AWS's own Nitro System and Elastic Fabric Adapter, the foundation AWS uses for its most demanding workloads.[1][2]
Two million GPUs is a capacity signal, not a price cut
Two million is the headline number, but it is a capacity plan rather than a product you can buy today. The GPUs span NVIDIA's upcoming Blackwell Ultra, Rubin and Rubin Ultra generations and are scheduled for 2027 and 2028, so the effect on pricing and availability will only show up over the next two years.[1]
AWS is also reserving 100,000 GPUs on secure infrastructure for United States federal and national security workloads, classified at Impact Level 6 and above. The announcement makes no equivalent commitment for Malaysia or any other Asia Pacific country, which is the detail Malaysian cloud buyers should note.[1]
Custom memory and networking tie the systems together
The partnership extends NVIDIA's NVLink Fusion interconnect with a custom high-bandwidth memory called NVHBM, developed with Amazon's Annapurna Labs, to lift Trainium performance. NVIDIA Vera CPUs also come to AWS for the first time.[1][2]
Every NVIDIA GPU and Trainium-based EC2 instance will run on the AWS Nitro System and connect through Elastic Fabric Adapter. In practice the raw GPU count matters less than how fast the accelerators, memory and network can move data together.[1]
Software and services fill out the stack
NVIDIA Nemotron open models become available through Amazon Bedrock and SageMaker, giving developers serverless and managed routes to open-weight models. Amazon EMR data processing gains GPU acceleration with cuDF, up to 3.7 times faster with 30 percent better price performance.[1]
Amazon OpenSearch adds cuVS-accelerated vector indexing, up to nine times faster at about a quarter of the cost, and Amazon Robotics will adopt NVIDIA's Jetson, Omniverse and Isaac physical AI stack. The clearest read is that AWS is standardising on NVIDIA across compute, data and robotics.[1]
Why Malaysia should care
AWS already operates a Malaysia Region, and Malaysia is positioning itself as a regional data centre hub. A two million GPU buildout is a signal of where global AI capacity is heading, but the announcement makes no Malaysia-specific commitment, so local cloud buyers should watch for capacity, pricing and data residency details on their own timeline.
Malaysian cloud customers
More AWS AI capacity should arrive over time, but the announcement sets no Malaysia timeline or pricing.[1]
Practical move: Track AWS Malaysia Region capacity announcements rather than assuming the two million GPUs change local costs soon.
Data centre operators and partners
Malaysia's data centre push competes for the same GPUs and energy the announcement signals demand for.[1][3]
Practical move: Watch how global GPU supply and energy constraints affect local data centre buildout plans.
Developers
Open Nemotron models and GPU-accelerated data tools lower the entry cost of building on AWS.[1]
Practical move: Try Nemotron on Bedrock and cuDF on EMR when your workload fits.
What Malaysians can do now
- Separate the 2027-2028 capacity plan from anything that changes your cloud bill today.
- Watch AWS Malaysia Region announcements for local capacity, pricing and data residency rather than assuming global plans apply here.
- Test Nemotron models and GPU-accelerated data tools in the AWS services you already use before committing to new infrastructure.
What we still do not know
The announcement leaves regional details open.
- Whether the two million GPU buildout will add capacity to the AWS Malaysia Region on any earlier timeline.
- What pricing and instance types Malaysian customers will actually see during 2027 and 2028.
- How energy, power and supply constraints will shape the pace of the global GPU rollout.
- Whether any sovereign AI or data residency options will be offered for Malaysia.
Sources
- 1.AWS and NVIDIA to Deliver 2 Million Additional GPUs and Next-Generation Infrastructure for Agentic and Physical AI NVIDIA, 26 August 2026
- 2.NVIDIA NVLink Fusion Expands With NVHBM Custom High-Bandwidth Memory NVIDIA, 26 August 2026
- 3.How XPUs Meet a World-Class AI Factory NVIDIA, 24 August 2026
More AI News
View all
