by Tan Aik Keong (AK)
Over the past few decades, the tech industry has gone through several infrastructure revolutions: the PC era moved computing from mainframes to personal devices; the internet era transformed business models through connectivity; the cloud era let businesses draw on elastic computing resources on demand.
Today, artificial intelligence is driving the next one. But AI's revolution is different from the ones before it: it doesn't just depend on software and algorithms — it depends on a new strategic resource, compute power.
Businesses used to talk about AI in terms of model capability — ChatGPT, Claude, Gemini and other large language models. But as AI moves from chatbots toward AI agents, demand for compute is rising fast. Future AI won't just answer questions; it will run continuously, analyse data, call tools and execute tasks — all of which needs serious GPU power behind it.
That's why GPUs are becoming one of the AI era's most important pieces of infrastructure. Just as businesses stopped building their own server rooms and moved to the cloud, more and more businesses will stop buying large numbers of GPUs and instead draw on AI compute through GPU-as-a-Service (GPUaaS).
For Malaysia, GPUaaS represents a real industry opportunity. With a growing data centre ecosystem, a strategic regional location and strong Southeast Asian market demand, Malaysia has a real chance to become an important player in the AI compute supply chain.
Why now is the time for GPU-as-a-Service
In the past, a company that wanted to build AI capability had to buy its own GPU servers. Training a large AI model requires a lot of high-performance GPUs, plus solving for data centre space, power supply, cooling, high-speed networking and software environment management — not just a financial burden, but a genuinely complex technical challenge for most businesses.
GPU-as-a-Service changes that. Instead of a huge upfront investment in a GPU cluster, businesses can rent AI compute as a service, matched to actual need — paying by GPU hours used, workload, inference demand, or even a dedicated GPU cluster.
This turns GPU compute from a resource only big tech companies could afford into a commercial service more businesses can use. In future, many companies won't own their own AI data centre, just as most don't build their own server rooms today — they'll draw on GPUaaS for the AI capability they need.
Why global tech giants are betting on AI infrastructure
In recent years, one trend in global AI competition has become increasingly clear: the AI race isn't just a race between models — it's a race between infrastructures.
NVIDIA founder and CEO Jensen Huang has repeatedly stressed that AI infrastructure will be the core construction project of the coming decades. Computing has spent decades centred on the CPU; the AI era is shifting it toward GPU-accelerated computing. Microsoft, Google and Amazon are each pouring billions of dollars into AI data centres, cloud platforms and GPU clusters.
The reason is simple: even the most advanced AI model can't run at scale without enough compute behind it. The real competition ahead isn't just "who has the smartest AI model" — it's who can provide more powerful, more stable, more cost-effective AI compute.
Why Malaysia has real potential in GPUaaS
Many people assume AI infrastructure only clusters around the US, China or Europe. In reality, Southeast Asia is becoming an important region for global data centre investment, and Malaysia sits right at the centre of that trend.
Over the past few years, Malaysia's data centre industry has grown fast. International cloud providers and data centre operators have kept increasing investment, helping Malaysia build a maturing data centre ecosystem — high-speed connectivity, cloud infrastructure, and enterprise-grade data centre capability. That's an important foundation for developing AI compute.
But AI data centres also bring new challenges, chief among them energy and water. Traditional data centres are mostly about server capacity; AI data centres, running dense GPU clusters, place much higher demands on power supply, cooling and energy efficiency. A large AI data centre needs stable, large-scale power, alongside efficient cooling to keep GPUs running continuously.
So the next phase of AI infrastructure competition won't just be about land, networking and data centre space — it will be a competition over energy infrastructure too. For Malaysia, power isn't a fully solved natural advantage; it's an area that needs continued investment and planning. How quickly we can expand reliable energy supply, strengthen the grid, and push forward green energy will be a real challenge in building out Malaysia's AI infrastructure.
Malaysia's biggest opportunity right now lies in what's already been built: a strategic location, a growing data centre ecosystem, international connectivity, and proximity to Southeast Asia's major markets. If we can further solve for energy supply, infrastructure expansion and purpose-built AI data centres, Malaysia has a real shot at moving from a data centre market to a genuine AI Compute Hub.
How to actually start a GPUaaS business
Building a GPU-as-a-Service business isn't as simple as buying a few GPUs and renting them out. A complete GPUaaS platform needs to combine hardware, data centre capability, a software platform and real operational capacity.
First, GPU hardware infrastructure. The AI market today runs mainly on high-performance NVIDIA GPUs — H100, H200, the B-series and other AI accelerator platforms — the core resource for training large models, running inference, and handling enterprise AI workloads. Businesses need to choose configurations based on their target customers: an AI startup might need a small-scale GPU environment for model development, while a large enterprise might need a dedicated GPU cluster running its own internal AI platform.
Second, data centre capability. GPU compute generates a lot of heat, so an ordinary server room can't handle AI workloads. A GPU data centre needs higher-density rack design, efficient cooling, high-speed networking and stable infrastructure management — which is why GPUaaS businesses usually need to partner with established data centres rather than simply buying hardware.
Third, an AI cloud platform. A GPUaaS service that actually delivers value doesn't just provide GPUs — it provides a simple, secure, manageable platform. Customers need to be able to request GPUs, deploy models, monitor usage, manage team permissions and track costs through that platform — similar to how AWS turned complex server infrastructure into an easy-to-use cloud service.
Fourth, commercial and compliance capability. As GPU compute becomes more strategically important, GPU providers need a complete customer management framework — KYC, use-case review, resource management, billing and security controls. In future, GPUs won't just be ordinary compute — they may also fall under AI chip supply-chain regulation, so GPU providers will need to track end users and end uses carefully. A successful GPUaaS business isn't just renting out GPUs — it's building a secure, trustworthy, scalable AI compute platform.
GPUaaS business models
GPUaaS can be built around several models: hourly billing, where customers pay by GPU time, suited to developers, research institutions and short-term projects; subscription, where businesses buy a fixed GPU quota and pay monthly, suited to long-term AI applications; dedicated AI compute, where large enterprises get a dedicated GPU cluster for internal model training, sensitive data processing or enterprise AI platforms; and AI Solution + GPU, where the bigger opportunity isn't renting out GPUs at all, but delivering a complete AI solution on top of them — enterprise AI agents, intelligent customer service, AI data analytics, industry-specific models and automated workflows. The real long-term commercial value isn't selling compute — it's helping businesses turn compute into productivity.
GPUaaS is the AI era's new infrastructure opportunity
Many people think of GPUaaS as simply "renting out GPUs." In reality, it's closer to how cloud computing developed. Cloud computing changed how businesses acquired servers. GPUaaS is changing how businesses acquire AI capability.
What businesses will need going forward isn't just GPUs — it's a full stack of AI capability: compute, data management, model deployment, AI agents, and enterprise systems integration. GPUs are the foundation, but the software platform and industry applications built on top are what ultimately determine the commercial value.
Malaysia's AI compute opportunity
The opportunity AI brings Malaysia isn't just becoming a market for AI applications — it's a real chance to become an important node in Southeast Asia's AI infrastructure. In the internet era, the critical asset was the network. In the cloud era, it was the cloud platform. In the AI era, it will be compute infrastructure.
Malaysia already has the foundation to build a data centre industry. The next step is pushing the GPU compute ecosystem further — connecting data centres, power, software platforms and AI applications together. Malaysian businesses won't just use AI in future — they can become suppliers of AI infrastructure.
Closing: the next AI race is a compute race
AI models keep moving faster, but every one of them shares the same underlying need: more, stronger, more stable compute. From GPUs to data centres, from software platforms to enterprise applications, GPU-as-a-Service is opening up a real new industry opportunity. Global tech giants have made it clear that AI infrastructure will shape the industry landscape ahead.
For Malaysia, now is an important window to build out an AI compute ecosystem. What will decide the AI race ahead isn't who has the most advanced model — it's who can provide the strongest compute infrastructure. And GPUaaS may well be Malaysia's most important entry point into the global AI industry chain.
For businesses exploring AI compute infrastructure, GPU cloud services, or enterprise AI deployment, the key first step is understanding your own needs, scale and business model. Agmo Group currently also offers GPU compute consulting, helping businesses plan their full journey — from GPU selection and infrastructure architecture through to AI application deployment.
Part of the AK AI Corner column. Originally published in Oriental Daily (东方日报) on Aug 13, 2026.