GPUaaS makes expensive piece of computing tech available to the wider world

GPUaaS lets businesses use the expensive chips as needed, paying only for what they use

chips chipset semiconductor — Representative Picture

5 min read Last Updated : May 12 2024 | 9:35 PM IST

Krutrim, the homegrown artificial intelligence startup, early in May introduced graphics processing unit as a-service (GPUaaS) for enterprises and developers to train their AI systems.

A GPU is a specialised computer chip that renders graphics and images by performing rapid mathematical calculations needed in AI processing. Demand for GPUaaS is expected to grow manifold as startups offer generative artificial intelligence (GenAI) — the core software of AI – products.

GPUaaS lets businesses use the expensive chips as needed, paying only for what they use. It is a flexible solution to avail of GPU and economical because businesses can avoid investing in hardware.

“The vision is to enable developers to do cutting-edge innovation. The platforms or tools that developers need are computing and services and that is what we have opened up. For instance, we have launched GPU-as-a-service, model-as-a-service, and location services for developers,” said Ravi Jain, head of strategy at Krutrim, which translates to "artificial" in Sanskrit.

GPUaaS is not new but has gained traction in the past few years as GenAI booms. “…(a) GPUaaS platform accelerates tasks like deep learning and rendering, and streamlines GPU resource management by automating allocation, scaling, and maintenance processes. Manually managing these tasks, especially at scale, can be complex and time-consuming,” said Rashid Khan, chief product officer and co-founder of Yellow.ai, a customer service automation platform.

ALSO READ: Lam Research plans semiconductor fabrication supply chain in India

There were 18 GenAI startups in India in 2021 and more than doubled to 60 as of the first quarter 2023, according to a Nasscom report, highlighting the demand for GPU-enabled computing infrastructure in the country.

Data from Nasscom, which represents India’s technology industry, shows that $8 billion was invested in AI between 2013 and 2022 ($3.2 billion in 2022 alone) across 1,900 AI startups in India.

The international GPUaaS market was worth $1.83 billion in 2021 and is projected to reach $19.56 billion by 2030 as it expands at a compound annual growth rate of 33.2 per cent, according to Allied Market Research.

India’s demand

Indian companies have just started implementing GenAI and are cautiously investing in computing capacity, so GPUaaS is helpful for them, said experts.

“Most firms use AI-as-a-service from a service provider and are not exposed to the cost of GPUs. Those few who want to deploy their own custom AI services are looking at alternatives to owning the infrastructure themselves. It is for such a segment that AI infrastructure as a service is positioned,” said Naresh Singh, senior analyst at market research firm Gartner.

Experts also say that Cloud service providers can offer GPUaaS to Indian companies, startups, global capability centres and governments.

“Rather than companies investing in GPU capacity in-house, which is both expensive and needs specialised skills, it makes sense to hire GPU on demand from these cloud providers,” said S Anjani Kumar, partner for consulting at Deloitte India.

Smaller organisations can use GenAi through GPUaaS. “GPUaaS providers leverage their bulk purchasing power to negotiate better terms, thus making GPUs more accessible and supporting a broader range of companies in harnessing advanced computational technologies,” said Ramprakash Ramamoorthy, director of AI research at ManageEngine, an information technology (IT) service company.

The government, as part of its Rs 10,000 crore India AI mission, might rent GPUs from domestic firms and provide the chips at subsidised rates to startups, according to media reports.

Yotta Infrastructure, an Indian data services company, aims to compete with global giants like Amazon and Meta Platforms, the owner of social media platforms, by providing GPU servers at a competitive price of $2-$2.5 per hour of usage. Globally, these rates are $9-$12.

“The government is going to source these GPUs as a service model from people like us who are creating the data centres by bringing in GPUs, and putting the software layer on the top,” said Sunil Gupta, chief executive officer of Yotta, earlier.

Governments will find GPUaaS suitable for their strategic needs, according to experts. “Government customers typically have sovereign needs and rigid security compliance requirements. Typically, they may begin with AI as a service but could take a strategic step to invest in their own infrastructure from regulatory and national security and strategic perspectives. This can drive good market demand for AI infrastructure (either captive or as a service),” said Naresh Singh, senior director analyst at Gartner.

Pay and go

A key benefit of GPUaaS for governments is the ‘pay-as-you-go’ feature, which eliminates the need for upfront investment in hardware, software, skills, and services, allowing the government to pay based on usage, said Anjani Kumar of Deloitte India.

Data centre and Cloud companies are expanding GPUaaS business as demand grows. Cloud computing firm CtrlS will invest $2 billion to develop AI and data centres in the next six years.

“Our expansion plans include integrating advanced technologies such as liquid cooling, AI, heating, ventilation, and air conditioning (HVAC), and optimised rack layouts to cater to the growing demand for AI applications and Cloud services,” said Ranjit Metrani, president for managed services at CtrlS Datacenters.

Neysa, an AI Cloud company, got $20 million in seed funding in April as it works on offering a GenAI solution as a service. The service will help clients “discover, plan, deploy, and manage their GenAI projects cost-effectively in a consumption-based model,” said the company earlier.

Technology-focused venture capital funds believe that the government's assistance in setting GPU clusters and giving access to training AI models at subsidised prices will help startups and academia.