IBM, Fujitsu, Oracle build next-generation infrastructure to accelerate Gen AI, cloud adoption

Representative image

At a time when demand for artificial intelligence (AI), generative AI (Gen AI), and cloud computing are surging, companies such as IBM, Fujitsu, and Oracle are building their next-generation enterprise infrastructure to accelerate the adoption of these nascent technologies which require massive bandwidth.

IBM and Fujitsu are developing chips and on-premise cloud facilities that are secure and efficient, apart from being rapidly scalable.

Enterprise architecture defines how software and hardware will interact and how technology and processes should align and work together. It ensures that all software, hardware, and workflows are organised and that all the tools, systems, and processes work smoothly together.

US-based IBM's hybrid cloud strategy and Japanese company Fujitsu’s integrated AI chips and cloud framework will address the challenges faced while implementing newer AI and cloud computing technologies.

Rolling out of these technologies by IT companies assumes importance at a time when increasing number of enterprises adopt cloud infrastructure and on-premise infrastructure, to scale up their AI offerings.

Traditionally, enterprise infrastructure serves multiple purposes, including providing servers for computing, storage systems for data management, and networking equipment for connectivity. It also incorporates security measures and acts as a backup system to ensure efficient operation of IT resources.

At crossroads: Infrastructure & AI

“Infrastructure is crucial for responsible AI deployment. (It) combines cutting-edge technology with robust governance and compliance to build trust and effectiveness in AI solutions,” said Akhtar Ali, vice president of Technical Computing Development, Systems & Technology Group at IBM.

The Bengaluru-headquartered IBM India Systems Development Lab is part of IBM’s global infrastructure business unit, which focuses on the development of technologies such as storage and cloud.

A recent report revealed that about 73 percent of CEOs prioritise technology investments in enterprise architecture to build organisational resilience. Catering to this, IBM provides a scalable cloud infrastructure for model training and tuning.

At a time when AI is being integrated into every aspect of business, as highlighted by recent management commentaries of top tech companies, the role of infrastructure has gained more prominence.

More so for clients of IT companies who operate in some of the highly regulated sectors such as banking, financial services, and insurance.

Jenny Tsai, senior vice president - overall database product management at Oracle, says that clients from these sectors are concerned about making sure that they don't send data out to a public large language model (LLM). Therefore, more and more customers want to bring LLMs into their local data centres, Tsai said.

“I'm sure at some point in time, they're going to have to look at how it works economically because that means they have to maintain their own LLM (and) also train it. So they need a lot more infrastructure to keep that going,” Tsai told Moneycontrol.

Also read: BFSI global capability centres bullish on Gen AI despite regulatory hurdles

The world’s sixth-largest IT company Fujitsu has capabilities to build an on-premise cloud-based facility for developing applications that require AI-based customised model training.

An on-premise cloud-based facility is a combination of traditional on-premise infrastructure with cloud computing technology, having the flexibility of the cloud with the control of having the equipment on-site.

GPUs & CPUs: Combined potential

What is known is that high-performance computing environments equipped with GPU is also being used for rapid training and deployment of AI models. However, a lesser-known fact is that when an AI model is being trained, it is first launched on the central processing unit (CPU) and GPU is an add-on support to any CPU architecture.

For example, Fujitsu is building a 2-nanometer CPU, named FUJITSU-MONAKA, being developed specifically for data centres. It uses proprietary technologies such as microarchitecture and delivers faster AI workload processing.

“So, what we are developing will be useful to anyone, including Nvidia… we very open to collaborating and we are actually already collaborating with a number of other tech giants (in) developing our software ecosystem,” Dr Priyanka Sharma, who is the head of MONAKA Software R&D Unit, Fujitsu Research of India, told Moneycontrol.

The FUJITSU-MONAKA CPU is scheduled to launch in 2027.

To train an AI model, after the CPU’s initial processing is complete, next comes the equally critical processing on GPUs. Several IT service providers offer GPUs to train models and process data for generative AI.

Tech giant Oracle has the capabilities to essentially create superclusters, with over a whopping 32,000 GPUs.

“So we have a lot of that built-in like an architectural design to our infrastructures that makes it very, very fast. This is part of the reason why like Microsoft is using our infrastructure to run the generative AI component of Bing Search,” Tsai said.

Ali added that IBM provides a scalable cloud infrastructure for model training and tuning, leveraging virtualisation, software-defined networking, and GPU resources.

Also read: The force called Nvidia in artificial intelligence