Impact of Generative AI on the Public and Private Cloud

Cloud and AI: the increasing need for computational power

Introduction

The cloud market encompasses public, private, hybrid, and multi-cloud models, each designed to meet the diverse and evolving needs of modern enterprises and the public sector.

Public cloud services, delivered by leading providers such as AWS, Microsoft Azure, and Google Cloud, are popular for their cost-effectiveness (e.g., pay-as-you-go pricing models), primarily for low-capacity workloads, scalability to handle fluctuating workloads, and flexibility to support a wide range of applications. Dominating in key segments like SaaS, IaaS, and PaaS (software-, infrastructure- and platform-as-a-service), these platforms have become the backbone of modern enterprise IT, enabling organisations to innovate rapidly and efficiently.

Private cloud solutions, whether managed on-premise or by vendors like VMware or IBM, offer organizations greater control over their data, enhanced security measures, and extensive customisation options, making them more suitable for businesses with stringent regulatory or operational requirements.

Hybrid cloud combines elements of on-premise, private, and public cloud environments, enabling secure and efficient data and application transfers across these infrastructures. This model provides the security and control of private clouds with the scalability of public cloud resources.

Multi-cloud strategies involve combining services from multiple cloud providers within a single architecture. This approach enhances redundancy and improves flexibility by leveraging each provider’s strengths. Often combined with hybrid cloud setups, multi-cloud strategies help organisations avoid vendor lock-in and create customised solutions for specific workloads.

Rise of Generative AI

Generative AI is transforming the cloud market by driving rapid growth in demand for computational power and adaptable, scalable infrastructure. Across the US and Europe, more than 54% of organisations identify AI as a core driver of their cloud strategies, with adoption rates exceeding 60% in industries such as banking and manufacturing (Wipro).

Looking ahead, global AI spending is expected to grow at a 29% CAGR from 2024 to 2028 and reach USD632bn by 2028 (IDC). This growth is expected to be mostly driven by generative AI with spending on related solutions growing at a 59% CAGR within the same period and reaching USD202bn by 2028 (IDC). Foundational AI models are expected to power 70% of NLP use cases by 2027, up from just 5% in 2022 (Gartner)

As a result, cloud computing faces increasing demand to support AI’s intensive computational needs while maintaining scalability, compliance and data security.

Fuelling AI innovation: key resources and tools for AI deployments

The creation and deployment of AI models is a complex process that demands substantial hardware and software resources. These include computational capacity for tasks like training, inference, and fine- tuning, which are essential to developing and optimising AI models. Figure 2 (below) illustrates the interplay between resources used in AI model development.

Hardware viewpoint

AI workloads rely on various processors, such as CPUs, GPUs, and TPUs, each better suited to different stages of AI computing processes. GPUs, for example, are well suited to handle parallel processing demand, while TPUs have a better fit for deep learning thanks to optimised efficiency. Efficient data storage is another vital component, as AI training needs quick access to large datasets. Data lakes, warehouses, and other cloud storage solutions ensure scalability for fluctuating data volumes. High-speed networks facilitate rapid data movement between storage and processing nodes, which is especially critical in distributed AI systems.

Software viewpoint

The AI software stack includes frameworks like TensorFlow and PyTorch, which provide tools for model design, training, and deployment while leveraging specialised hardware to ensure efficiency and scalability in AI workflows.

MLOps platforms like Azure ML and Vertex AI manage data collection, model training, versioning, and deployment, automating much of the process and enabling continuous model optimisation.
APIs provide access to pre-trained models, data, and computational resources, allowing developers to add AI features without starting from scratch. Initially proprietary and closed, these APIs made developers use specific cloud platforms, historically strengthening the reliance of developers on them. Over time, the APIs were gradually opened and made public, lowering the dependence of the clients on one single provider and favouring multi-cloud and hybrid models adoption.

The role of cloud service providers

Cloud service providers offer specialised hardware and software instances, enabling access to high- performance computational resources without requiring large upfront investments. While the value propositions of public, private, and hybrid cloud models differ, all of them appeal to those seeking to utilise AI capabilities without the need to build their own infrastructure or develop specialised in-house expertise, lowering entry barriers and accelerating AI adoption. Between 2018 and 2023, the global business AI adoption rate among organizations remained around 50%, substantially increasing to 72% in 2024 (Statista). This is largely driven by the increasing use of GenAI which jumped from 33% in 2022 to 65% in 2023 (McKinsey).

Unlocking the potential of the private cloud: empowering hybrid solutions for AI

DThe private cloud model offers a distinct value proposition for developers and organisations. It provides a dedicated infrastructure with enhanced security and privacy, allowing for full control over resources and predictable costs.

It benefits industries like finance, healthcare, manufacturing and the public sector, primarily due to its enhanced data security, compliance capabilities and control over sensitive information. In finance and healthcare, private cloud environments ensure compliance with regulations, safeguarding critical data. In manufacturing, it securely centralises IoT and predictive maintenance data, protecting intellectual property and allowing for real-time, secure data analysis. For government organisations, the private cloud supports data sovereignty, helping control access to critical information.

At the same time, private cloud environments present certain challenges for AI workloads. An on-premise private cloud environment involves high upfront costs (capex), required for dedicated infrastructure such as processing units, as opposed to the opex-driven public cloud model. Rapid scalability is often constrained in the private cloud model, where resource expansion relies on available physical infrastructure, making it less flexible for AI’s resource-intensive and fluctuating demands.

On the other hand, a private cloud environment hosted by a third party like IBM, VMware or HPE, while also providing the customer with dedicated resources and customised solutions, does not involve high upfront costs compared to the fully on-premise model. Its cost structure is more predictive compared to the public 5% 7% cloud model, which can be an optimal solution for organisations with predictable and less fluctuating workloads. However, while it is possible to scale resources up or down, in the private cloud model this would require physical hardware changes that can be expensive and time-consuming. This is where the hybrid cloud model steps in, leveraging the advantages of public and private cloud solutions.

The hybrid cloud setup allows sensitive data to remain in a private environment while utilising public cloud for intensive, scalable AI tasks. In a typical hybrid model, proprietary and sensitive data is stored and processed on-premise or in the hosted private cloud. At the same time, AI models are trained and tested in the public cloud using large, non-sensitive datasets, leveraging access to the latest APIs and scalable compute resources when required. The combined use of private and public clouds has emerged as a preferred option for many organisations, with 78% of enterprise IT decision- makers utilising hybrid cloud for building and deploying generative AI solutions (Enterprise Strategy Group).

This trend has been reflected in the revenue of private and hybrid cloud providers. IBM delivered a strong performance in the second quarter of 2024, directly attributed to its AI-related services, with its bookings for generative AI-related services exceeding USD2bn. In the third quarter of 2024, HPE saw strong growth for its AI server and AI system segments, with the AI server segment achieving a record of USD1.5bn in revenue and contributing 16 p.p. to the 32% YoY growth of the overall service segment. Similarly, Dell’s infrastructure segment reached USD11.4bn in revenue in the same period, marking 34% YoY growth, with AI server sales playing a critical role.