t cht lk
t cht lk
Tony Bartlett, Director of Data
Centre Compute, Dell Technologies South Africa
However, 57 % struggle with integrating AI workloads across cloud platforms, underscoring the need for a seamless infrastructure.
As IT teams consider the best resources for deploying cloud workloads, the choice often comes down to onpremises versus cloud infrastructure via public cloud services. Public cloud resources offer extraordinary scalability and access to next-generation technologies. Private cloud or on-premises infrastructure provides more control, security and visibility.
As cloud solutions have proliferated, enterprises have adopted both on-premises and cloud infrastructure to create a multicloud architecture. Yet deploying multiple public clouds and on-premises technology can create a complex management experience that adds cost, risk and administrative burden to the task of running workloads on cloud resources.
To effectively implement AI in a multicloud world, enterprises must focus on four key pillars: computing power, data management, storage, and efficiency. Each of these play a crucial role in supporting AI workloads at scale.
Computing power
AI’ s potential is fully realised when enterprises have the right computing power and network capabilities to support large-scale data processing. These components serve as the foundation for AI workloads, ensuring models run efficiently and deliver meaningful results.
Advanced processing power specialised accelerators are necessary for training AI on large datasets.
For example, financial institutions leverage AIoptimised GPUs for real-time fraud detection. Whether through on-premises data centres or cloud-based AI-optimised instances, businesses must ensure they have the right computing resources.
Connectivity with high-speed networks
AI applications require fast, uninterrupted data transfer. High-bandwidth, low-latency connections between cloud environments ensure smooth AI operations. Businesses should leverage software-defined networking, SDN and network optimisation tools for seamless connectivity.
Data management
AI thrives on high-quality, accessible data, but managing data across multiple clouds can be challenging. Without seamless data integration, AI models risk being trained on outdated or incomplete datasets, leading to unreliable insights. Effective data management strategies are key to AI success.
Unified data governance
With data spread across different clouds, security, compliance, and consistency are critical. Enterprises need strong governance frameworks to ensure regulatory compliance, e. g., GDPR, CCPA and data security. AI-specific governance policies should address concerns like bias in training datasets and privacy.
AI models, particularly those using machine learning, ML and deep learning, DL, require extensive computational power. High-speed Graphics Processing Units, GPUs, Tensor Processing Units, TPUs, and
Seamless data integration
AI models pull data from multiple sources, including legacy systems, cloud storage, and real-time streams. Integration tools that ensure seamless interoperability across these sources help businesses consolidate and access data efficiently.
Real-time data access
Many AI-driven applications, such as fraud detection and predictive maintenance, depend on real-time insights. Enterprises should invest in cloud-native solutions for real-time data ingestion and processing.
Storage
AI workloads generate and consume vast amounts of data. Inefficient storage strategies can inflate operational costs as businesses struggle to balance access speed with budget constraints. Therefore,
68 INTELLIGENTCIO AFRICA www. intelligentcio. com