Business Info - issue 151 magazine 24 INFRASTRUCTURE Enterprises around the globe are ramping up their AI investments in a bid to understand customers better, enhance service delivery and create more agility in their supply chains and distribution networks. They’re also leveraging AI to help employees do their jobs better and deploying analytics and AI to identify where and how to optimise business and operational processes. With business resilience and relevance now top of the corporate agenda, following the disruptions caused by the global pandemic, enterprise demand for AI capabilities that augment human productivity and support business resiliency and innovation is accelerating at pace. So much so that Ritu Jyoti, group vice president for AI and Automation Research at IDC, recently suggested “we have now entered the domain of AI-augmented work and decision-making across all the functional areas of a business”, adding that “responsible creation and use of AI solutions that can sense, predict, respond and adapt at speed is an important business imperative.” As AI becomes prevalent across the enterprise, organisations will need to ensure they can support the growing volume of data-intensive workloads in a highly agile and cost-effective manner. Scalability, high performance computing (HPC) capacity and storage are just some of the critical considerations that will need to be prioritised to assure effective performance both now and into the future. Infrastructure and platform requirements While CPU-based environments are capable of handling basic AI workloads, deep learning involves multiple large data sets and scalable neural network algorithms, so organisations will need to ensure that they have sufficient compute resources, featuring AI-focused GPUs, to cope with these demands, as well as scalable storage capacity. Is your infrastructure fit for AI? In addition to determining how much AI data is likely to be generated by applications, organisations will need to evaluate what types of storage will be needed. For example, businesses that use AI to make real-time execution decisions, such as financial services firms, will need fast all-flash storage. Similarly, organisations that build out big data and analytics environments may well find that traditional network-attached storage architectures present scaling issues in relation to I/O and latency. Networking is another key infrastructure component that requires careful consideration. High bandwidth, low-latency, reliable and scalable networks will be vital for enabling AI and machine learning models, for optimising the delivery of AI results and for ensuring robust transmission of huge volumes of data from connected IoT devices and sensors. Alongside ensuring their networks can react in real time, organisations should ensure that their service wrap and technology stack are consistent in all regions. Data management and security Data quality is critical to the effective operation of AI systems, so organisations will need to deploy automated data cleansing tools to remediate against the risk of inaccurate or out of date data leading to flawed decision-making or inaccurate predictive models. As well as ensuring that machines and people experience fast and easy access to the data they need, from any endpoint, organisations will need to have strong identity and access controls in place to protect data access and usage and a proper data management infrastructure that will track, control and monitor the flow of data in an end-toend enterprise analytics system. Since AI can involve sensitive data, such as patient records, financial information and personal data, any data breach will be a significant disaster for the enterprise. For this reason it will be critical to ensure that the end-to-end AI infrastructure is secured using state-ofthe-art encryption technologies. Tackling AI infrastructure challenges The unique demands of AI workloads make cloud – specifically hybrid cloud – the ideal foundation for AI. As data sets grow and AI complexity increases, the escalating cost of compute cycles, data movement and storage can spiral out of control. Hybrid cloud enables enterprises to utilise on-premises infrastructure for ongoing, steady-state AI workloads, supplemented with cloud services whenever additional capacity is needed. The versatility of hybrid cloud solutions such as Infrastructure-as-aService enables organisations of all sizes to meet the evolving technology demands of AI at a sustainable cost that makes it possible to develop and implement AI without sacrificing performance. Choosing a suitable IaaS provider will be key to the overall success and failure of projects, so organisations will need to evaluate providers carefully to ensure they represent the best fit for current and future plans, including the offer of cost-effective dedicated servers as a means to boost performance. Providing ease of access to infrastructure designed for AI workloads that can be flexed on demand and delivered via a convenient Opex-based model will enable organisations to draw on computing power, storage and connectivity when and where they’re needed and enable the highly scalable and secure architecture needed to power today’s innovative AI systems. Terry Storrar, Managing Director of Leaseweb UK, explains what to look for when building enterprise infrastructures capable of supporting greater use of Artificial Intelligence in business decision-making Terry Storrar