The wheel, steam engine and the internet created revolutionary jumps in the way people work and play. Today, artificial intelligence is reshaping science, business and personal interactions with equal magnitude. In every industry, including agriculture, healthcare, customer service, finance, manufacturing, retail and more, companies are quickly adopting AI to ensure theyre not left behind during this tectonic shift.

AI workloads have unique requirements, including strategic planning to ensure data scientists and researchers work efficiently on delivering successful projects. For IT teams just starting out, its helpful to know that while AI workloads require accelerated infrastructure and software, many of the solutions IT is most familiar with are already AI-ready for integration into an innovative strategy for creating an AI Centre of Excellence.

Few resources are as readily available and easy to use as the cloud, and this easy access to infrastructure extends to AI workloads. With GPU-accelerated instances available from every cloud service provider, these resources are ideal for prototyping AI projects. They provide the scale needed when training new models. The cloud also can serve enterprises well as infrastructure for AI inference workloads, where AI models are deployed for things like computer vision, conversational AI, speech, language and translation, and recommendation systems.

The challenge here is that data governance and cloud costs can complicate AI adoption. Training models generally require processing large datasets, and as AI projects grow, hosting all the data on the cloud can result in unexpected costs. Additionally, when AI is deployed in applications, many apps require real-time responsiveness for automation or user experience, which can become a challenge when data makes a round trip from the cloud.

To overcome these hurdles, enterprises are building AI Centres of Excellence with on-prem systems for AI that connect with cloud-based AI computing for prototyping and scale. This involves planning for data gravity and putting computing closer to the source of data to ensure costs are balanced and resources are at the ready. It also helps enterprises start with small projects in the cloud that grow into the hybrid ecosystem when its time to deploy. All major cloud service providers offer hybrid accelerated computing solutions, making it easier to harness both on-prem and cloud-based compute resources as needed.

With this hybrid approach, enterprise data scientists always have the resources they need to stay as productive as possible whether theyre creating new models, training AI, or evaluating a deployed model to ensure its still accurate.

Its also important to consider the big picture when looking at the cost of accelerated computing in the hybrid cloud. On paper, high-performance instances may at first look costly, but they end up delivering significant cost savings. They enable large datasets to be processed much more quickly, which results in lower total costs. Most importantly, these instances provide faster time-to-market for products and services. In addition, software technology can help right-size accelerated computing resources to maximise efficiency on diverse AI training and inference workloads.

For AI use cases like conversational AI services, accelerated computing platforms train large, sophisticated networks in hours instead of weeks. When deployed as AI-powered services, these networks deliver immediate, natural-sounding replies to complex questions.

Central to every AI project is a software architecture built to deliver on enterprise AI objectives. Workloads for conversational AI, recommender systems, robotics automation and computer vision all depend on specialised software designed for these unique applications.

These software requirements can present the biggest challenges for AI teams getting started on new projects. To help companies hit the ground running on their AI Centres of Excellence, NVIDIA offers free software resources for developers and data scientists. The NVIDIA AI platform also offers a single architecture to develop and optimise the applications while offering the flexibility to run them anywhere.

For businesses, one size rarely fits all. The same is true for AI workloads. With a hybrid cloud strategy to augment an enterprise AI Centre of Excellence, IT teams can deliver AI acceleration thats both on demand and within budgets. By keeping AI software in mind and developing a strategy to keep pace with software innovation, enterprises will be ready to scale easily from the data centre, to the cloud, to the edge.

Interested in hearing industry leaders discuss subjects like this and sharing their experiences and use-cases? The Data Centre Congress, 4th March 2021 is a free virtual event exploring the world of data centres. Learn more here and book your free ticket:https://datacentrecongress.com/

Original post:

How the hybrid cloud is key to enterprise AI infrastructure strategies - Cloud Tech