Omdia Consulting surveyed 304 respondents from fields such as telecommunications, industrial/manufacturing, financial services, healthcare, retail, energy/utilities, and government to determine key drivers of Enterprise AI initiatives. Alexander Harrowell, Principal Analyst, Advanced Computing wrote an excellent white paper highlighting the findings and offering interpretations around six subject areas.
Enterprise AI is all about the developer due to the following reasons:
- Diverse AI Models: Enterprises adopting AI have more distinct and diverse models than previously assumed. These models are usually trained in the enterprise in support of specific, time-bound projects. Enterprises also re-train or fine-tune their models relatively frequently, sometimes aligning with their software release cycle.
- High Training and Development Requirements: Model training and development requirements are high, while the audience using the model for inference is often small. Enterprises use more of their computing resources for model training and development than they do for inference.
- Avoiding Porting Applications: Enterprise respondents want to avoid porting AI applications from the training environment to a different chip architecture for inference. Respondents with more AI models in production were more likely to practice a faster software release cycle, typical of modern development methods such as Agile, Scrum, and Continuous Integration/Continuous Delivery.
- Focus on Project Delivery: The problem is not so much how to serve AI inference faster, but how to deliver AI projects and changes to existing AI projects faster.
- Importance of Software Support and Compatibility: Real enterprises value software support and compatibility as much as they do performance. When building infrastructure for your AI development pipeline, the widest possible selection of CPUs with standard programming tools is crucial.
- Increasing Ambition and Confidence: Data scientists and developers are showing increasing ambition and confidence in tackling more demanding business problems with AI. This points to the importance of data preparation, development, training, and evaluation relative to inference.
In summary, as the enterprise AI market shifts gears and adoption ramps up across new verticals and applications, developers are working with more models, more complex/diverse models and more model training projects than ever before. Good AI outcomes therefore come from considering these needs from the outset, in enabling developers to deliver.
Again, you may download the full white paper here.
I am fascinated by the potential for AI and ML to transform business and society, and occasionally say interesting things about it. My educational background includes a CS degree, AI/ML post-grad work, and AWS certifications. I have served in AI marketing roles at IBM, Lenovo, and now Intel.