Industry executives and experts share their predictions for 2025. Read them in this 17th annual VMblog.com series exclusive.
By Sijie Guo, CEO and Co-Founder, StreamNative
Artificial Intelligence (AI) is undoubtedly evolving from buzzword to business necessity. 2025 will mark a decisive shift, however, in how enterprises approach AI implementation. The focus will move from model development and fine-tuning to the more practical challenges of building scalable, production-ready AI applications. This transition represents a maturation of the AI industry, where the emphasis on foundation models gives way to the complex realities of enterprise integration.
All in all, I see the industry moving beyond the model wars. We’re entering an era where a handful of dominant foundation models, like Llama, will serve as standardized building blocks. The real differentiation will come from how effectively organizations can architect their data pipelines and infrastructure to support AI applications at scale.
This architectural shift brings new challenges to the forefront. Organizations must now grapple with creating resilient, real-time data pipelines that can feed AI applications with fresh, contextual data. The traditional batch-oriented approach to data processing, while suitable for historical analytics, falls short when supporting dynamic AI applications that require instantaneous decision-making capabilities.
The Impact of Infrastructure in 2025
The implications I’ve outlined for data center infrastructure are significant. As AI workloads become more prevalent, organizations face mounting pressure to optimize their architecture for both cost and performance. The networking demands of AI applications can be particularly intense, with data transfer loads reaching up to 1000x that of traditional workloads. This reality is driving innovation in how data moves between systems and how processing is distributed.
We’re seeing the emergence of what I call “‘Streaming-First AI Architectures or Streaming Augmented Lakehouse.” These designs prioritize real-time data flows, enabling AI applications to maintain awareness of current conditions and adapt their responses accordingly without creating a huge cost for infrastructure. It’s a fundamental shift from static, batch-processed AI to truly dynamic, context-aware systems.
This architectural evolution demands new approaches to data management. Organizations are increasingly looking to combine the benefits of real-time streaming with robust yet affordable data storage and processing capabilities. The goal is to enable AI systems that can operate on fresh data while maintaining access to the historical context needed for accurate decision-making.
Enterprise Implications
For enterprise IT leaders, this shift necessitates a rethinking of their AI strategy. Instead of focusing primarily on model selection and training, organizations must now consider:
- How to architect data pipelines that can support real-time AI applications;
- Ways to minimize data movement while maintaining performance;
- Methods for ensuring data consistency across streaming and historical stores; and
- Approaches to scaling infrastructure cost-effectively as AI workloads grow.
The Reality of Implementation
As I’ve mentioned, this transition won’t be without challenges. Organizations must balance the desire for real-time capabilities with practical considerations around cost and complexity. Not all AI applications require instantaneous processing, and businesses need to carefully evaluate where real-time capabilities deliver meaningful value.
As we move into 2025, the success of enterprise AI initiatives will increasingly depend on architectural decisions rather than model selection. Organizations that invest in building robust, scalable data infrastructure will be better positioned to deploy AI applications that can deliver real business value.
The focus on architecture over models represents a maturing of the AI industry, where practical implementation challenges take center stage. For data center professionals, this shift emphasizes the need to build flexible, efficient infrastructure that can support the next generation of AI applications while maintaining cost-effectiveness and operational reliability.
This evolution signals a new phase in enterprise AI adoption, where the ability to effectively integrate AI into existing operations becomes more critical than raw model capabilities. As organizations navigate this transition, the emphasis on architectural innovation will drive new approaches to data center design and management, reshaping how we think about AI infrastructure in the years to come.
##
ABOUT THE AUTHOR
Sijie Guo is the Founder and CEO of Streamnative. Sijie’s journey with Apache Pulsar began at Yahoo! where he was part of the team working to develop a global messaging platform for the company. He then went to Twitter, where he led the messaging infrastructure group and co-created DistributedLog and Twitter EventBus. In 2017, he co-founded Streamlio, which was acquired by Splunk, and in 2019 he founded StreamNative. He is one of the original creators of Apache Pulsar and Apache BookKeeper, and remains VP of Apache BookKeeper and PMC Member of Apache Pulsar. Sijie lives in the San Francisco Bay Area of California.
