AI beyond the hype at Red Hat Summit: 7 key announcements

As AI continues to reshape the enterprise world, the conversation is shifting from “what if” to “how.” During Red Hat Summit, we showcased a series of advancements designed to evolve your AI initiatives from concept to confident production. This roundup covers some of the primary articles we published during Red Hat Summit, covering everything from practical deployment strategies and data management to cutting-edge performance optimization and the rise of AI agents. Explore how Red Hat is building the foundational pieces to help your organization harness AI’s full potential, helping you turn your AI experiments into real business value. 

1. Models-as-a-Service: Let’s use AI, not just talk about it

Models-as-a-Service (MaaS) focuses on making AI more efficient and usable. The fundamental idea of MaaS is that a small team of AI experts can provide AI models as consumable API endpoints, enabling everyone in your organization to harness the power of these AI models without needing to be an expert.

At Red Hat, we’ve found MaaS significantly boosts innovation while cutting costs by deploying models once for widespread access. It also accelerates an organization's ability to adopt new AI models while enhancing privacy and data security by enabling organizations to control their own AI environments. MaaS optimizes your AI footprint and models for maximum gain, proving a practical solution for cost, speed and privacy in enterprise AI. 

2. EDB and Red Hat: A powerful combination for the AI-driven enterprise

As AI becomes more common, EBD and Red Hat are teaming up to help businesses get ahead with it. EBD Postgres AI, built on PostgreSQL, offers advanced security and high availability. It includes pgvector, which lets you store and search AI data (vector embeddings) right in the database. This makes building AI apps much simpler. Red Hat OpenShift AI provides a comprehensive platform to manage both generative and predictive AI lifecycles across hybrid clouds, complete with MLOps and LLMOps capabilities. 

This integration lets you build knowledge bases by using EDB Postgres AI’s pgvector within OpenShift AI, transforming domain-specific data into vector embeddings. This helps retrieval-augmented generation (RAG) to ground large language models (LLMs) with internal data, delivering more accurate, contextually relevant AI responses. OpenShift AI also provides an ideal environment for agentic AI frameworks, enabling AI systems to perform complex, autonomous tasks. This combined approach simplifies AI development, improving accuracy and providing a modern and consistent hybrid cloud data infrastructure for diverse intelligent applications. 

3. Beyond tokens per second: Unlocking smarter enterprise AI with inference-time scaling

While speed is often the focus in AI deployment, accuracy drives true business value. Inference-time scaling (ITS) boosts AI model accuracy without retraining by optimizing how models operate during use. This technique, especially verifier-guided parallel search, generates multiple responses and picks the best one. Our tests showed significant accuracy gains for Llama models on finance questions, with a quantized 70B model even reaching GPT-4o level performance. For enterprises, it’s crucial to prioritize accuracy over speed. You can scale AI with ITS, rather than simply increasing model size, and integrate ITS into your existing infrastructure to achieve smaller AI. 

4. Feast: The open source feature store for AI

Getting AI applications into production is tough, often due to data challenges. Feast, an open source feature store, tackles this by centralizing how organizations store, manage and serve data for AI models. It bridges the gap between data, software and AI engineers. Feast helps address critical problems like training-serving skew and feature computation redundancy, while also serving data more efficiently at scale. Red Hat has integrated Feast into OpenShift AI 2.20, providing a standardized way to deploy AI models and their data. Feast is flexible, adapting to existing data infrastructure and is already powering real-time personification, fraud detection and RAG applications across industries. 

5. Introducing Red Hat AI Inference Server: High-performance, optimized LLM serving anywhere

To deploy LLMs efficiently, we’re introducing Red Hat AI Interference Server. It’s a core component of Red Hat AI, bringing optimized LLM inference with hybrid cloud portability. Powered by the high-performance vLLM engine, it uses advanced techniques like PagedAttention and various parallelism methods for AI models. It also includes LLM Compressor for model optimization. Delivered as a container, it simplifies consistent deployment across OpenShift, RHEL and other systems, supporting multiple accelerators (NVIDIA, AMD, Google TPUs). We're also offering access to a curated, optimized model repository. Red Hat AI Inference Server delivers state-of-the-art performance and flexibility for enterprise AI. 

6. Building enterprise-ready AI agents: Streamlined development with Red Hat AI

AI is rapidly moving towards agentic systems that actively work to achieve goals, not just respond to individual queries. Building these agents often involves managing many separate tools, which can be complex. To simplify this, Red Hat is integrating Llama Stack and Model Context Protocol (MCP) into our AI portfolio, creating a more unified platform for AI models and tools.

Red Hat OpenShift AI provides the flexible foundation for agent development, offering streamlined processes through an AI API server (Llama Stack). This server acts like an AI equivalent of a Kubernetes control plane, providing consistent APIs and an extensible design that abstracts complexity. This makes building agents easier, whether you use Llama Stack directly or bring your own frameworks. The platform delivers an open source core, strong hybrid cloud capabilities, options for model customization and an enterprise-grade security posture for creating powerful AI agents. 

7. From hype to confidence: What makes a model “validated” for enterprise AI

Beyond demo hype, enterprises need “validated” AI models. Fragmented validation methods lead to costly and potentially unreliable AI deployments. True validation requires scalable performance and reproducible accuracy under real-world conditions. Red Hat offers validated third-party AI models, providing clear insights into performance, accuracy and cost. We rigorously test models across our AI platforms, enabling confident, transparent deployment of optimized AI solutions. 

Discover more about AI at Red Hat 

Red Hat is actively shaping the future of enterprise AI through powerful and flexible open source solutions. We are simplifying how you use AI, from practical deployment with MaaS to managing your data with Feast. Our focus extends to optimizing performance with inference-time scaling and building advanced AI agents. By integrating solutions like EDB Postgres AI with OpenShift AI and emphasizing validated models for predictable outcomes, Red Hat is dedicated to innovation, accessibility and long-term reliability across the AI ecosystem. 

Learn more about Red Hat AI.