Guest: Madelyn Olson (LinkedIn)
Company: AWS 
Show: 2026 Predictions
Topic: AI Infrastructure

The explosion of specialized databases is reversing. As AI agents and real-time systems demand faster, more accessible data, organizations are consolidating infrastructure around a smaller set of flexible platforms. Madelyn Olson, Valkey Project Maintainer and Principal Engineer for AWS In-Memory Databases, argues that 2026 will see a decisive shift toward expert teams managing fewer databases—and those databases will need to do a lot more.

The Case for Database Consolidation

“We’ve seen a very strong consolidation into fewer databases,” Olson explains. “There was a big explosion of databases around vector search, and now we’re starting to see the reverse. Teams trying to run their applications optimally aren’t as interested in managing 10 different types of databases. They really just want to run a couple.”

The trend mirrors the pre-2015 era, but with a critical difference: AI-powered tooling now enables small expert teams to manage complex, data-intensive systems at scale. These “Tiger Teams” consolidate expertise, standardize infrastructure, and ensure data accessibility for agentic workflows—all while leveraging automation to force-multiply their impact.

The winners? Postgres, Valkey, and platforms like OpenSearch. The losers? Specialized systems that can’t adapt. “You’re not going to see systems like Milvus achieve wide adoption when Valkey can handle the high-performance side of it and Postgres can manage the analytics side,” Olson says.

Real-Time Data for AI Agents

The second prediction Olson offers is rooted in infrastructure reality: agents need real-time access to structured data. “There’s still increasing usage of data, especially as more agents come online. A lot of that data needs to be accessed in real time. Stuff like Postgres and Valkey will become increasingly more useful for these agentic systems to access data very quickly.”

Valkey, which forked from Redis OSS 7.2 when Redis changed its license nearly two years ago, was designed for high-performance in-memory caching. But as AI workloads mature, the project is expanding into semantic caching, hybrid search (combining full-text and vector similarity), and improved durability.

“We’re looking at AI workloads like semantic caching with vector similarity search and full-text search capabilities, which we’re hoping to launch and expand upon in 2026,” Olson says. “We’re also looking to improve durability so that you can actually use Valkey as a foundational database product.”

That hybrid search capability is critical for retrieval-augmented generation (RAG), the technique that grounds large language models in external, domain-specific data. Valkey’s roadmap includes making these semantic queries fast, reliable, and cost-effective—key requirements for organizations building agent-powered systems.

Addressing Cost and Accessibility Challenges

Rising RAM costs are creating a bottleneck. “Cost is going to become more of a problem. The cost of RAM has been going up, and we expect the same thing to be happening in the SSD space,” Olson notes. “Making sure you’re able to keep all this data accessible quickly is going to get more expensive over the next couple of years.”

Valkey’s response: aggressive compression and SSD integration. “We’re able to very efficiently compress data at rest so that you can store more of your user data using less DRAM,” Olson explains. “We’re also looking at storing stuff in SSD—not quite as fast as DRAM, but in many cases, it can serve the same functionality and still get results returned quickly.”

The trade-off leverages modern CPUs, particularly AWS Graviton chips, which include built-in instructions for fast compression. The result: more cost-effective infrastructure that doesn’t sacrifice performance for real-time workloads.

Human-Centered AI Adoption

Olson’s advice for enterprise leaders is pragmatic: “Make sure you’re thinking through why and how you’re integrating with these agentic technologies. You should definitely be adopting them, but really focus on what problems can be solved, and more importantly, making sure you have the data available to those systems so that they can be solved.”

She emphasizes that adoption hinges on developer experience. “People will take the path of least resistance to get something done. People will readily accept and adopt technology that makes their lives easier. You have to keep humans in mind when thinking about how AI systems work with them—not just on their own.”

At AWS, Olson credits accessible data analytics systems and MCP (Model Context Protocol) servers for making it easy to integrate agents into workflows. “It’s been really great because I’ve had access to agents that are able to do analytic queries. The systems in place really make that easy to do, and that just makes my job so much easier.”

For organizations struggling with data sprawl, the message is clear: consolidation around expert teams and flexible databases isn’t optional—it’s the foundation for effective AI adoption.

Valkey’s 2026 Roadmap

Valkey’s priorities align with these predictions. The project is focused on:

• First-class durability: Moving beyond async replication to provide stronger consistency guarantees, enabling Valkey to serve as a foundational database rather than just a cache.
• Hybrid search: Launching full-text and vector similarity search in the coming months to support semantic queries and RAG workflows.
• Cost optimization: Compression and SSD integration to address rising infrastructure costs without sacrificing performance.

“We want to make Valkey as a project easier to use,” Olson says. “You can run one system, believe that the data is going to be there, and not have to think as much about the system moving forward.”

As organizations navigate the shift to AI-powered infrastructure, Olson’s predictions point to a future where fewer, more capable databases serve as the backbone for real-time data access—and where expert teams, empowered by automation, make that future scalable.