Why Vector Data Services For AI Are A Moveable Feast

Cloud is a combo. It wasn’t long in the early evolution and revolution of cloud computing that we realized one cloud service in one place from one Cloud Services Provider (CSP, aka hyperscaler) was okay, but a combination cloud made a lot more sense. By combining a proportion of public cloud services (major-scale cloud delivered by a hyperscaler) with a degree of private cloud (server-level technology operated inside an organization’s own datacenter on-premises), we achieve scale with the former and additional governance, privacy and mission-critical lockdown with the latter.

The hybrid combination of both cloud resources - sometimes even now split out across poly-cloud architectures where some parts of a monolithic application are handled by different cloud services - made sense two decades ago in the early cloud ascendancy, as it still does today.

The hybrid cloud story has been happily unfolding for some time now, with its next chapter possibly driven by the need to use the vector database services so well suited to Large Language Models (LLMs) in the Artificial Intelligence (AI) space in more hybrid-style deployments.

Guided (managed) hybrid

Aiming to join the queue and make an impact in this space is vowel-economization enthusiast company Qdrant, a firm known for its open source vector database. Pronounced with its missing U & A as ‘quadrant’, Qdrant is bidding for recognition as one of the first technology vendors to offer a vector database in a managed hybrid cloud model.

Now explaining how its newly available Qdrant Hybrid Cloud service works, the company says that this is a run-anywhere self-service solution. That means Qdrant is supposed to allow businesses to deploy and manage vector databases across any cloud provider, on-premise or edge location. Why do we need hybrid portability (place-ability would be more accurate) to be able to put different elements of data and cloud service in more than one place? Because it enables a business to maximize performance, security and cost efficiency for AI-driven applications. It also helps organizations to control and protect where and how sensitive data is used.

“Enterprises need to run their vector database applications in any environment with full control over their data, and Qdrant Hybrid Cloud is built to address exactly this,” said André Zayarni, CEO & co-founder of Qdrant. “With Hybrid Cloud, Qdrant takes the next step in enabling large enterprises to face complex challenges and better build and implement robust, next-gen AI applications while meeting strict risk and compliance standards.”

Zayarni reminds us that vector databases have emerged as a critical component for building AI-based (and especially generative AI-based) applications. He says that Qdrant Hybrid Cloud marks an advancement in the field of vector search and enterprise AI, bringing vector search applications a level where they can redefine the standard for enterprise-grade AI applications. It provides a rich set of features for performance optimization and can handle billions of vectors with efficiency, scale and ‘memory safety’ i.e. data is protected towards the point of persistence.

“Qdrant Hybrid Cloud [is built with] architecture to ensure complete database isolation, which allows customers to deploy a vector database in their chosen environment without sacrificing the benefits of a managed cloud service,” notes Bastian Hofmann, director of enterprise solutions at Qdrant. “It gives users maximum control over their data and vector search workloads, enabling companies to unlock unprecedented opportunities to deliver personalized customer experiences in the era of AI while upholding privacy and data sovereignty.”

The company cements its point by stressing its open source principles, which it says will foster a new level of trust and reliability as companies navigate the evolving enterprise AI landscape.

The vector database market is of course on a trajectory for further growth, fueled by the integration of generative AI and Retrieval-Augmented Generation (RAG) techniques that enhance Large Language Models with external, proprietary data for additional awareness and validation accuracy. The company says that this growth is integral to the expansion of the generative AI market, where RAG plays a significant role in driving innovation, customized user experiences and application diversity.

Three core takeaways

Has Qdrant pushed forward vector database technologies into a more adept, flexible and manageable space with Qdrant Hybrid Cloud as a managed hybrid cloud service? Almost certainly yes.

Is Qdrant the only vector database company offering a managed hybrid cloud service as it claims? It’s a cloudy (pun not intended) area in terms of definition, database vendors like DataStax remind us that organizations can run Apache Cassandra in hybrid environments - some nodes in a company’s own datacenter and some on a hyperscaler such as Azure, Google or AWS - and that would still in technical terms be considered one single deployment.

In fairness to Qdrant, the company qualifies its gambit here by suggesting that it is the industry’s first ‘dedicated’ vector database to be offered in a managed hybrid cloud model. According to Qdrant CEO Andre Zayarni, add-on vector search capabilities may be a good starting point for experimenting with vector search, but users will hit limitations when it comes to more complex scenarios. Added-on solutions lack specialized storage and retrieval systems needed for vector data. Also, their generalized architecture limits customization options. Dedicated native vendor databases, meanwhile, focus on vector search. They give users and developers the speed, memory safety and scale they need even when processing billions of vectors. Analyst house Gartner specifically calls out ‘scalable vector databases’ as a critical enabler in its Impact Radar for 2024.

Does this make the whole discussion any less valid and intrinsically interesting as we now create the data services needed to run really smart AI applications everywhere and on any foundational format and configuration of cloud computing? Good heavens no. The still-nascent technologies in this space are characterized by a huge amount of prototyping and experimentation, we need to work out where all the pieces go and all this stuff helps, a lot.

While AI itself is still very much on the move, the use of managed hybrid cloud services can give us the vector data layer we need to make this whole meal a moveable feast.