Approximately two years ago, Redis, a widely-used cache database, found itself at the forefront of a burgeoning trend among technology vendors, as many began integrating vector search functionalities into their offerings. This movement was largely influenced by the exponential rise in interest surrounding generative artificial intelligence (AI), which has transformed how businesses and developers interact with data.

Vector embeddingsmathematical representations generated by foundational models like OpenAI's ChatGPTare pivotal in this context. They effectively encapsulate chunks of language, such as words or phrases, facilitating nuanced searches. By incorporating vector search into databases, enterprises can optimize their systems to store, index, and search through millions of vectors, thereby extending the capabilities of foundational models to harness enterprise-level data more effectively.

Since the beginning of 2023, a plethora of database systems have rolled out vector search as a primary feature. This rapid adoption has made it increasingly challenging for vendors to stand out in a crowded marketplace. Prominent database solutions, including MongoDB, Cassandra, PostgreSQL, Snowflake, and SingleStore, have all announced similar capabilities in the span of a single year.

As competition intensifies, many vendors are looking to differentiate themselves by introducing more sophisticated features aimed at solidifying their positions within the evolving AI landscape. Redis, for instance, unveiled LangCachea fully managed REST service designed to mitigate costly and latency-prone calls to large language models (LLMs). This innovative service caches previous responses to queries that are semantically similar, thus enhancing efficiency.

LangCache acts as an intermediary between your application and the LLMs you utilize. When a query is submitted, we determine if a similar response already exists in our cache and return it, eliminating the need to engage the LLM inference engine directly, explained Rowan Trollope, CEO of Redis, in an interview with The Register earlier this month.

Similarly, YugabyteDB, which offers a PostgreSQL-compatible database with a distributed backend, is striving to adapt its software more comprehensively to support AI initiatives. With notable clients such as Paramount, GM, and Kroger, the company has stated that integrating vector indexing libraries like Usearch, HNSWLib, and Faiss will significantly enhance the performance of its PostgreSQL vector search extension, pgvector.

Karthik Ranganathan, founder and co-CEO of Yugabyte, discussed the companys approach, noting that they have replicated and automatically sharded these indexes within their distributed backend to optimize pgvector's performance. He emphasized that, although pgvector benefits from the extensive PostgreSQL community, it requires performance enhancements to compete effectively with specialized vector databases. In the fast-paced world of AI, we are directly interfacing with various open-source libraries to keep up, he added.

According to research from Gartner, many ambitious AI projects in business, often dubbed moonshot projects, have been experiencing considerable failure rates. As the hype around generative AI begins to wane, expectations regarding its capabilities are likely to be recalibrated, with Gartner predicting 2025 will be a pivotal year for assessing AI's utility in business.

Gartner also raised concerns regarding the lack of readiness in data infrastructure, which hampers the integration of generative AI into real-world applications. It forecasted that, by 2025 and 2026, organizations might abandon up to 60% of their AI initiatives due to insufficient data preparation. Roxane Edjlali, a senior director analyst at Gartner, underscored the importance of including vector data stores as a means to enhance data management practices suitable for innovative applications like generative AI.

AI-ready data is not a one-time achievement. Its an ongoing process, necessitating continuous improvements in data management infrastructures that adapt to existing and upcoming AI use cases, she stated, urging organizations to invest in AI and develop robust data management practices.

While both Redis and YugabyteDB focus on enhancing their transactional systems to support AI, Oracle remains the leading transactional database, recognized for pioneering natural language vector search functionalities within relational business systems.

On the analytics front, vendors are also evolving their technologies to align with the demands of generative AI. Teradata, which boasts a 40-year legacy in business intelligence and data warehousing, serves clients including HSBC, Unilever, and American Airlines. Last year, Teradata announced its integration of Nvidia NeMo and NIM microservices into its Vantage cloud platform, aiming to accelerate AI workloads and support the development of both foundational and customized large language models (LLMs), as well as retrieval-augmented generation (RAG) applications.

Teradatas journey into machine learning commenced with its acquisition of specialist analytics vendor Aster for $263 million, further demonstrating its commitment to advancing analytics capabilities. Martin Willcox, VP of analytics and architecture at Teradata, noted the growing demand for analytics and business intelligence services as organizations explore AI agents and LLM-driven customer interactions. He highlighted Teradatas Massively Parallel Processing (MPP) architecture, which is adept at running inferences for AI models with parameter ranges of 150 to 250 million, and how they are enhancing API integrations with hyperscalers to facilitate better support for LLMs.

Concurrently, clients are amassing extensive stores of unstructured dataincluding images, audio, PDFs, and emailsand utilizing vector search to distill meaningful insights from this information alongside LLMs. Current vector database technologies essentially come in two flavors, Willcox explained. There are specialized tools that excel with smaller datasets but struggle to scale, and then there are Model-View-Presenter (MVP) frameworks like Apache Spark that can scale linearly but perform inadequately for specific tasks. We believe theres a significant opportunity for a vector store that merges both scalability and performance while maintaining the traditional robustness expected from an enterprise-class database management system.

SingleStore, formerly known as MemSQL, has also developed a hybrid database capable of supporting both Online Transaction Processing (OLTP) and Online Analytical Processing (OLAP) capabilities, and it counts clients like Uber, Kellogg's, and GE among its user base. Recognizing the increasing demand for generative AI, SingleStore has included vector search functionalities since 2017, employing an exact-nearest-neighbor approach. Last year, they showcased an indexed approximate-nearest-neighbor (ANN) search, which they claimed would dramatically enhance vector search speed and simplify application development.

Technical evangelist Akmal Chaudhri stated that SingleStore currently serves 400 customers worldwide, with around 45 of them leveraging the companys generative AI capabilities to some extent, including vector search features. He noted their diverse clientele spans various industries, including those with stringent regulatory requirements, many of whom are developing chatbots tailored for compliance. Organizations today accumulate vast amounts of data without clear strategies for utilization, and vector capabilities empower them to interrogate their data more effectively, he added.

A recent survey by Gartner revealed that 63% of organizations either lack or are uncertain about their data management practices suitable for AI integration. Despite the current enthusiasm for generative AI, Gartner also indicated that other machine learning approaches, such as simulation, may be more effective for certain business challenges, such as planning and forecasting.

Looking ahead, Gartner predicts that the overall Database Management Systems (DBMS) marketa sector that is already well-establishedwill experience growth of around 16% by 2025, potentially reaching a market value of approximately $137 billion. As vendors continue to chase customer investments, it is anticipated that they will strive to showcase how their technologies can address and resolve business challenges through AI solutions.