What’s most disconcerting about the AI outlook for 2026 is that the vast majority of organizations involved are still merely wielding the buzzword equivalent of a hammer. In the 2010s, the buzzword was “deep learning.”  In 2022, the buzzword became “generative AI.” In 2024, “agentic AI” became the term of art.

Most AI solutions today are merely rooted in the same statistical machine learning approach that’s been predominant since deep learning was popularized during the 2010s. Most enterprises don’t seem to want to incorporate complementary approaches.  Passivity and a lack of due diligence are among the reasons.

I’m not saying statistical machine learning is unnecessary; that’s hardly the case. But it’s certainly not sufficient.  Many seem entirely unaware that hybrid or “neurosymbolic” AI (a blending of probabilistic and deterministic, as well as neural net plus knowledge representation) exists and has been benchmarked to be much more accurate. 

It’s common sense that companies need to build a proper data foundation for AI, including contextually rich relationships. Which is why “contextual engineering” has also become a buzzword.

AI as narrowly and monolithically defined has become the hammer. Therefore, every problem becomes a nail. Including data problems.

Algorithms on their own, as I’ve said before, don’t solve data problems. Data is the prime mover, and deep learning algorithms are black boxes. Data determines the shape of the algorithms. If the input is garbage, the output will be garbage also.

40+ years of relative data layer inertia

Part of the issue is that most enterprises aren’t taking their data problems seriously enough. This is despite the fact that Gartner has consistently bemoaned the lack of AI-ready data in enterprises and points out in its survey reports that 70 percent of companies are unhappy with the return on investment in what they think of as AI.

In that sense, the information technology mindset is still just as monolithic now as it was in the 1980s. Enterprises locked in on relational database management systems (RDBMSes) for data storage, logic in application code. The focus was on tabular, transactional data and narrowly defined BI. 

In the 1990s, web technology threatened to change the way enterprises did things at the data layer. But vendors came up with workarounds instead. MySQL became the de facto standard RDBMS for web app development.

By the 2000s, Google and its search engine peers moved away from relational DBMSes…for some things. Quite a bit of experimentation with NoSQL databases took place. Plenty of alternatives to RDBMSes exist, and there’s plenty of competition for niches every year. But the graph DBMSes that can solve the hard problems with AI-ready data continue to be underused.

I’m not saying RDBMSes should be replaced, just that they’re overused. That’s particularly the case when it comes to large-scale data integration and data, content, and knowledge management, which should be the exclusive province of graph DBMSes considering their potential for scaling and relationship richness.

I checked the DB Engines site for their current ranking. 12 of the top 20 ranked DBMSes are relational. Six of the top 20 are “multi-model”, which means relational, document and/or key-value.  One is a vector DBMS. And only one, being number 20 out of 20 in the ranking, is a graph DBMS. But not a graph DBMS that harnesses the full power of specific relationships and standards.

Graph DBMSes: 60+ years later, a Tool That’s Still Missing from Most Toolboxes

The need for graph database management systems (or network DBMSes, as they were called in the 1960s) and logically connecting semantic metadata that assumes a graph data model hasn’t changed. AI-ready data demands these capabilities. Gartner continues to emphasize the point.

But there’s an immediate obstacle. The audience not versed in DBMS alternatives doesn’t often understand what aficionados mean by a “graph database”. Quite a few aren’t used to thinking in graphs or haven’t studied graph theory at all. 

As a result, I’ve been borrowing the term “network database” to convey more directly the notion of nodes and edges or any-to-any connections. It’s a term that goes back to 1963.

Back then, Charles Bachman designed and built the Integrated Data Store (IDS) when he was at General Electric. IDS, according to computing history professor Thomas Haigh of the University of Wisconsin, was the first DBMS. 

In essence, this network or graph DBMS was the first DBMS. Haigh includes this illuminating figure in his Communications of the ACM article that came from a presentation Bachman gave at a conference. IDS wasn’t an any-to-any graph DBMS in terms of individual entities, but it was designed to be much more relationship rich than the RDBMSes that followed.

2023-2025’s Silver Lining

What’s been surprising considering this relative data layer inertia is that semantics technology company mergers and acquisitions have picked up significantly in recent years. And major incumbents are making a number of these acquisition commitments. 

In October 2025, Steve Hedden of TopQuadrant posted on this M&A, noting that “The rise of GraphRAG reignited interest in knowledge graphs akin to when Google launched its Knowledge Graph in 2012. The sudden demand for structured context and explainable retrieval gave them new relevance.” 

In the post, Hedden listed evidence as to how much interest major technology providers such as Samsung and ServiceNow have demonstrated in the coming together of probabilistic machine learning and deterministic semantic graph databases:

• “January 23, 2023 — Digital Science acquired metaphacts, creators of the metaphactory platform: “a platform that supports customers in accelerating their adoption of knowledge graphs and driving knowledge democratization.” 
• February 7, 2023 — Progress acquired MarkLogic in February of 2023. MarkLogic is a multimodal NoSQL database, with a particular strength in managing RDF data, the core data format for graph technology.
• July 18, 2024 — Samsung acquired Oxford Semantic Technologies, makers of the RDFox graph database, to power on-device reasoning and personal knowledge capabilities.  
• October 23, 2024 — Ontotext and Semantic Web Company merged to form Graphwise, explicitly positioning around GraphRAG. “The announcement is significant for the graph industry, as it elevates Graphwise as the most comprehensive knowledge graph AI organization and establishes a clear path towards democratizing the evolution of Graph RAG as a category.” 
• May 7, 2025 — ServiceNow announced its acquisition of data.world, integrating a graph-based data catalog and semantic layer into its enterprise workflow platform.”

2026: Hope Springs Eternal, 16 Years Later

I’ve been exploring the same basic semantic graph value theme since 2009, when I led the development of this issue of PwC’s Technology Forecast on web semantics and its value in supporting broader BI, BI that goes beyond tabular data.

In 2018, when I was still at PwC, I was invited to deliver a keynote on current trends at the Semantics conference in Vienna. During the keynote I talked quite a bit about AI and the need for the Symbolist tribe (those who supported knowledge representation including knowledge graphs and other related methods of linked together facts) to work together more closely with the Connectionist tribe (those focused on neural networks and deep learning).

It’s a bit disconcerting to think back to previous decades and realize how much things really  haven’t changed, at least when it comes to the data layer and how most enterprises are managing it. But I’m optimistic, especially considering that “contextual engineering” seems to be becoming a buzzword. That’s right in my wheelhouse. 

For More Information

Haigh, Thomas. “Historical Reflections: How Charles Bachman Invented the DBMS, a Foundation of Our Digital World.” Communications of the ACM, vol. 59, no. 7, July 2016, pp. 26-31, https://doi.org/10.1145/2935880. 

Thomas Haigh’s article explores the creation of the Integrated Data Store (IDS) in 1963 at General Electric. It highlights how Bachman’s practical engineering background led to the first true database management system, which introduced essential concepts like metadata, data independence, and transaction processing. 

Haigh points out in this ACM article that “IDS and CODASYL systems (a standards body that also supported the COBOL language) did not use the relational data model, formulated years later by Ted Codd, which underlies today’s dominant SQL database management systems. Instead it introduced what would later\ be called the ‘network data model.’ This encoded relationships between different kinds of records as a graph….” 

Hedden, Steve. “Is RAG Dead? The Rise of Context Engineering and Semantic Layers for Agentic AI.” Towards Data Science, 21 Oct. 2025, towardsdatascience.com/is-rag-dead-the-rise-of-context-engineering-and-semantic-layers-for-agentic-ai-7a3b3e2d6b1d.

Steve Hedden’s article describes the evolution of Retrieval-Augmented Generation (RAG) into context engineering for agentic AI. Standard RAG is maturing into governed systems using knowledge graphs and semantic layers to link disparate enterprise data. Future retrieval will be an iterative “reasoning loop” incorporating multimodal data, tool metadata, and agent memories to ensure AI is explainable, trustworthy, and semantically grounded.

Petkova, Teodora. “Okay, You Got a Knowledge Graph Built with Semantic Technology… and Now What?” Graphwise, 9 Dec. 2024, https://graphwise.ai/blog/okay-you-got-a-knowledge-graph-built-with-semantic-technology-and-now-what/.

This blog post from Teodora Petkova at Graphwise explores how semantic knowledge graphs break data silos by assigning machine-readable meaning to disparate sources. Beyond just building a graph, the technology enables semantic data integration, annotation, and search. These capabilities allow enterprises to analyze complex relationships, detect patterns, and infer new facts, ultimately powering better automation and discoverability in fields like healthcare, pharmaceuticals, and business intelligence.

Includes a case study description with a knowledge graph illustration: “In 2017 NuMedii wanted to build an expert knowledge graph with concepts from genomics, proteomics, metabolomics, disease conditions, drug products, scientific literature and various biomedical ontologies, integrated information from more than 20 open data sets.This massive integration helped the enterprise access highly normalized and semantically interlinked data, discover knowledge locked in documents and identify patterns and correlations between biomedical concepts.” I’ve used this same illustration in several talks over the years:

Graphwise/NuMedii, 2024

PricewaterhouseCoopers. Technology Forecast: Spinning a Data Web. no. 2, 2009, https://www.pwc.com/cl/es/publicaciones/assets/pronostico-de-tecnologia.pdf.

The Spring 2009 Technology Forecast on the Semantic Web and Linked Data technologies I led the development of answered the research question, “How do enterprises fix business intelligence?”  It explored how standards like RDF, SPARQL, and ontologies could revolutionize enterprise data management by enabling “Web-scale” data federation without moving information from its original source.

Key themes included:

• Overcoming Silos: Traditional relational databases create rigid silos; Linked Data provides a flexible graph-based alternative that improves decision-making by linking disparate internal and external sources.
• Ontology-Driven Scalability: While requiring more front-end effort, ontologies allow businesses to manage shared meaning across domains, making them more scalable than traditional integration.
• Industry Use Cases: Interviews with leaders from the BBC, Chevron, and M. D. Anderson Cancer Center illustrated practical applications in media, oil and gas, and healthcare.
• CIO Strategy: PwC recommended CIOs lead the development of a “business ontology” to bridge the gap between strategy and operations.