The integration of generative artificial intelligence into corporate workflows is increasingly exposing a structural bottleneck: raw foundation models are only as effective as the enterprise data they can access. As companies move beyond initial AI pilots, the focus is shifting toward data architecture. Retailer Lowe’s, for instance, has begun deploying semantic layers and knowledge graphs to improve the accuracy of its AI agents, which were developed in partnership with OpenAI over the past two years.

This push for structured data is triggering a territorial dispute among enterprise software providers, with Microsoft, Databricks, and SAP competing for control over the semantic layer. Simultaneously, the capital requirements at the foundation model layer continue to escalate to unprecedented levels. Unverified reports from Crunchbase News indicate that Anthropic is raising a $65 billion Series H round at a valuation approaching $1 trillion. Together, these signals suggest the AI industry is bifurcating: a fierce battle for data management at the application layer, and massive capital concentration at the infrastructure level.

The strategic value of enterprise data architecture

Semantic layers function as critical data management tools that standardize definitions for core business metrics, such as revenue and customer segments. By establishing a single source of truth, these layers enable AI systems to navigate corporate databases more accurately and efficiently. For Lowe’s, the U.S. home improvement retailer, this architecture is already yielding operational dividends. According to Chandhu Nair, a senior vice president at the company, Lowe’s is utilizing its semantic layer alongside knowledge graphs—tools that map the relationships between different data types—to enhance an AI-powered shopping assistant and an internal sales coach for employees.

The realization that proprietary data is the true differentiator in AI deployments is reshaping competitive dynamics across multiple sectors. Microsoft, the tech giant and primary backer of OpenAI; Databricks, the prominent data analytics platform; and SAP, the German enterprise software incumbent, are all aggressively positioning themselves to control this semantic layer. A similar dynamic is unfolding in the legal sector, where major law firms are increasingly developing internal AI capabilities. According to The Information, this internal development poses a direct threat to specialized legal AI startups like Harvey and Legora, as incumbent firms opt to leverage their own highly structured, proprietary data rather than relying entirely on third-party vendors.

Capital consolidation at the foundation layer

While enterprises focus on data structuring, the foundation model layer is experiencing capital demands that rival sovereign budgets. An unverified report from Crunchbase News suggests that Anthropic, the AI research company behind the Claude family of models, is nearing a $1 trillion valuation on the back of a $65 billion Series H funding round. While these figures remain unconfirmed and represent a staggering leap in private market valuations, they underscore the prevailing market consensus that training frontier AI models requires capital at a scale previously unseen in venture capital.

Supporting these massive models also requires deep integration at the infrastructure level. Alongside the capital arms race, cloud providers are continuously optimizing their underlying systems for AI workloads. Unverified reports from InfoQ suggest Microsoft is preparing to announce Azure Linux 4.0, positioning it as a general-purpose server distribution. If accurate, this points to Microsoft further vertically integrating its cloud infrastructure to squeeze out operational efficiencies, a necessary step when supporting the compute-intensive demands of partners like OpenAI and the broader enterprise AI ecosystem.

The current trajectory of enterprise AI points to a dual reality: massive, capital-intensive foundation models at the bottom, and highly specific, proprietary data architectures at the top. As software giants and specialized startups fight for control over the semantic layer, the ultimate utility of generative AI will likely depend on how seamlessly these trillion-dollar models can interface with the structured realities of corporate data.

With reporting from The Information, Crunchbase News, InfoQ

Source · The Information