Between May 5 and June 8, 2026, a pattern emerged that most enterprise AI coverage missed. Dun & Bradstreet signed partnerships with Anthropic, Microsoft, and OpenAI — all three major AI platform providers — in four weeks. S&P Global announced a strategic collaboration with Cohere to bring financial intelligence into Cohere's sovereign AI platform, North. And Snowflake deepened its $200 million partnership with Anthropic at Snowflake Summit 26, reporting accelerating enterprise adoption of Claude in Cortex AI across customers including Basis, Block, Carvana, eSentire, Indeed, and Notion.
Three legacy data companies. Five AI platform partnerships. One architectural thesis: the moat in enterprise AI is not the model. It is the verified, governed, domain-specific data that agents need to act with confidence.
The numbers explain the urgency. 73% of enterprise data leaders rank poor data quality as the number one barrier to AI success — not model accuracy, not compute cost, not talent. Gartner predicts that 60% of AI projects through 2026 will be abandoned specifically because companies lack AI-ready data. The hallucination tax — the cost of wrong AI answers propagating through enterprise workflows — hit an estimated $112 billion in 2025 and is climbing. Enterprises spend an average of 4.3 hours per employee per week verifying AI output, at a cost of roughly $14,200 per employee per year according to Forrester data compiled by Suprmind.
These data companies are not selling AI. They are selling the verified foundation that makes AI trustworthy enough for production. And every major AI platform just decided it needs them.
This article unpacks what each deal means, why data companies are emerging as the most important layer in the enterprise AI stack, and provides two frameworks: the Enterprise AI Data Partner Portfolio Map for identifying which data partnerships your AI stack needs, and the AI-Ready Data Maturity Assessment for evaluating whether your own enterprise data is ready for agentic workloads.
Dun & Bradstreet: One Company, Three AI Platforms, Four Weeks
D&B's strategy is the most aggressive in the batch — and the most revealing. The company did not pick a side in the AI platform war. It signed with all three.
Deal 1: D&B + Anthropic (May 5)
D&B announced a collaboration with Anthropic to bring risk data directly inside Claude via Model Context Protocol (MCP). The integration enables users to create customized KYC/KYB workflows in minutes, automating onboarding processes that previously took days. A financial institution can now use D&B data in Claude to onboard new corporate clients in seconds — automatically verifying identity, ownership structure, risk profile, and generating audit-ready documentation.
Alex Zuck, General Manager of Risk at D&B, framed it precisely: "Claude isn't just being given more data; it's being given the verified context and decision logic required to act. Outputs that are not only personalized but also explainable, auditable, and consistent — all essential for high-stakes, regulated environments."
Use case: Risk and compliance workflows. KYC/KYB onboarding, third-party risk evaluation, audit documentation.
Deal 2: D&B + Microsoft (June 2)
D&B launched a Graph Connector for Microsoft 365 Copilot that feeds the D&B Commercial Graph directly into Microsoft Graph — the data substrate powering Copilot across the entire M365 ecosystem. The connector provides free access to a curated sample of verified business data including company summaries, locations, contact details, employee counts, and annual revenue ranges.
The architectural significance is specific: the D&B Graph Connector makes Copilot query a clean, structured source of truth for commercial identity instead of fabricating it from unstructured web content. This is a direct countermeasure to the hallucination problem — grounding Copilot's business answers in a dataset that performs 100 billion verifications, tests, and checks per month.
Use case: Productivity workflows. Sales prospecting, market research, supplier discovery, partner evaluation.
Deal 3: D&B + OpenAI (June 3)
D&B announced a collaboration with OpenAI enabling users to access the D&B Commercial Graph in ChatGPT and Codex via MCP servers. Financial professionals can now bring verified business identity, ownership, relationship, credit, and risk data directly into their workspace to accelerate due diligence, financial reporting, and credit origination.
Scott Spencer, General Manager of Finance & Credit at D&B, said the integration "helps teams of all sizes, including small and mid-sized businesses, to embrace the power of AI with confidence in their workflows." The D&B Finance Analytics tools are also accessible via MCP server, enabling automated business credit decisions powered by a rules-based engine.
Use case: Finance workflows. Credit origination, due diligence, financial reporting, risk management.
The Pattern
Three deals. Three different AI platforms. Three different enterprise workflow categories. One data asset: the D&B Commercial Graph, anchored by the D-U-N-S Number — the global standard for identifying commercial entities since 1963. D&B is not competing with Anthropic, Microsoft, or OpenAI. It is making itself indispensable to all three by providing the verified business identity layer that every AI agent needs when it operates in regulated, high-stakes enterprise workflows.
The delivery mechanism across all three deals is the same: Model Context Protocol (MCP). D&B has bet its entire AI distribution strategy on MCP as the universal connector between data providers and AI platforms. That bet is paying off — MCP has become the standard integration pattern for enterprise data-to-AI pipelines in 2026.
S&P Global + Cohere: Financial Intelligence Meets Sovereign AI
S&P Global announced a strategic collaboration with Cohere on June 8 to bring its financial data into Cohere's enterprise platform, North. The integration enables financial institutions to run sensitive on-premise workloads directly within North, combining S&P Global's trusted data with their own enterprise data to generate faster, more accurate answers to complex financial questions.
The strategic logic targets a specific segment: regulated financial institutions that need sovereign AI — AI that runs on-premise or in jurisdiction-controlled environments, not in a U.S.-based hyperscaler cloud. Cohere's positioning as a "sovereign AI provider for governments and regulated industries" makes it the distribution channel for S&P Global's data into markets where data residency, compliance, and jurisdictional control are non-negotiable.
Bhavesh Dayalji, Chief AI Officer of S&P Global and CEO of Kensho, described the strategy: "We've done the work on the backend to make our data AI-ready, build the retrieval infrastructure, and partner with best-in-class AI providers, so that customers can simply put S&P Global to work in the platforms they already use."
The technical foundation is the Kensho LLM-ready API — optimized for function calling patterns, citation-backed retrieval, and natural language querying of S&P Capital IQ Financials, earnings call transcripts, and other datasets. Unlike typical APIs, it is designed specifically for AI integration, with a Python library that streamlines authentication and LLM connectivity.
S&P Global's approach mirrors D&B's multi-platform strategy. The Cohere deal builds on what the company calls a broader strategy of "making its data available across the AI platforms that customers use" — platform-agnostic distribution of domain-specific financial intelligence.
Snowflake + Anthropic: $200M and the Governed Data Layer
The Snowflake-Anthropic partnership, announced at Snowflake Summit 26 on June 1, represents the largest dollar commitment in the data-AI partnership wave: $200 million in joint investment established in December 2025, now showing accelerating enterprise traction.
The architecture is different from D&B and S&P Global. Snowflake is not a domain-specific data company. It is the governed data environment — the infrastructure layer where enterprise data already lives, with security, governance, observability, and compliance controls already in place. Claude operates directly on data within Snowflake via Cortex AI, meaning sensitive data never leaves the governed environment.
Christian Kleinerman, EVP of Product at Snowflake, captured the market shift: "Customers want AI that works directly on their governed data, not in isolated systems. Snowflake Cortex Code is becoming the fastest-growing product in Snowflake's history."
The customer list — Basis, Block, Carvana, eSentire, Indeed, Notion — spans cybersecurity investigations, financial analysis, production data apps, and knowledge work. The common thread: all are deploying Claude on governed enterprise data through Snowflake rather than sending data to external AI systems.
Steve Corfield, Head of Global Business Development at Anthropic, framed the value proposition: "Snowflake brings the governed data environment enterprises already rely on, and Claude brings the reasoning to put that data to work."
Why Data Companies Are the New AI Power Brokers
The five partnerships share a structural thesis that is reshaping enterprise AI architecture: AI agents are only as good as the data they can access, and the most valuable data in the enterprise is not the data you can scrape from the web. It is the verified, governed, proprietary data that only specialized providers can deliver.
Consider what happens when an AI agent operates without verified data:
- A compliance agent onboarding a new client fabricates an ownership structure because it has no access to verified corporate hierarchy data. Result: regulatory violation.
- A finance agent running credit analysis hallucinates revenue figures for a counterparty. Result: mispriced risk.
- A sales agent recommends targeting a prospect based on outdated firmographic data. Result: wasted pipeline.
These are not hypothetical failure modes. They are the $112 billion hallucination tax — the aggregate cost of AI systems operating without grounded, verified data. The data companies partnering with AI platforms are directly addressing this tax by inserting a verified layer between the model and the enterprise workflow.
The economics favor the data companies. AI platform providers — Anthropic, OpenAI, Microsoft, Cohere — are competing fiercely on model quality, pricing, and features. That competition compresses their margins and commoditizes their products. Data companies face no such compression. D&B's Commercial Graph, S&P Global's financial intelligence, and Snowflake's governed data environment are each defensible moats built over decades. They cannot be replicated by training a larger model.
Forrester's State of Agentic AI, 2026 found that 75% of enterprises are adopting agentic AI but fewer than 10% have scaled it beyond pilot. The top structural barrier is not technology — it is ROI uncertainty driven by data quality concerns. Data partnerships directly address this barrier by providing production-grade data that makes agents trustworthy enough to operate autonomously.
Framework #1: Enterprise AI Data Partner Portfolio Map
Use this map to identify which data partnerships your AI agent stack needs. Map your current data sources against the six data domains below. Gaps in any domain limit the reliability and scope of your agent deployments.
| Data Domain | What Agents Need | Who Provides It | Integration Pattern | Priority If Missing |
|---|---|---|---|---|
| Business Identity | Verified company identity, ownership, hierarchy, D-U-N-S | Dun & Bradstreet | MCP servers → Claude, ChatGPT, Copilot | Critical for compliance, onboarding, KYC/KYB |
| Financial Intelligence | Market data, credit ratings, earnings, benchmarks, pricing | S&P Global (Kensho), Bloomberg, Refinitiv | LLM-ready APIs, sovereign AI platforms | Critical for finance, risk, investment workflows |
| Governed Enterprise Data | Customer data, operational data, analytics in governed environment | Snowflake, Databricks, Microsoft Fabric | Native AI integration (Cortex AI, Mosaic AI) | Critical for any agent operating on internal data |
| Customer & Revenue Data | CRM records, account history, pipeline, support tickets | Salesforce Data Cloud, HubSpot | Platform-native agents (Agentforce) | Critical for sales, service, marketing agents |
| Procurement Intelligence | Vendor pricing, contract terms, negotiation benchmarks | Vertice (Vendr), Coupa | Platform-native AI agents | Critical for procurement and spend agents |
| Document Intelligence | Invoices, contracts, POs, regulatory filings | Rossum (Coupa), Kofax, ABBYY | IDP integration into workflow platforms | Critical for finance automation |
How to use it: Audit your current AI agent deployments against these six domains. For each domain, ask: "Does our agent have access to verified, governed, up-to-date data for this domain?" If the answer is no, the agent is either hallucinating or producing generic outputs that a competitor's agent — with proper data access — will outperform.
Three portfolio patterns we're seeing:
-
Financial services enterprises need Business Identity (D&B) + Financial Intelligence (S&P Global) + Governed Enterprise Data (Snowflake) as the minimum stack. All three are now available through platform-agnostic partnerships.
-
B2B enterprises need Business Identity (D&B) + Customer & Revenue Data (Salesforce) + Procurement Intelligence (Vertice/Coupa). The D&B-across-all-platforms strategy means business identity data is available regardless of which AI platform you chose.
-
Regulated enterprises add Document Intelligence and Governed Enterprise Data as non-negotiable requirements. Sovereign AI providers like Cohere become the preferred platform because data never leaves jurisdiction-controlled environments.
Framework #2: AI-Ready Data Maturity Assessment
Before partnering with external data providers, assess whether your own enterprise data is ready for agentic AI workloads. Score each criterion 1–5 (1 = not started, 5 = production-grade). The total score indicates your maturity level and next steps.
| # | Criterion | What to Assess | Score 1–5 |
|---|---|---|---|
| 1 | Identity Resolution | Can your systems uniquely identify entities (companies, people, products) across all internal datasets with a single persistent identifier? | |
| 2 | Data Freshness | How current is your data? Real-time (<1 hour), daily, weekly, or "whenever someone updates the spreadsheet"? Agents operating on stale data make stale decisions. | |
| 3 | Schema Consistency | Are your data schemas consistent across systems, or does "customer" mean something different in your CRM, ERP, and data warehouse? AI agents cannot reason across inconsistent schemas. | |
| 4 | Citation & Provenance | Can your data layer provide source citations for every data point? Agents in regulated environments must show their work. If your data cannot tell an auditor where a number came from, the agent cannot either. | |
| 5 | Access Governance | Are permissions inherited automatically, or do you maintain a separate access control layer for AI? Agents should see only the data their operator is authorized to access — no more, no less. | |
| 6 | API Readiness | Is your data accessible via function-calling APIs optimized for LLM integration (like Kensho's LLM-ready API), or do agents need to scrape dashboards and parse CSVs? | |
| 7 | Verification Cadence | How often is your data verified against authoritative sources? D&B runs 100 billion checks per month. What is your equivalent? |
Scoring:
- 28–35: Production-ready. Your data can support agentic AI at scale. Focus on external data partnerships to fill domain gaps.
- 21–27: Pilot-ready. Your data supports experimental deployments but will create trust issues at scale. Prioritize citation, provenance, and API readiness.
- 14–20: Foundation work needed. Invest in identity resolution and schema consistency before deploying agents beyond basic Q&A.
- 7–13: Pre-AI. Your data infrastructure was designed for human interpretation, not machine consumption. The Gartner prediction that 60% of AI projects will be abandoned due to data quality applies directly to this tier.
The action rule: External data partnerships (D&B, S&P Global, Snowflake) fill domain gaps. Internal data maturity fills trust gaps. You need both. A D&B partnership does not fix your internal data quality problems, and fixing your internal data does not give your agents access to 100 billion verified business checks per month. The highest-performing enterprises in 2026 are doing both simultaneously.
The MCP Factor
One technical detail connects all three D&B partnerships and deserves specific attention: Model Context Protocol (MCP).
D&B's Anthropic integration uses MCP. The OpenAI integration uses MCP servers for ChatGPT and Codex. The Microsoft integration uses Graph Connector (Microsoft's equivalent bridging pattern). In every case, the architectural pattern is the same: a standardized protocol that lets data providers expose verified data to AI platforms without building custom integrations for each one.
MCP is becoming the USB standard of enterprise AI data — the universal connector that allows any data provider to plug into any AI platform. For enterprise architects, this means the data partnership landscape is becoming composable: you can mix and match data providers and AI platforms without lock-in to either side.
For data companies, MCP is the distribution layer that makes the multi-platform strategy viable. D&B did not build three separate integrations with three separate architectures. It built MCP servers once and connected them to Claude, ChatGPT, and Codex with minimal incremental effort. That efficiency is what made the "four weeks, three deals" pace possible.
What This Means for Enterprise AI Strategy
The data company partnership wave inverts a common assumption in enterprise AI planning. Most enterprises start with the model: "Should we use Claude, GPT, or Gemini?" Then they layer data on top. The partnerships announced in the past five weeks suggest the market is flipping: start with the data, then let the model be interchangeable.
D&B's strategy proves the point. The same verified business data — the same Commercial Graph, the same D-U-N-S Numbers, the same 100 billion monthly verifications — is now accessible through Claude, ChatGPT, Copilot, and Codex. The model is the variable. The data is the constant.
For CIOs and Chief Data Officers, three actions in Q3 2026:
1. Audit your data partner portfolio against the six-domain map. If your agents are operating without verified business identity data, financial intelligence, or governed enterprise data access, they are hallucinating in domains where hallucination has material business consequences.
2. Run the AI-Ready Data Maturity Assessment on your internal data. The external partnerships only work if your internal data is ready to be combined with them. A D&B integration connected to an internal CRM with inconsistent entity resolution will produce worse results than no integration at all.
3. Invest in MCP-native data infrastructure. The three D&B deals prove that MCP is the winning distribution pattern for data-to-AI integration. If your enterprise data is not accessible via MCP or equivalent function-calling APIs, it is invisible to the AI agent stack — no matter how good that data is.
The model is commoditized. The execution layer is being acquired. The context layer is being built. But the data layer — the verified, governed, domain-specific intelligence that gives every other layer its credibility — is where the durable value lives. The companies that own that data are the new power brokers of enterprise AI.
Continue Reading
-
5 Acquisitions in 14 Days. The AI Agent Stack Is Being Bought. — How Asana, Salesforce, Coupa, Vertice, and Palo Alto Networks are acquiring the execution layer for AI agents.
-
Microsoft IQ Is GA. The Enterprise Agent Context War Just Reset. — Microsoft's four-layer unified context system and what it means for the agent grounding substrate.
-
78% Can't Pass an AI Audit. Alation Just Made It a Score. — Why data governance is the prerequisite for every AI agent reaching production.
Rajesh Beri is Head of AI Engineering at Zscaler.