The News
CData Software used its virtual conference, Foundations 2026, to formally discuss the CData Connect AI Gateway, a platform designed to sit between enterprise AI models and business data systems. The product combines deep connectivity to hundreds of enterprise data sources (Salesforce, NetSuite, SAP, and others), a context engine that encodes business definitions and tacit organizational knowledge, intelligent model routing, and row-level data security and governance controls. The launch was accompanied by a live customer case study from Anaqua CIO Erik Bailey, internal team demonstrations, and a CData Labs research segment presenting benchmarking results showing the platform achieved 98.5% accuracy versus a 65–75% rate from other MCP implementations, and a 178x cost difference between a small and large model when the gateway was properly configured.
Analyst Take
The real problem isn’t which model you pick
The enterprise AI deployment conversation has been dominated by model selection for two years. Which LLM is smarter? Which scores highest on the leaderboard? CData’s message at this event cuts against that framing cleanly, and they’re right to do it. The CData Labs segment made the point explicitly: a model that ranks well on a generalized benchmark may still fail on your specific business data because it doesn’t know what “at risk” means in your company, how your Salesforce customizations relate to your NetSuite contracts, or where the join path is between your CRM and your support ticketing system. Context, not model intelligence, is the deciding variable. That’s a defensible analytical position, not just a marketing claim. The 178x cost differential between a small and large model on the same task, once the gateway was properly configured, is the kind of benchmark number that makes CFOs pay attention.
Why governance is the hidden adoption blocker
The Anaqua case study was more instructive than most vendor-curated customer stories tend to be. Bailey’s framing of write access as fundamentally different from read access (“a hallucination on a read might be embarrassing, but a hallucination on a write—that’s a whole other matter”) captures the practical governance question that most AI deployment conversations skip past. His organization’s approach, running an internal AI transformation office to gate write access requests individually, reflects a real operational pattern: go fast on reads, deliberate on writes. The developer demonstration reinforced this with a concrete example: three lines of configuration change moved a raw Anthropic API connection behind the gateway, adding logging, budget enforcement, PII masking, model routing, and read-only tool scoping, all without changing application code. For developers evaluating AI gateway architectures, that implementation story matters. The control plane is separated from the application layer entirely.
This governance dimension connects to a broader finding from ECI Research’s GovTech survey work. According to ECI Research’s Google GovTech Survey Results, 31.8% of respondents identified “FedRAMP/compliance approval friction for AI vendors” as the single largest blocker preventing widespread AI adoption in their developer workflows. CData is a commercial enterprise vendor, not a government-specific one, but the dynamic is the same in any regulated or risk-sensitive environment: the AI tool itself may be capable, but the enterprise adoption question is always about control, auditability, and accountability. CData is positioning the gateway as the answer to exactly that gap. The full audit trail, the user-scoped tokens, the policy engine with natural-language rule authoring — these are features designed for organizations where someone has to explain every query to a compliance officer.
The MCP sprawl problem is real, and it’s arriving fast
Bailey used the term “MCP sprawl” without prompting, and it’s a signal worth noting. The Model Context Protocol is being adopted quickly, and first-party MCP servers from individual SaaS vendors (HubSpot, for example) are proliferating without centralized oversight. What CData is building is a governance layer that sits above the MCP layer, capable of managing third-party MCP servers under a single policy and logging framework. The value proposition is not that the CData connectors are deeper than a native HubSpot connector (though they claim they are). It’s that consistency, central logging, and permission management are worth accepting some functional trade-off. That’s a real architectural decision enterprises are going to face in the next 12–18 months as agentic deployments multiply. The internal CData team demonstrations (sales coaching from Gong transcripts, daily demand pipeline refresh without BI team involvement, automated month-end close aggregation) showed this is not theoretical; it’s in production.
For ITDMs evaluating AI infrastructure investments, the economic case is straightforward: uncontrolled AI spend is already a problem. The CData executive team cited their own experience with token costs before deploying the gateway. According to ECI Research’s Google GovTech Survey Results, 49.6% of respondents estimated that between 26% and 50% of their organization’s code will be assisted or generated by AI within the next 12 months. Organizations moving that fast on AI-assisted development and agentic workflows need a budget enforcement and routing layer, not just a model API key.
Looking Ahead
CData is entering an AI gateway market that is crowded at the routing and observability layer but notably thinner at the data connectivity and context layer. The differentiation they’re betting on, pre-built semantic knowledge of hundreds of enterprise systems including the “hard” ones like SAP, Oracle, and Workday, is genuinely difficult to replicate quickly. If they can sustain that connectivity moat while adding governance and context features at the pace the Anaqua case study suggests, they have a credible position as the infrastructure layer for enterprise agentic AI. The CData Labs commitment to open benchmarking is a smart move: publishing reproducible datasets and evaluation harnesses that anyone can run is the kind of credibility-building that enterprise buyers, especially in regulated industries, respond to.
The twelve-month risk for CData is that the major cloud providers continue pushing their own native AI integration layers deeper into their respective ecosystems. Microsoft Copilot’s limitations with third-party connectors, which Bailey described as the reason Anaqua moved to Connect AI, may not persist. What CData needs to demonstrate over the coming year is that its cross-system, multi-cloud, vendor-neutral position is more durable than any single vendor’s native integration story. The self-learning context engine, which accumulates organizational tacit knowledge over time and promotes it to shared company context, could be that differentiator: it creates compounding switching costs that raw connectivity alone does not.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
