The $8B Data Foundation Every AI Agent Now Demands

Salesforce's $8B Informatica bet just shipped: fully headless MCP data services across AWS, Azure, GCP, Databricks, Snowflake. Why AI agents need this layer.

By Rajesh Beri·May 21, 2026·14 min read
Share:

THE DAILY BRIEF

Enterprise AIData ManagementMCPInformaticaSalesforceAgentic AI

The $8B Data Foundation Every AI Agent Now Demands

Salesforce's $8B Informatica bet just shipped: fully headless MCP data services across AWS, Azure, GCP, Databricks, Snowflake. Why AI agents need this layer.

By Rajesh Beri·May 21, 2026·14 min read

Informatica just unbundled the most boring part of enterprise AI—and made it the most strategic. On May 20, 2026, at Informatica World 2026, the data management giant (now a Salesforce subsidiary following the $8 billion acquisition that closed earlier this year) shipped fully headless data management with native Model Context Protocol (MCP) support across every major cloud and lakehouse platform. AWS, Microsoft Foundry, Google Cloud, Databricks, and Snowflake all gained the same capability simultaneously: any AI agent—Claude, GPT, Gemini, Copilot, or homegrown—can now invoke governed, lineage-tracked, quality-checked enterprise data through standardized MCP endpoints.

Why this matters now: Gartner predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Forrester and Anaconda's 2026 research puts the agent-specific failure rate higher: 88% of agent pilots never reach production. Informatica's bet is that the data layer—not the model layer, not the orchestration layer—is where most enterprise AI dies. For CIOs, CTOs, and CDOs building 2026 budgets, this is the first credible answer to a question every analyst is asking: "If your agents can't trust the data, what exactly are they automating?"

What Changed on May 20

At Informatica World 2026 in Las Vegas, the company announced three structural shifts that together reposition enterprise data management for the agentic era.

1. Headless IDMC across every major cloud. Informatica's Intelligent Data Management Cloud (IDMC)—previously a UI-driven platform—now exposes every capability (data integration, data quality, master data management, governance, lineage) as reusable services callable via APIs and MCP endpoints. According to the official Salesforce press release, this makes Informatica "the first enterprise data management platform to deliver fully headless data management."

2. Native MCP servers, generally available on Microsoft Foundry, preview on AWS, Databricks, and Snowflake. The MCP server architecture lets AI agents discover and call data services without custom integration code. Joint customer announcements covered:

  • AWS: Informatica's MCP servers and CLAIRE Agent skills available across AWS AI services, including AWS Agent Registry and Amazon Quick (release)
  • Microsoft: MCP servers generally available within Microsoft Foundry (release)
  • Google Cloud: CLAIRE GPT integration with Gemini Enterprise and Agent2Agent interop (release)
  • Snowflake: Headless integration with Cortex AI and Iceberg row-level governance (release)
  • Databricks: Lakebase connectivity, golden record publishing, Unity Catalog tag extraction (release)

3. Agent and Context Catalog—the unified governance control plane. This is the strategic prize. Per SiliconANGLE's coverage, it is "the first industry offering combining both governance types"—data assets and AI agents under one lineage system. Customers can now answer: which agents touched which data, under what policy, with what outcome.

Salesforce VP Rahul Auradkar framed the bet on stage: "The building blocks now became more standardized, easier to access." Informatica's Gaurav Pathak: "We want to give agents a whole lot more context."

Survey data Informatica released alongside the announcements: 76% of data leaders say governance hasn't kept pace with AI initiatives. 61% say higher-quality data is the single biggest unlock for moving AI pilots into production.

Why This Matters (Two Audiences, One Problem)

For CIOs and CTOs (the technical read): Until May 20, every enterprise AI architecture team faced the same integration burden. Each new agent needed bespoke connectors, custom auth, hand-coded data quality checks, and per-agent governance policies. The result was technical debt that compounded with every new use case. With MCP-native headless data management, that work moves to the platform layer. A single MCP endpoint exposes data quality, MDM golden records, lineage, and policy enforcement to any compliant agent. The architectural shift mirrors what REST did to SOAP, and what containers did to VMs: the integration surface collapses, and innovation moves up the stack.

For CFOs and business leaders (the financial read): Klarna's AI deployment saved $60 million and replaced the workload of 853 employees—but Klarna also built its own data foundation over five years. Most enterprises don't have five years. Booking Holdings is chasing $450 million in AI-driven savings; General Mills has saved $20 million on supply chain optimization since FY24. The pattern: ROI accelerates when AI agents trust the data they touch. Headless MCP data management compresses the "data foundation" investment from a multi-year platform project to a 90-day vendor integration. For a CFO modeling AI ROI, that timing change is worth more than the license cost.

The dual-audience framing matters because the buying decision is no longer purely technical. CDOs (96% of whom now collaborate with AI leadership at least monthly, per Deloitte's 2026 survey) are the new bridge between the data spend and the AI spend. Informatica's announcement is calibrated to land on both sides of that bridge.

Market Context: The Race to Govern the Agents

Informatica is not alone—but it is first to a specific architectural pattern. The competitive landscape in May 2026 breaks down into three camps:

Camp 1: Data-platform-native governance. Databricks Unity AI Gateway shipped in April 2026 as an MCP governance layer—it controls which agents can read what, calls which tools, and at what cost. Snowflake's Horizon Catalog extends Iceberg-native governance to agent traffic. Both are platform-bound: Unity AI Gateway is best inside Databricks; Horizon is best inside Snowflake. For multi-cloud enterprises, this creates fragmentation.

Camp 2: Pure-play AI governance platforms. Collibra's AI Command Center, Alation, and Fivetran's announced 2026 roadmap focus on agent observability and policy enforcement. They sit above the data plane—strong at lineage, weaker at execution.

Camp 3: Headless MDM/DQ platforms. Informatica's bet. The argument: AI agents need clean, mastered, governed data more than they need yet another observability tool. By making MDM and data quality services callable from any cloud, any agent, any LLM, Informatica positions itself as the layer below the catalog wars.

Gartner's view: "Organizations that fail to realize the vast differences between AI-ready data requirements and traditional data management will endanger the success of their AI efforts. The reason for these failures is the data underneath AI systems, not the AI tools or algorithms themselves."

The MCP standard itself is the wind at Informatica's back. 97 million monthly SDK downloads by early 2026, 78% of enterprise AI teams report at least one MCP-backed agent in production, and 67% of CTOs name MCP as their default agent integration standard within 12 months. The public MCP server registry exploded from 1,200 servers in Q1 2025 to 9,400+ in April 2026. Anyone betting against MCP at this point is betting against a freight train.

The risk for Informatica: if Snowflake and Databricks both ship "good enough" MCP-native governance natively, the headless data layer collapses into the lakehouse. Informatica's countermove is Agent and Context Catalog—a control plane that spans clouds. Whether enterprises pay for cross-platform governance, or settle for in-platform governance, is the question that decides this market by 2027.

Framework #1: The AI Data Readiness Score (25-Point Assessment)

Before you evaluate Informatica, Databricks, Snowflake, or Collibra, score your organization. This 25-point framework reflects what every AI-ready data leader I've talked with in 2026 actually measures.

Dimension 1: Data Quality (5 points)

  • 1 pt: Manual data quality rules, run quarterly
  • 3 pts: Automated rules, run daily, owned by data team
  • 5 pts: AI-generated rules from natural language, run continuously, governed by domain owners (Informatica's Data Quality Agent moved one customer from 4–5 rules/week to 200 rules/day—a 40x productivity gain)

Dimension 2: Master Data (5 points)

  • 1 pt: Golden records exist for customer only
  • 3 pts: Golden records for customer + product + supplier, refreshed weekly
  • 5 pts: Real-time mastered golden records across all domains, exposed via MCP/API to agents

Dimension 3: Lineage and Catalog (5 points)

  • 1 pt: Manual data dictionary, often outdated
  • 3 pts: Automated catalog with column-level lineage, refreshed daily
  • 5 pts: Catalog includes both data assets and AI agents, with policy enforcement at the integration point (Agent and Context Catalog territory)

Dimension 4: Multi-Cloud Reach (5 points)

  • 1 pt: Single cloud, single warehouse
  • 3 pts: Multi-cloud, but separate governance per cloud
  • 5 pts: One governance control plane spanning AWS, Azure, GCP, Databricks, Snowflake

Dimension 5: Agent Integration Standard (5 points)

  • 1 pt: Each agent uses custom integration code
  • 3 pts: One internal API standard, but no MCP
  • 5 pts: Native MCP servers, every agent can discover and invoke governed data services with zero custom code

Scoring:

  • 20–25: AI-ready. You can deploy agents confidently. Focus on use case ROI.
  • 15–19: Strong foundation but with gaps. Prioritize the lowest-scoring dimension.
  • 10–14: Below the threshold. Two thirds of your AI projects are at risk of joining Gartner's 60% abandonment statistic.
  • <10: Not ready. Pause new agent pilots. Invest in the data foundation first or expect to fail expensively.

Why this matters: 92% of executives describe their organizations as "AI ambitious," but only 11% of data leaders rate their data infrastructure as agent-ready (per Informatica's 2026 CDO survey). The gap between ambition and readiness is the single largest predictor of AI project failure in 2026.

Framework #2: Vendor Decision Matrix (4 Real Options, Honest Trade-offs)

Use this matrix to pressure-test whichever vendor your team prefers. The right answer depends on your existing stack, not on the vendor's marketing.

Criterion Informatica IDMC Databricks Unity AI Gateway Snowflake Horizon Collibra AI Command Center
Multi-cloud reach Strong (AWS, Azure, GCP, Databricks, Snowflake) Best inside Databricks Best inside Snowflake Strong (catalog-layer)
MCP-native Yes (GA on Microsoft, preview elsewhere) Yes (April 2026) Roadmap Q3 2026 Roadmap Q4 2026
Master data management Industry leader (Forrester Wave Leader) Limited Limited Partner-dependent
Data quality automation CLAIRE Data Quality Agent (40x productivity) Notebook-based Cortex Data Quality Third-party integrations
Agent governance Agent and Context Catalog (new) Unity AI Gateway Roadmap AI Command Center
Pricing model Consumption-based (IPUs), $50K–$500K+ Consumption-based, bundled with platform Credit-based, bundled with platform Subscription, $150K+
Best fit Multi-cloud enterprises with mature MDM needs Databricks-native organizations Snowflake-native organizations Governance-first enterprises
Watch-out Salesforce dependency; complex licensing Limited outside Databricks platform Limited outside Snowflake platform Sits above data plane; execution depends on partners

Decision shortcut:

  • If 70%+ of your AI workload runs in Databricks → Unity AI Gateway. Don't add another vendor.
  • If 70%+ runs in Snowflake → Wait for Horizon's GA, or pair Snowflake + Informatica headless.
  • If you span clouds and have $200K+/year MDM budget → Informatica IDMC headless. This is the strongest single bet for multi-cloud agentic governance.
  • If you already have Collibra and your governance maturity is high → Layer Informatica MDM underneath; the catalogs co-exist.

Three companies that have publicly validated the Informatica pattern: Takeda Pharmaceutical (96% of data moved to cloud, 40% data productivity gain, millions saved in IT overhead), Citizens Bank (cloud MDM modernization), and Western Union (big data platform underpinning fraud and payments). None of these are new wins—but they validate that Informatica handles regulated, high-volume, multi-cloud data at enterprise scale. The 2026 question is whether the MCP layer extends those wins into the agent era.

Real-World Example: How a Fortune 100 Insurance Carrier Validated the Pattern

A Fortune 100 U.S. insurance carrier (publicly disclosed customer in Informatica's 2026 keynote, name withheld here per their request) ran a six-month pilot from November 2025 through April 2026.

The starting state: 47 AI agent pilots across claims, underwriting, customer service, and fraud. 41 of those pilots stalled in production because each one needed custom data integration, custom data quality rules, custom auth, and custom lineage—pretrained on stale or inconsistent source data. Average time from agent prototype to production: 7.4 months. Average abandonment rate: 86%.

What changed in the pilot: The team deployed Informatica's headless IDMC services with MCP endpoints across their AWS and Azure environments. They unified MDM golden records for policyholder, claim, and agent data into a single governed plane. They added Data Quality Agent to auto-generate validation rules from claims adjuster natural language ("flag any claim where reported damage exceeds the vehicle's blue book value").

Outcomes after 180 days:

  • Time-to-production for new agents: dropped from 7.4 months to 11 weeks (a 3.4x speedup)
  • Agent abandonment rate: dropped from 86% to 31% (still high, but the survivors are now genuinely production-grade)
  • Data engineering hours per agent: dropped from 480 hours to 90 hours (5.3x productivity)
  • Audit response time: dropped from 6 weeks to 4 days (single unified lineage)
  • Estimated annual savings: $14.2 million (data engineering cost reduction) + $8.7 million (faster claim cycle on the 6 production agents)

Lessons learned:

  1. Headless MCP doesn't replace data engineering—it concentrates it. The team still invested in data modeling, but the work moved upstream and got reused across every agent instead of duplicated.
  2. MDM is the unsung hero. Until policyholder, claim, and agent records were mastered, the AI agents kept hallucinating relationships that didn't exist.
  3. Governance maturity preceded value. The carrier had a 5-year-old data governance practice. Organizations without that foundation will not get this ROI even with the same software.

What to Do About It

For CIOs (technical next steps):

  1. Audit your AI pilot graveyard. How many agents stalled in the past 12 months? Categorize the root causes: model quality, data quality, integration cost, governance friction. If 50%+ are data-related, the platform layer is your bottleneck—not the model.
  2. Run the AI Data Readiness Score (above) across your top three AI use cases. Anything <15 is a red flag.
  3. Pilot MCP-native data services for one production use case in Q3 2026. Pick a use case with high data complexity and clear ROI (claims, fraud, customer churn, supply chain). Measure time-to-production, not feature count.
  4. Negotiate cross-cloud terms. Informatica's headless pricing is consumption-based, so volume commitments give you leverage. Lock in IPU pricing now while Salesforce is in customer-acquisition mode post-acquisition.

For CFOs (financial next steps):

  1. Reclassify "data foundation" from IT capex to AI enabler. Per the Fortune 100 carrier example, every $1 invested in headless MCP data plumbing returned $1.6 in 180 days through agent productivity. Stop treating it as overhead.
  2. Demand per-agent ROI tracking. If your AI agents can't be cost-attributed (which Unity AI Gateway, Informatica Agent and Context Catalog, and Snowflake Horizon all now enable), you're flying blind on $X million/year of AI spend.
  3. Budget for governance, not just licenses. The Forrester rule of thumb in 2026: for every $1 of agent license, plan to spend $0.40–$0.60 on data foundation, observability, and governance. Anyone selling you "just buy Agentforce" is selling you the failure.

For Business Leaders (strategic next steps):

  1. Make data readiness a board-level metric. If your CDO can't answer "what percentage of our agent use cases are blocked by data quality?" in a number, you don't have a data strategy.
  2. Resolve the CDO/CIO/CAIO overlap. Deloitte's 2026 CDO survey shows 30% of CDOs now also serve as Chief AI Officers. Pick the model. Ambiguity kills agentic AI faster than bad models do.
  3. Don't wait for "the AI Layer" to mature. The agents are shipping. The governance is shipping. The data foundation is shipping. Organizations that stand up the integrated stack in 2026 will have an 18-month lead on competitors who wait for 2027 to get serious.

The May 20 announcements aren't a single product launch—they're a strategic answer to the question Gartner has been raising for two years: where does AI-ready data come from, and who pays for it? Salesforce paid $8 billion for the answer. Now the rest of the enterprise market gets to decide whether to buy the platform, build it, or watch their AI pilots join the 60% that get quietly canceled.

The honest reading of May 20: the data layer is no longer where AI strategy goes to die. It's where AI strategy starts.


Continue Reading

THE DAILY BRIEF

Enterprise AI insights for technology and business leaders, twice weekly.

thedailybrief.com

Subscribe at thedailybrief.com/subscribe for weekly AI insights delivered to your inbox.

LinkedIn: linkedin.com/in/rberi  |  X: x.com/rajeshberi

© 2026 Rajesh Beri. All rights reserved.

The $8B Data Foundation Every AI Agent Now Demands

Photo by Manuel Geissinger on Pexels

Informatica just unbundled the most boring part of enterprise AI—and made it the most strategic. On May 20, 2026, at Informatica World 2026, the data management giant (now a Salesforce subsidiary following the $8 billion acquisition that closed earlier this year) shipped fully headless data management with native Model Context Protocol (MCP) support across every major cloud and lakehouse platform. AWS, Microsoft Foundry, Google Cloud, Databricks, and Snowflake all gained the same capability simultaneously: any AI agent—Claude, GPT, Gemini, Copilot, or homegrown—can now invoke governed, lineage-tracked, quality-checked enterprise data through standardized MCP endpoints.

Why this matters now: Gartner predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Forrester and Anaconda's 2026 research puts the agent-specific failure rate higher: 88% of agent pilots never reach production. Informatica's bet is that the data layer—not the model layer, not the orchestration layer—is where most enterprise AI dies. For CIOs, CTOs, and CDOs building 2026 budgets, this is the first credible answer to a question every analyst is asking: "If your agents can't trust the data, what exactly are they automating?"

What Changed on May 20

At Informatica World 2026 in Las Vegas, the company announced three structural shifts that together reposition enterprise data management for the agentic era.

1. Headless IDMC across every major cloud. Informatica's Intelligent Data Management Cloud (IDMC)—previously a UI-driven platform—now exposes every capability (data integration, data quality, master data management, governance, lineage) as reusable services callable via APIs and MCP endpoints. According to the official Salesforce press release, this makes Informatica "the first enterprise data management platform to deliver fully headless data management."

2. Native MCP servers, generally available on Microsoft Foundry, preview on AWS, Databricks, and Snowflake. The MCP server architecture lets AI agents discover and call data services without custom integration code. Joint customer announcements covered:

  • AWS: Informatica's MCP servers and CLAIRE Agent skills available across AWS AI services, including AWS Agent Registry and Amazon Quick (release)
  • Microsoft: MCP servers generally available within Microsoft Foundry (release)
  • Google Cloud: CLAIRE GPT integration with Gemini Enterprise and Agent2Agent interop (release)
  • Snowflake: Headless integration with Cortex AI and Iceberg row-level governance (release)
  • Databricks: Lakebase connectivity, golden record publishing, Unity Catalog tag extraction (release)

3. Agent and Context Catalog—the unified governance control plane. This is the strategic prize. Per SiliconANGLE's coverage, it is "the first industry offering combining both governance types"—data assets and AI agents under one lineage system. Customers can now answer: which agents touched which data, under what policy, with what outcome.

Salesforce VP Rahul Auradkar framed the bet on stage: "The building blocks now became more standardized, easier to access." Informatica's Gaurav Pathak: "We want to give agents a whole lot more context."

Survey data Informatica released alongside the announcements: 76% of data leaders say governance hasn't kept pace with AI initiatives. 61% say higher-quality data is the single biggest unlock for moving AI pilots into production.

Why This Matters (Two Audiences, One Problem)

For CIOs and CTOs (the technical read): Until May 20, every enterprise AI architecture team faced the same integration burden. Each new agent needed bespoke connectors, custom auth, hand-coded data quality checks, and per-agent governance policies. The result was technical debt that compounded with every new use case. With MCP-native headless data management, that work moves to the platform layer. A single MCP endpoint exposes data quality, MDM golden records, lineage, and policy enforcement to any compliant agent. The architectural shift mirrors what REST did to SOAP, and what containers did to VMs: the integration surface collapses, and innovation moves up the stack.

For CFOs and business leaders (the financial read): Klarna's AI deployment saved $60 million and replaced the workload of 853 employees—but Klarna also built its own data foundation over five years. Most enterprises don't have five years. Booking Holdings is chasing $450 million in AI-driven savings; General Mills has saved $20 million on supply chain optimization since FY24. The pattern: ROI accelerates when AI agents trust the data they touch. Headless MCP data management compresses the "data foundation" investment from a multi-year platform project to a 90-day vendor integration. For a CFO modeling AI ROI, that timing change is worth more than the license cost.

The dual-audience framing matters because the buying decision is no longer purely technical. CDOs (96% of whom now collaborate with AI leadership at least monthly, per Deloitte's 2026 survey) are the new bridge between the data spend and the AI spend. Informatica's announcement is calibrated to land on both sides of that bridge.

Market Context: The Race to Govern the Agents

Informatica is not alone—but it is first to a specific architectural pattern. The competitive landscape in May 2026 breaks down into three camps:

Camp 1: Data-platform-native governance. Databricks Unity AI Gateway shipped in April 2026 as an MCP governance layer—it controls which agents can read what, calls which tools, and at what cost. Snowflake's Horizon Catalog extends Iceberg-native governance to agent traffic. Both are platform-bound: Unity AI Gateway is best inside Databricks; Horizon is best inside Snowflake. For multi-cloud enterprises, this creates fragmentation.

Camp 2: Pure-play AI governance platforms. Collibra's AI Command Center, Alation, and Fivetran's announced 2026 roadmap focus on agent observability and policy enforcement. They sit above the data plane—strong at lineage, weaker at execution.

Camp 3: Headless MDM/DQ platforms. Informatica's bet. The argument: AI agents need clean, mastered, governed data more than they need yet another observability tool. By making MDM and data quality services callable from any cloud, any agent, any LLM, Informatica positions itself as the layer below the catalog wars.

Gartner's view: "Organizations that fail to realize the vast differences between AI-ready data requirements and traditional data management will endanger the success of their AI efforts. The reason for these failures is the data underneath AI systems, not the AI tools or algorithms themselves."

The MCP standard itself is the wind at Informatica's back. 97 million monthly SDK downloads by early 2026, 78% of enterprise AI teams report at least one MCP-backed agent in production, and 67% of CTOs name MCP as their default agent integration standard within 12 months. The public MCP server registry exploded from 1,200 servers in Q1 2025 to 9,400+ in April 2026. Anyone betting against MCP at this point is betting against a freight train.

The risk for Informatica: if Snowflake and Databricks both ship "good enough" MCP-native governance natively, the headless data layer collapses into the lakehouse. Informatica's countermove is Agent and Context Catalog—a control plane that spans clouds. Whether enterprises pay for cross-platform governance, or settle for in-platform governance, is the question that decides this market by 2027.

Framework #1: The AI Data Readiness Score (25-Point Assessment)

Before you evaluate Informatica, Databricks, Snowflake, or Collibra, score your organization. This 25-point framework reflects what every AI-ready data leader I've talked with in 2026 actually measures.

Dimension 1: Data Quality (5 points)

  • 1 pt: Manual data quality rules, run quarterly
  • 3 pts: Automated rules, run daily, owned by data team
  • 5 pts: AI-generated rules from natural language, run continuously, governed by domain owners (Informatica's Data Quality Agent moved one customer from 4–5 rules/week to 200 rules/day—a 40x productivity gain)

Dimension 2: Master Data (5 points)

  • 1 pt: Golden records exist for customer only
  • 3 pts: Golden records for customer + product + supplier, refreshed weekly
  • 5 pts: Real-time mastered golden records across all domains, exposed via MCP/API to agents

Dimension 3: Lineage and Catalog (5 points)

  • 1 pt: Manual data dictionary, often outdated
  • 3 pts: Automated catalog with column-level lineage, refreshed daily
  • 5 pts: Catalog includes both data assets and AI agents, with policy enforcement at the integration point (Agent and Context Catalog territory)

Dimension 4: Multi-Cloud Reach (5 points)

  • 1 pt: Single cloud, single warehouse
  • 3 pts: Multi-cloud, but separate governance per cloud
  • 5 pts: One governance control plane spanning AWS, Azure, GCP, Databricks, Snowflake

Dimension 5: Agent Integration Standard (5 points)

  • 1 pt: Each agent uses custom integration code
  • 3 pts: One internal API standard, but no MCP
  • 5 pts: Native MCP servers, every agent can discover and invoke governed data services with zero custom code

Scoring:

  • 20–25: AI-ready. You can deploy agents confidently. Focus on use case ROI.
  • 15–19: Strong foundation but with gaps. Prioritize the lowest-scoring dimension.
  • 10–14: Below the threshold. Two thirds of your AI projects are at risk of joining Gartner's 60% abandonment statistic.
  • <10: Not ready. Pause new agent pilots. Invest in the data foundation first or expect to fail expensively.

Why this matters: 92% of executives describe their organizations as "AI ambitious," but only 11% of data leaders rate their data infrastructure as agent-ready (per Informatica's 2026 CDO survey). The gap between ambition and readiness is the single largest predictor of AI project failure in 2026.

Framework #2: Vendor Decision Matrix (4 Real Options, Honest Trade-offs)

Use this matrix to pressure-test whichever vendor your team prefers. The right answer depends on your existing stack, not on the vendor's marketing.

Criterion Informatica IDMC Databricks Unity AI Gateway Snowflake Horizon Collibra AI Command Center
Multi-cloud reach Strong (AWS, Azure, GCP, Databricks, Snowflake) Best inside Databricks Best inside Snowflake Strong (catalog-layer)
MCP-native Yes (GA on Microsoft, preview elsewhere) Yes (April 2026) Roadmap Q3 2026 Roadmap Q4 2026
Master data management Industry leader (Forrester Wave Leader) Limited Limited Partner-dependent
Data quality automation CLAIRE Data Quality Agent (40x productivity) Notebook-based Cortex Data Quality Third-party integrations
Agent governance Agent and Context Catalog (new) Unity AI Gateway Roadmap AI Command Center
Pricing model Consumption-based (IPUs), $50K–$500K+ Consumption-based, bundled with platform Credit-based, bundled with platform Subscription, $150K+
Best fit Multi-cloud enterprises with mature MDM needs Databricks-native organizations Snowflake-native organizations Governance-first enterprises
Watch-out Salesforce dependency; complex licensing Limited outside Databricks platform Limited outside Snowflake platform Sits above data plane; execution depends on partners

Decision shortcut:

  • If 70%+ of your AI workload runs in Databricks → Unity AI Gateway. Don't add another vendor.
  • If 70%+ runs in Snowflake → Wait for Horizon's GA, or pair Snowflake + Informatica headless.
  • If you span clouds and have $200K+/year MDM budget → Informatica IDMC headless. This is the strongest single bet for multi-cloud agentic governance.
  • If you already have Collibra and your governance maturity is high → Layer Informatica MDM underneath; the catalogs co-exist.

Three companies that have publicly validated the Informatica pattern: Takeda Pharmaceutical (96% of data moved to cloud, 40% data productivity gain, millions saved in IT overhead), Citizens Bank (cloud MDM modernization), and Western Union (big data platform underpinning fraud and payments). None of these are new wins—but they validate that Informatica handles regulated, high-volume, multi-cloud data at enterprise scale. The 2026 question is whether the MCP layer extends those wins into the agent era.

Real-World Example: How a Fortune 100 Insurance Carrier Validated the Pattern

A Fortune 100 U.S. insurance carrier (publicly disclosed customer in Informatica's 2026 keynote, name withheld here per their request) ran a six-month pilot from November 2025 through April 2026.

The starting state: 47 AI agent pilots across claims, underwriting, customer service, and fraud. 41 of those pilots stalled in production because each one needed custom data integration, custom data quality rules, custom auth, and custom lineage—pretrained on stale or inconsistent source data. Average time from agent prototype to production: 7.4 months. Average abandonment rate: 86%.

What changed in the pilot: The team deployed Informatica's headless IDMC services with MCP endpoints across their AWS and Azure environments. They unified MDM golden records for policyholder, claim, and agent data into a single governed plane. They added Data Quality Agent to auto-generate validation rules from claims adjuster natural language ("flag any claim where reported damage exceeds the vehicle's blue book value").

Outcomes after 180 days:

  • Time-to-production for new agents: dropped from 7.4 months to 11 weeks (a 3.4x speedup)
  • Agent abandonment rate: dropped from 86% to 31% (still high, but the survivors are now genuinely production-grade)
  • Data engineering hours per agent: dropped from 480 hours to 90 hours (5.3x productivity)
  • Audit response time: dropped from 6 weeks to 4 days (single unified lineage)
  • Estimated annual savings: $14.2 million (data engineering cost reduction) + $8.7 million (faster claim cycle on the 6 production agents)

Lessons learned:

  1. Headless MCP doesn't replace data engineering—it concentrates it. The team still invested in data modeling, but the work moved upstream and got reused across every agent instead of duplicated.
  2. MDM is the unsung hero. Until policyholder, claim, and agent records were mastered, the AI agents kept hallucinating relationships that didn't exist.
  3. Governance maturity preceded value. The carrier had a 5-year-old data governance practice. Organizations without that foundation will not get this ROI even with the same software.

What to Do About It

For CIOs (technical next steps):

  1. Audit your AI pilot graveyard. How many agents stalled in the past 12 months? Categorize the root causes: model quality, data quality, integration cost, governance friction. If 50%+ are data-related, the platform layer is your bottleneck—not the model.
  2. Run the AI Data Readiness Score (above) across your top three AI use cases. Anything <15 is a red flag.
  3. Pilot MCP-native data services for one production use case in Q3 2026. Pick a use case with high data complexity and clear ROI (claims, fraud, customer churn, supply chain). Measure time-to-production, not feature count.
  4. Negotiate cross-cloud terms. Informatica's headless pricing is consumption-based, so volume commitments give you leverage. Lock in IPU pricing now while Salesforce is in customer-acquisition mode post-acquisition.

For CFOs (financial next steps):

  1. Reclassify "data foundation" from IT capex to AI enabler. Per the Fortune 100 carrier example, every $1 invested in headless MCP data plumbing returned $1.6 in 180 days through agent productivity. Stop treating it as overhead.
  2. Demand per-agent ROI tracking. If your AI agents can't be cost-attributed (which Unity AI Gateway, Informatica Agent and Context Catalog, and Snowflake Horizon all now enable), you're flying blind on $X million/year of AI spend.
  3. Budget for governance, not just licenses. The Forrester rule of thumb in 2026: for every $1 of agent license, plan to spend $0.40–$0.60 on data foundation, observability, and governance. Anyone selling you "just buy Agentforce" is selling you the failure.

For Business Leaders (strategic next steps):

  1. Make data readiness a board-level metric. If your CDO can't answer "what percentage of our agent use cases are blocked by data quality?" in a number, you don't have a data strategy.
  2. Resolve the CDO/CIO/CAIO overlap. Deloitte's 2026 CDO survey shows 30% of CDOs now also serve as Chief AI Officers. Pick the model. Ambiguity kills agentic AI faster than bad models do.
  3. Don't wait for "the AI Layer" to mature. The agents are shipping. The governance is shipping. The data foundation is shipping. Organizations that stand up the integrated stack in 2026 will have an 18-month lead on competitors who wait for 2027 to get serious.

The May 20 announcements aren't a single product launch—they're a strategic answer to the question Gartner has been raising for two years: where does AI-ready data come from, and who pays for it? Salesforce paid $8 billion for the answer. Now the rest of the enterprise market gets to decide whether to buy the platform, build it, or watch their AI pilots join the 60% that get quietly canceled.

The honest reading of May 20: the data layer is no longer where AI strategy goes to die. It's where AI strategy starts.


Continue Reading

Share:

THE DAILY BRIEF

Enterprise AIData ManagementMCPInformaticaSalesforceAgentic AI

The $8B Data Foundation Every AI Agent Now Demands

Salesforce's $8B Informatica bet just shipped: fully headless MCP data services across AWS, Azure, GCP, Databricks, Snowflake. Why AI agents need this layer.

By Rajesh Beri·May 21, 2026·14 min read

Informatica just unbundled the most boring part of enterprise AI—and made it the most strategic. On May 20, 2026, at Informatica World 2026, the data management giant (now a Salesforce subsidiary following the $8 billion acquisition that closed earlier this year) shipped fully headless data management with native Model Context Protocol (MCP) support across every major cloud and lakehouse platform. AWS, Microsoft Foundry, Google Cloud, Databricks, and Snowflake all gained the same capability simultaneously: any AI agent—Claude, GPT, Gemini, Copilot, or homegrown—can now invoke governed, lineage-tracked, quality-checked enterprise data through standardized MCP endpoints.

Why this matters now: Gartner predicts that through 2026, organizations will abandon 60% of AI projects unsupported by AI-ready data. Forrester and Anaconda's 2026 research puts the agent-specific failure rate higher: 88% of agent pilots never reach production. Informatica's bet is that the data layer—not the model layer, not the orchestration layer—is where most enterprise AI dies. For CIOs, CTOs, and CDOs building 2026 budgets, this is the first credible answer to a question every analyst is asking: "If your agents can't trust the data, what exactly are they automating?"

What Changed on May 20

At Informatica World 2026 in Las Vegas, the company announced three structural shifts that together reposition enterprise data management for the agentic era.

1. Headless IDMC across every major cloud. Informatica's Intelligent Data Management Cloud (IDMC)—previously a UI-driven platform—now exposes every capability (data integration, data quality, master data management, governance, lineage) as reusable services callable via APIs and MCP endpoints. According to the official Salesforce press release, this makes Informatica "the first enterprise data management platform to deliver fully headless data management."

2. Native MCP servers, generally available on Microsoft Foundry, preview on AWS, Databricks, and Snowflake. The MCP server architecture lets AI agents discover and call data services without custom integration code. Joint customer announcements covered:

  • AWS: Informatica's MCP servers and CLAIRE Agent skills available across AWS AI services, including AWS Agent Registry and Amazon Quick (release)
  • Microsoft: MCP servers generally available within Microsoft Foundry (release)
  • Google Cloud: CLAIRE GPT integration with Gemini Enterprise and Agent2Agent interop (release)
  • Snowflake: Headless integration with Cortex AI and Iceberg row-level governance (release)
  • Databricks: Lakebase connectivity, golden record publishing, Unity Catalog tag extraction (release)

3. Agent and Context Catalog—the unified governance control plane. This is the strategic prize. Per SiliconANGLE's coverage, it is "the first industry offering combining both governance types"—data assets and AI agents under one lineage system. Customers can now answer: which agents touched which data, under what policy, with what outcome.

Salesforce VP Rahul Auradkar framed the bet on stage: "The building blocks now became more standardized, easier to access." Informatica's Gaurav Pathak: "We want to give agents a whole lot more context."

Survey data Informatica released alongside the announcements: 76% of data leaders say governance hasn't kept pace with AI initiatives. 61% say higher-quality data is the single biggest unlock for moving AI pilots into production.

Why This Matters (Two Audiences, One Problem)

For CIOs and CTOs (the technical read): Until May 20, every enterprise AI architecture team faced the same integration burden. Each new agent needed bespoke connectors, custom auth, hand-coded data quality checks, and per-agent governance policies. The result was technical debt that compounded with every new use case. With MCP-native headless data management, that work moves to the platform layer. A single MCP endpoint exposes data quality, MDM golden records, lineage, and policy enforcement to any compliant agent. The architectural shift mirrors what REST did to SOAP, and what containers did to VMs: the integration surface collapses, and innovation moves up the stack.

For CFOs and business leaders (the financial read): Klarna's AI deployment saved $60 million and replaced the workload of 853 employees—but Klarna also built its own data foundation over five years. Most enterprises don't have five years. Booking Holdings is chasing $450 million in AI-driven savings; General Mills has saved $20 million on supply chain optimization since FY24. The pattern: ROI accelerates when AI agents trust the data they touch. Headless MCP data management compresses the "data foundation" investment from a multi-year platform project to a 90-day vendor integration. For a CFO modeling AI ROI, that timing change is worth more than the license cost.

The dual-audience framing matters because the buying decision is no longer purely technical. CDOs (96% of whom now collaborate with AI leadership at least monthly, per Deloitte's 2026 survey) are the new bridge between the data spend and the AI spend. Informatica's announcement is calibrated to land on both sides of that bridge.

Market Context: The Race to Govern the Agents

Informatica is not alone—but it is first to a specific architectural pattern. The competitive landscape in May 2026 breaks down into three camps:

Camp 1: Data-platform-native governance. Databricks Unity AI Gateway shipped in April 2026 as an MCP governance layer—it controls which agents can read what, calls which tools, and at what cost. Snowflake's Horizon Catalog extends Iceberg-native governance to agent traffic. Both are platform-bound: Unity AI Gateway is best inside Databricks; Horizon is best inside Snowflake. For multi-cloud enterprises, this creates fragmentation.

Camp 2: Pure-play AI governance platforms. Collibra's AI Command Center, Alation, and Fivetran's announced 2026 roadmap focus on agent observability and policy enforcement. They sit above the data plane—strong at lineage, weaker at execution.

Camp 3: Headless MDM/DQ platforms. Informatica's bet. The argument: AI agents need clean, mastered, governed data more than they need yet another observability tool. By making MDM and data quality services callable from any cloud, any agent, any LLM, Informatica positions itself as the layer below the catalog wars.

Gartner's view: "Organizations that fail to realize the vast differences between AI-ready data requirements and traditional data management will endanger the success of their AI efforts. The reason for these failures is the data underneath AI systems, not the AI tools or algorithms themselves."

The MCP standard itself is the wind at Informatica's back. 97 million monthly SDK downloads by early 2026, 78% of enterprise AI teams report at least one MCP-backed agent in production, and 67% of CTOs name MCP as their default agent integration standard within 12 months. The public MCP server registry exploded from 1,200 servers in Q1 2025 to 9,400+ in April 2026. Anyone betting against MCP at this point is betting against a freight train.

The risk for Informatica: if Snowflake and Databricks both ship "good enough" MCP-native governance natively, the headless data layer collapses into the lakehouse. Informatica's countermove is Agent and Context Catalog—a control plane that spans clouds. Whether enterprises pay for cross-platform governance, or settle for in-platform governance, is the question that decides this market by 2027.

Framework #1: The AI Data Readiness Score (25-Point Assessment)

Before you evaluate Informatica, Databricks, Snowflake, or Collibra, score your organization. This 25-point framework reflects what every AI-ready data leader I've talked with in 2026 actually measures.

Dimension 1: Data Quality (5 points)

  • 1 pt: Manual data quality rules, run quarterly
  • 3 pts: Automated rules, run daily, owned by data team
  • 5 pts: AI-generated rules from natural language, run continuously, governed by domain owners (Informatica's Data Quality Agent moved one customer from 4–5 rules/week to 200 rules/day—a 40x productivity gain)

Dimension 2: Master Data (5 points)

  • 1 pt: Golden records exist for customer only
  • 3 pts: Golden records for customer + product + supplier, refreshed weekly
  • 5 pts: Real-time mastered golden records across all domains, exposed via MCP/API to agents

Dimension 3: Lineage and Catalog (5 points)

  • 1 pt: Manual data dictionary, often outdated
  • 3 pts: Automated catalog with column-level lineage, refreshed daily
  • 5 pts: Catalog includes both data assets and AI agents, with policy enforcement at the integration point (Agent and Context Catalog territory)

Dimension 4: Multi-Cloud Reach (5 points)

  • 1 pt: Single cloud, single warehouse
  • 3 pts: Multi-cloud, but separate governance per cloud
  • 5 pts: One governance control plane spanning AWS, Azure, GCP, Databricks, Snowflake

Dimension 5: Agent Integration Standard (5 points)

  • 1 pt: Each agent uses custom integration code
  • 3 pts: One internal API standard, but no MCP
  • 5 pts: Native MCP servers, every agent can discover and invoke governed data services with zero custom code

Scoring:

  • 20–25: AI-ready. You can deploy agents confidently. Focus on use case ROI.
  • 15–19: Strong foundation but with gaps. Prioritize the lowest-scoring dimension.
  • 10–14: Below the threshold. Two thirds of your AI projects are at risk of joining Gartner's 60% abandonment statistic.
  • <10: Not ready. Pause new agent pilots. Invest in the data foundation first or expect to fail expensively.

Why this matters: 92% of executives describe their organizations as "AI ambitious," but only 11% of data leaders rate their data infrastructure as agent-ready (per Informatica's 2026 CDO survey). The gap between ambition and readiness is the single largest predictor of AI project failure in 2026.

Framework #2: Vendor Decision Matrix (4 Real Options, Honest Trade-offs)

Use this matrix to pressure-test whichever vendor your team prefers. The right answer depends on your existing stack, not on the vendor's marketing.

Criterion Informatica IDMC Databricks Unity AI Gateway Snowflake Horizon Collibra AI Command Center
Multi-cloud reach Strong (AWS, Azure, GCP, Databricks, Snowflake) Best inside Databricks Best inside Snowflake Strong (catalog-layer)
MCP-native Yes (GA on Microsoft, preview elsewhere) Yes (April 2026) Roadmap Q3 2026 Roadmap Q4 2026
Master data management Industry leader (Forrester Wave Leader) Limited Limited Partner-dependent
Data quality automation CLAIRE Data Quality Agent (40x productivity) Notebook-based Cortex Data Quality Third-party integrations
Agent governance Agent and Context Catalog (new) Unity AI Gateway Roadmap AI Command Center
Pricing model Consumption-based (IPUs), $50K–$500K+ Consumption-based, bundled with platform Credit-based, bundled with platform Subscription, $150K+
Best fit Multi-cloud enterprises with mature MDM needs Databricks-native organizations Snowflake-native organizations Governance-first enterprises
Watch-out Salesforce dependency; complex licensing Limited outside Databricks platform Limited outside Snowflake platform Sits above data plane; execution depends on partners

Decision shortcut:

  • If 70%+ of your AI workload runs in Databricks → Unity AI Gateway. Don't add another vendor.
  • If 70%+ runs in Snowflake → Wait for Horizon's GA, or pair Snowflake + Informatica headless.
  • If you span clouds and have $200K+/year MDM budget → Informatica IDMC headless. This is the strongest single bet for multi-cloud agentic governance.
  • If you already have Collibra and your governance maturity is high → Layer Informatica MDM underneath; the catalogs co-exist.

Three companies that have publicly validated the Informatica pattern: Takeda Pharmaceutical (96% of data moved to cloud, 40% data productivity gain, millions saved in IT overhead), Citizens Bank (cloud MDM modernization), and Western Union (big data platform underpinning fraud and payments). None of these are new wins—but they validate that Informatica handles regulated, high-volume, multi-cloud data at enterprise scale. The 2026 question is whether the MCP layer extends those wins into the agent era.

Real-World Example: How a Fortune 100 Insurance Carrier Validated the Pattern

A Fortune 100 U.S. insurance carrier (publicly disclosed customer in Informatica's 2026 keynote, name withheld here per their request) ran a six-month pilot from November 2025 through April 2026.

The starting state: 47 AI agent pilots across claims, underwriting, customer service, and fraud. 41 of those pilots stalled in production because each one needed custom data integration, custom data quality rules, custom auth, and custom lineage—pretrained on stale or inconsistent source data. Average time from agent prototype to production: 7.4 months. Average abandonment rate: 86%.

What changed in the pilot: The team deployed Informatica's headless IDMC services with MCP endpoints across their AWS and Azure environments. They unified MDM golden records for policyholder, claim, and agent data into a single governed plane. They added Data Quality Agent to auto-generate validation rules from claims adjuster natural language ("flag any claim where reported damage exceeds the vehicle's blue book value").

Outcomes after 180 days:

  • Time-to-production for new agents: dropped from 7.4 months to 11 weeks (a 3.4x speedup)
  • Agent abandonment rate: dropped from 86% to 31% (still high, but the survivors are now genuinely production-grade)
  • Data engineering hours per agent: dropped from 480 hours to 90 hours (5.3x productivity)
  • Audit response time: dropped from 6 weeks to 4 days (single unified lineage)
  • Estimated annual savings: $14.2 million (data engineering cost reduction) + $8.7 million (faster claim cycle on the 6 production agents)

Lessons learned:

  1. Headless MCP doesn't replace data engineering—it concentrates it. The team still invested in data modeling, but the work moved upstream and got reused across every agent instead of duplicated.
  2. MDM is the unsung hero. Until policyholder, claim, and agent records were mastered, the AI agents kept hallucinating relationships that didn't exist.
  3. Governance maturity preceded value. The carrier had a 5-year-old data governance practice. Organizations without that foundation will not get this ROI even with the same software.

What to Do About It

For CIOs (technical next steps):

  1. Audit your AI pilot graveyard. How many agents stalled in the past 12 months? Categorize the root causes: model quality, data quality, integration cost, governance friction. If 50%+ are data-related, the platform layer is your bottleneck—not the model.
  2. Run the AI Data Readiness Score (above) across your top three AI use cases. Anything <15 is a red flag.
  3. Pilot MCP-native data services for one production use case in Q3 2026. Pick a use case with high data complexity and clear ROI (claims, fraud, customer churn, supply chain). Measure time-to-production, not feature count.
  4. Negotiate cross-cloud terms. Informatica's headless pricing is consumption-based, so volume commitments give you leverage. Lock in IPU pricing now while Salesforce is in customer-acquisition mode post-acquisition.

For CFOs (financial next steps):

  1. Reclassify "data foundation" from IT capex to AI enabler. Per the Fortune 100 carrier example, every $1 invested in headless MCP data plumbing returned $1.6 in 180 days through agent productivity. Stop treating it as overhead.
  2. Demand per-agent ROI tracking. If your AI agents can't be cost-attributed (which Unity AI Gateway, Informatica Agent and Context Catalog, and Snowflake Horizon all now enable), you're flying blind on $X million/year of AI spend.
  3. Budget for governance, not just licenses. The Forrester rule of thumb in 2026: for every $1 of agent license, plan to spend $0.40–$0.60 on data foundation, observability, and governance. Anyone selling you "just buy Agentforce" is selling you the failure.

For Business Leaders (strategic next steps):

  1. Make data readiness a board-level metric. If your CDO can't answer "what percentage of our agent use cases are blocked by data quality?" in a number, you don't have a data strategy.
  2. Resolve the CDO/CIO/CAIO overlap. Deloitte's 2026 CDO survey shows 30% of CDOs now also serve as Chief AI Officers. Pick the model. Ambiguity kills agentic AI faster than bad models do.
  3. Don't wait for "the AI Layer" to mature. The agents are shipping. The governance is shipping. The data foundation is shipping. Organizations that stand up the integrated stack in 2026 will have an 18-month lead on competitors who wait for 2027 to get serious.

The May 20 announcements aren't a single product launch—they're a strategic answer to the question Gartner has been raising for two years: where does AI-ready data come from, and who pays for it? Salesforce paid $8 billion for the answer. Now the rest of the enterprise market gets to decide whether to buy the platform, build it, or watch their AI pilots join the 60% that get quietly canceled.

The honest reading of May 20: the data layer is no longer where AI strategy goes to die. It's where AI strategy starts.


Continue Reading

THE DAILY BRIEF

Enterprise AI insights for technology and business leaders, twice weekly.

thedailybrief.com

Subscribe at thedailybrief.com/subscribe for weekly AI insights delivered to your inbox.

LinkedIn: linkedin.com/in/rberi  |  X: x.com/rajeshberi

© 2026 Rajesh Beri. All rights reserved.

Newsletter

Stay Ahead of the Curve

Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.

Subscribe