Good morning, AI enthusiasts. Cloudera just released its State of Enterprise AI in 2025 survey, capturing the views of 1,500+ IT executives and revealing a critical paradox: AI is everywhere, but it isn’t yet fully unlocked.

Leaders see clear value in AI, yet they’re dealing with infrastructure gaps like expensive compute, broken data, and governance issues that decide whether AI scales or fails.

To understand these problems (and their solutions) better, we partnered and sat down with Cloudera’s CTO, Sergio Gago, for an exclusive Q&A.

In today’s AI rundown:

• Why only 21% enterprises have full AI integration

• AI playbook for organizations starting from scratch

• Securing early business wins with AI

• Measuring the wins for building on growth

• Taking AI to data for full security

• Baking in compliance by design

• Making ‘AI everywhere’ a reality

LATEST DEVELOPMENTS

AI INTEGRATION

🤖 Why only 21% enterprises have full AI integration

Image: Kiki Wu / The Rundown

The Rundown: While enterprises are bullish on AI and continue to cite heavy investments and confidence in the tech, only 21% of the leaders in Cloudera’s survey said they have fully integrated AI into their core business processes.

Cheung: Why is full AI integration so hard even today? What are the biggest factors holding companies back?

Gago: One of the biggest shifts our survey uncovered was the cost of training models. Compared to our survey one year ago, we found that the cost to access computer capacity for training AI is on the rise, jumping from 8% in 2024 to 42% now.

Just as important is access to the right data. To train AI models effectively, organizations need access to one hundred percent of their data in all forms and wherever it resides. Without full access, models are limited in scope and accuracy. This also applies to RAG (Retrieval Augmented Generation) techniques, giving the LLMs contextual access to your enterprise information.

When AI can be applied to all this data—whether in the cloud, in the data center, or at the edge—it becomes more trustworthy, more contextual, and ultimately more valuable to the business.

Why it matters: For AI practitioners and decision-makers, this finding from Cloudera highlights that it’s not the running or scaling that’s hindering AI integration — but the core foundation that lies underneath. The only path to trustworthy, enterprise-wide AI goes through solving infrastructure efficiency and unlocking all organizational data.

PLAYBOOK

🔥 AI playbook for organizations starting from scratch

Image: Kiki Wu / The Rundown

The Rundown: To reach 100% AI integration, organizations need to follow a structured path: first anchoring efforts with clear business goals, then breaking down data and infrastructure barriers, and finally scaling through focused, value-driven use cases.

Cheung: What measures should organizations with zero to little AI take to go up to the full 100% AI integration mark?

Gago: First, clarify your goals: define which business problems you're trying to solve and who owns those decisions. Next, ensure your data is clean, contextual, and accessible. That means unifying structured, semi-structured, and unstructured data across environments: clouds, in the data center, or edge.

From there, build a flexible infrastructure that can evolve as AI models and frameworks change. Prioritize security, governance, and transparency from the start, because trust is foundational.

Finally, use reference architectures or accelerators to move quickly with targeted, high-impact use cases. The organizations that succeed are the ones who move with focus and responsibility and scale from there.

Why it matters: Gago’s roadmap makes AI integration a step-by-step journey. By following this approach, organizations can turn AI from scattered experiments with fragmented data efforts into a trusted capability that delivers measurable impact across the business.

EARLY WINS

🦄 Securing early business wins with AI

Image: Kiki Wu / The Rundown

The Rundown: While processes across industries are being reshaped by AI, organizations should start off with select, tightly-scoped use cases that can deliver measurable results.

Cheung: What business processes are being reshaped by AI, and what’s an easy, high-confidence AI win you recommend shipping first?

Gago: The use cases span industries from manufacturing to banking. Whether an organization is trying to get ahead of maintenance on the factory floor, wants to revamp its customer experience, or leverages AI agents to help identify fraud and security risks, AI has become a ubiquitous asset in every IT leader's toolbelt.

Early use cases come from well-defined, ROI-driven domains. In the case of formats like AI agents, this would mean areas like IT helpdesk agents and DevOps assistants. Prioritizing adopting AI in these domains gives IT leaders a great opportunity to introduce automation, while driving tangible results.

Gago added: Helpdesk agents can be deployed to automate micro-tasks such as password resets, respond to tier-one support tickets, and recommend knowledge base content. DevOps assistants can detect anomalies, automate remediation, improve cost control, or generate alerts for infrastructure management.

Why it matters: AI can feel overwhelming when applied everywhere at once, but focus wins first. By starting with ROI-driven domains like IT helpdesk, enterprises can deliver quick, measurable results — providing early wins that build confidence, prove value to stakeholders, and create momentum for scaling AI responsibly across functions.

MEASURING OUTCOMES

🧠 Measuring AI wins for building on growth

Image: Kiki Wu / The Rundown

The Rundown: In Cloudera’s survey, operational efficiency was cited as the biggest ROI from their AI projects, but measuring impact shouldn’t stop at cost and speed. It should also account for customer/user satisfaction.

Cheung: How do you measure if an AI project is actually helping?

Gago: Our survey asked respondents to share where they expect the biggest ROI from AI over the year. 29% pointed to operational efficiency, followed by 18% citing customer experience, 15% product innovation, 14% revenue generation, 13% risk management, and 11% talent productivity.

Gago added: To measure if it’s helping, organizations should look at metrics tied to speed, cost, and satisfaction. That might include ticket resolution time, reduction in manual workload, incident frequency, or internal user feedback. AI’s impact becomes evident when it consistently shortens cycles, reduces costs, and improves outcomes.

Why it matters: Measuring AI’s impact through efficiency and satisfaction makes its value tangible. When organizations track outcomes and show real benefits, they build a strong case for AI’s use — and drive executive confidence to fuel broader adoption across the enterprise.

SECURITY

⚡️ Taking AI to data for full security

Image: Kiki Wu / The Rundown

The Rundown: As AI adoption accelerates, so do security risks. But proactive governance, enforcing lineage, and bringing AI to the data (rather than moving data to AI) can help harness AI’s power, without compromising trust or security.

Cheung: AI ties to several security concerns, with 50% of survey respondents worrying about training data leaks and 48% about unauthorized access. How does Cloudera bridge this gap for secure AI?

Gago: Governance is critical. Without consistent governance and security standards in place, anytime an organization opens its data up to train AI models there is a risk that it becomes susceptible to leakage or a third-party actor. The industry was generally very good at that with classical machine learning, but somehow many companies forgot about data governance in the world of Generative AI.

Aside from Cloudera's governance tooling, our main advantage is bringing AI to your data. By partnering with Cloudera, you can maintain data ownership, keep it wherever it resides, and apply AI on top of it — capturing all insights without opening yourself and your business to increased risk. This is: data access, fine-grained controls, catalog, and lineage as the building blocks for safe and private AI deployments.

Cloudera also delivers data lineage to ensure data quality and help teams understand how AI is applying it to make decisions. This eliminates the black box conundrum, giving users visibility into the data AI is using to respond or take action.

Why it matters: Strong governance and keeping AI close to the data not only reduce the risk of leaks and unauthorized access but also improve visibility into how decisions are made. The benefit is clear: organizations can unlock more value from AI while protecting sensitive data and building trust with customers and regulators.

COMPLIANCE

🧪 Baking in compliance by design

IImage: Kiki Wu / The Rundown

The Rundown: Teams can often get stuck on writing and implementing policies for enforcing security and governance compliance. But, as Gago points out, it should be baked in right from the beginning — not as an afterthought.

Cheung: What’s a practical, non‑scary way to put basic security rules in place across a setup — and actually enforce them?

Gago: Start by embedding rules directly into your data architecture, not layered on top of it. That means things like encryption, access controls, lineage, and audit trails should be baked in from the beginning, not retrofitted after the fact.

Write policies once, then apply them universally across public cloud, private cloud, and in the data center, wherever the data lives. Enforcing policy should feel automatic, not manual. The best systems don’t rely on someone remembering to check a box; they enforce compliance by design.

Gago added: Focus on a few high-impact rules: who can see what, where sensitive data lives, and how it’s tracked. Start small, then scale up. Most importantly, make policy transparent and explainable — involve your legal, IT, cybersecurity, and compliance teams from the beginning. Governance can’t be an afterthought.

Why it matters: When policies are enforced by design and made explainable, teams understand the rules, the reasoning behind them, and their implementation — and accept the guardrails for ensuring security across business processes using AI.

SCALING AI

💡 Making ‘AI everywhere’ a reality

Image: Kiki Wu / The Rundown

The Rundown: AI may soon be deeply embedded across most enterprises, but the journey won’t be simple. Teams must overcome barriers around integration, management, and security, and above all, focus on building trust in their systems.

Cheung: If we zoom out 5 years, do you believe ‘AI everywhere’ will be a reality? And what’s your personal north star for Cloudera to that end?

Gago: ‘AI everywhere’ is possible even today, but only if organizations build with governance and flexibility at the core and create access to data anywhere. The biggest barriers (data silos, cost, and compliance) can be solved with an open and policy-driven architecture. But the real challenge won’t just be infrastructure. It’ll be trust.

Gago added: The future belongs to teams that can scale AI responsibly, with visibility into how decisions are made and confidence in the data behind them To that end, Cloudera’s north star is clear: bringing AI to data, anywhere. That means enabling large enterprises to securely apply and scale AI to 100% of their data. Cloudera aims to be the platform enterprises trust most to innovate confidently, govern effectively, and drive lasting value.

Why it matters: Trust is emerging as the true currency of enterprise AI. Gago makes it clear that scaling AI responsibly is all about ensuring decisions are explainable, governed, and grounded in reliable data. Enterprises that embed trust at the core of their AI efforts will be the winners.