1. Introduction
In this episode of “Impact of AI: Explored”, we (James O’Regan and Gerjon Kunst) sit down with Matt Soltau, Global Director of Strategy & Operations at IntelliPaaS, to unpack a topic that’s often invisible but rapidly becoming critical: digital waste and the sustainability impact of AI.
We’re collectively pouring billions into AI, yet very few organizations have a clear picture of how much energy their AI and digital operations actually consume. Behind every “good morning” prompt to your favorite large language model sits an infrastructure of GPU‑powered data centers, zombie jobs running in the background, and duplicate data spinning on disks for years.
We explore how enterprises can make AI part of the sustainability solution rather than just another part of the problem—from integration debt and data silos to ESG reporting, AI agents, and what it really means to build AI‑ready, sustainable data foundations.
2. Meet the Guest: Matt Soltau
Matt Soltau is the Global Director of Strategy & Operations at IntelliPaaS, an AI‑powered enterprise Integration Platform as a Service (eiPaaS) that connects cloud, on‑premises, and legacy systems without forcing organizations to rip and replace their existing stack.
With a background that spans integration, automation, and global go‑to‑market leadership, Matt has helped enterprises across APAC, EMEA, and North America untangle complex, legacy tech stacks so they can deploy secure, compliant, and scalable AI and automation. He’s an MBA graduate based out of Hong Kong, and his current work at IntelliPaaS focuses on building the data backbone organizations need to operationalize AI in regulated and hybrid environments.
At IntelliPaaS, Matt and his team provide:
- Global deployment flexibility – deploy anywhere: on‑prem, private cloud, public cloud, containers, and even air‑gapped high‑security environments.
- AI‑enhanced data transformation and workflow automation – enabling real‑time, low‑code integration across modern and legacy systems.
- Strong fit for regulated industries where data sovereignty, compliance, and security are non‑negotiable.
In short, Matt lives at the intersection of data integration, AI readiness, and sustainable digital operations—which made him the perfect guest for this episode.
Meanwhile, AI‑optimized racks in data centers can draw 3–4x the power of standard server racks, and we’re rapidly building more of them. Estimates suggest that by 2035, AI data centers in the US alone could require over 30x the power they did in 2024, underscoring how quickly this footprint is growing.
3. Setting the Stage
Most organizations see AI primarily as a productivity and innovation engine, but very few see it as a massive energy consumer hiding inside the cloud bill. The waste is invisible: nobody sees it, nobody owns it, and almost nobody meters it.
At the same time, AI is also being used to optimize delivery routes, reduce CO₂ emissions in logistics, and streamline operations—meaning it can both consume more energy and help save it elsewhere. That tension is exactly why this conversation matters.
In this blogpost, based on our discussion with Matt, you can expect to:
- Understand what digital waste looks like in real enterprises.
- See why data, integration, and processes are the unsexy but essential foundation for AI.
- Explore how AI can both help and hurt sustainability efforts, depending on how it’s architected.
- Walk away with practical ideas to reduce digital waste while still moving fast with AI.
4. Episode Highlights
Highlight 1 – “Zombie jobs” and invisible waste
Matt describes how many enterprises have “zombie jobs”—background processes that have been running for years or even decades, consuming compute and delivering no value. There’s no meter, no clear owner, and they simply disappear into a large cloud bill that nobody wants to dissect.
He compares it to leaving all your appliances on standby at home and then being surprised when the energy bill spikes—except, in enterprises, the problem is far harder to trace.
Highlight 2 – AI power demand vs. real efficiency
When we talked about the power draw of AI data centers, Matt contrasted traditional server racks (roughly 5–15 kW) with AI‑optimized GPU racks that can hit 40–60 kW, or even around 100 kW per rack at the cutting edge. Then he added the image everyone can relate to:
“Now imagine a hallway full of them… thousands of them.”
His core question: if power demand might grow 30x, why aren’t we aiming to be 30% (or more) productive with the power we already have?
5. Deep Dive: Grounding AI in Clean, Connected Data
If there’s one recurring theme in this conversation, it’s that AI is only as good—and as safe—as the systems and data underneath it.
From “spaghetti data” to a nervous system
Enterprises often live with data silos: finance keeps financial data, sales and marketing own their customer datasets, and legacy systems quietly run business‑critical workflows that nobody fully understands anymore. The result is what Matt elsewhere calls “spaghetti data”—fragmented, duplicated, and inconsistent.
Matt uses a human nervous system analogy:
- Your body constantly processes signals—like feeling a cold glass of water in your hand—in real time.
- That live, interconnected flow is what lets you react quickly and correctly.
For organizations, the equivalent is integration:
- Real‑time data pipelines between departments and systems.
- Clear sources of truth that both humans and AI can trust.
- The ability to transform and enrich data at the moment it’s needed, not as an afterthought.
Why integration comes before AI
Matt’s view—and one we very much share—is that AI readiness starts with digital transformation and integration, not with a model.
He shares a story about an insurance company in Singapore that tried to automate its customer service with AI:
- Live calls were transcribed in real time.
- The AI would surface the “right” answer to the operator on screen.
- It worked on day zero, but there was no continuous retraining as policies changed, objections emerged, and new edge cases appeared.
- It didn’t take long before the AI started giving incorrect advice, and the project had to be shelved.
The lesson:
- If your data pipelines, integration layer, and processes are not designed for continuous updates, your AI will drift out of reality.
- Sandbox success doesn’t equal production success—the “Ferrari engine in a golf cart” analogy Matt used captures how misaligned architectures can be.
In his words, “start with the plumbing layer first.” Once your integration backbone is solid, AI can safely sit on top, pulling from clean, current sources of truth rather than trying to improvise its way through six different legacy systems.
6. Real-Life Stories & Examples
The invisible data center behind “good morning”
Every time you type “good morning” into ChatGPT or another LLM, you’re kicking off a computation that happens in a power‑hungry AI data center, potentially cooled by substantial water usage and backed by large‑scale energy generation. It’s invisible, but it’s not free.
Route optimization vs. emissions trade‑offs
Gerjon shared a real customer scenario: using AI to optimize delivery routes.
- On the one hand, smarter route planning reduces CO₂ emissions on the road.
- On the other, the AI models and infrastructure required for that optimization consume more energy in the data center.
The net impact isn’t always obvious, which is why measurement and transparency are so important.
The bank server nobody dared to touch
We also swapped war stories from IT: James recalled working at an Irish bank where a mysterious server in the corner was effectively the core banking system—no one knew exactly how it worked, and everyone was afraid to touch it.
“What’s that do?”
“Don’t touch it. It’s pretty much running the entire banking system.”
Matt’s take:
“The most dangerous system in your business is the one that works perfectly and nobody knows why.”
Now add AI on top of that:
- That “dusty attic” of systems might contain incredibly valuable data and business logic.
- AI could help analyze, modernize, and consolidate what’s there—ideally without deleting anything critical in the process.
- Done right, you could shut down old, power‑hungry servers and move the value (the data) into modern, efficient platforms.
AI vs. Google for real-life questions
Matt even used a simple French toast anecdote: with a few mismatched ingredients and no flour—only cornstarch—he turned to Perplexity instead of Google.
- A traditional approach would mean multiple search queries, recipe browsing, and manually figuring out substitutions.
- With an AI assistant, he got one tailored recipe and a great breakfast.
We don’t know the exact relative power consumption of those paths—but it illustrates how AI can consolidate steps, potentially offsetting some of the overhead when used well.
7. Key Takeaways
- Digital waste is real—and mostly invisible.
Enterprises run zombie jobs, keep duplicate data, and accumulate integration debt that quietly consumes compute without delivering value. - AI is both a power hog and a powerful optimizer.
GPU racks draw significantly more power than traditional servers, yet AI can also deduplicate data, optimize routes, and streamline operations to reduce waste elsewhere. - Clean, connected data beats fancy models.
Without integration, data governance, and clear sources of truth, AI becomes unreliable, hallucinates more, and often fails when moved from sandbox to production. - Integration is the foundation of AI readiness.
Platforms like IntelliPaaS let organizations deploy integrations anywhere—cloud, on‑prem, air‑gapped—while creating a real‑time data backbone that AI can safely build on. - ESG and IT are still disconnected.
ESG reporting is becoming mandatory, but IT teams often can’t measure the energy cost of their workflows, making digital operations a genuine blind spot in sustainability strategies. - Accountability is non‑negotiable.
True sustainability starts with data governance and ownership—for example, assigning a CFO or CIO clear responsibility for digital sustainability metrics. - Aim for efficiency, not just more power.
Instead of accepting a 30x increase in AI power demand, we should be asking how to be 30% more productive with the power we already use—through better architecture, integration, and process design.
8. Closing Thoughts
Talking with Matt reinforced something we keep coming back to in “Impact of AI: Explored”: the boring fundamentals—data, integration, governance, process documentation—are exactly what make AI powerful, safe, and sustainable in the real world.
AI isn’t going anywhere. It will continue to expand into more industries, workflows, and everyday tasks. The question is whether we treat it as an unmetered energy sink or as a disciplined, well‑architected layer on top of clean data and strong governance.
If you’re an IT leader, architect, or business stakeholder, now is the time to:
- Map your data flows and integrations.
- Identify zombie jobs and legacy systems you’re afraid to touch.
- Start giving someone clear accountability for digital sustainability.
We’d love to hear how you’re approaching AI, integration, and sustainability in your own organization. What’s your biggest challenge: data quality, ESG reporting, legacy systems, or governance around AI agents?
You can watch this episode on YouTube or listen on your favorite podcast platform, and as always—like, subscribe, and join the conversation so we can keep exploring the real impact of AI together.
