David Ly is the founder of Iveda, having served as CEO and Chairman of the Board of Directors since the company’s inception in 2003.
There’s a version of AI’s future that looks like a giant brain in the cloud—every device feeding data upward and every decision flowing back down. It’s a nice picture (I bet you’re picturing it now!), but it’s misaligned with what real-world deployments demand.
Over the past several years, I’ve watched the assumption of centralized AI run into the same wall over and over again. The wall isn’t theoretical; it’s a delayed alert, a failed connection at the worst possible moment or a privacy regulation nobody anticipated until the contract was already signed. These are far from one-off cases; they’re often the norm.
The shift toward edge AI—processing intelligence closer to where data is generated—is a genuine course correction that I believe will help us overcome this wall. Understanding why requires a simpler question: What does our environment demand?
When Milliseconds And Uptime Aren’t Optional
Speed matters differently depending on what’s at stake. In a consumer app, a two-second lag is a nuisance. In a mission-critical environment (e.g., a facility monitoring for safety or a system managing physical infrastructure in real time), that lag can mean the difference between an alert that arrives in time and one that doesn’t.
Centralized models carry inherent latency. Data travels to the cloud, that data gets processed and a decision comes back. And real-world conditions are rarely ideal. Network congestion, bandwidth constraints, intermittent connectivity—any of these can introduce delays at exactly the wrong moment.
I saw this play out in a healthcare deployment spanning 12 hospitals, each streaming video to a centralized command center for AI analysis. When the system started missing detections, the initial assumption was that the AI needed tuning. But the bandwidth between the hospitals and the central hub was inconsistent, so the AI was working with incomplete data. Once they fixed the network, the detection issues disappeared. The lesson? AI can only be as reliable as the infrastructure delivering its inputs.
Resilience must be built into the design from the start. An edge-based system can not only perform better under pressure—I’ve found it can keep performing even after a connection drops. And in the environments where AI matters most, you can’t afford to find out how resilient you are.
The Privacy Problem Is Structural, Not Cosmetic
Data privacy conversations in AI tend to focus on policy—compliance checklists, consent frameworks and regulatory timelines. Those things matter, but the more immediate problem is that the simple act of moving data creates exposure.
Every time raw data travels from a camera, sensor or device to a centralized server, it passes through infrastructure that can be intercepted or compromised. The more sensitive the data, the more that transmission pathway becomes a liability.
Processing data locally can change that equation. If the intelligence lives at the edge, the raw data doesn’t have to leave. What gets transmitted is a result, an alert or an anonymized signal—not the underlying footage or sensor readings that created it.
At my company, we’ve seen this with partners in Spain and Germany. Customers refused to transmit video data to the cloud at all, simply because they didn’t want the data leaving their facilities, full stop. By deploying AI at the edge and keeping everything on-premises, they were able to reduce regulatory exposure, simplify compliance and give customers full ownership of their data.
Public trust in AI-powered systems is fragile right now. When people understand that their data isn’t being shipped somewhere and stored indefinitely—and that the system sees only what it needs to see—the conversation changes. Trust is easier to build when the architecture supports it.
The Leadership Trade-Off Nobody Truthfully Talks About
I’ll be the first to say that cloud infrastructure offers real advantages. In my experience, it can enhance scalability, centralize management and lower upfront costs. In many contexts, it’s exactly the right choice. But the decision between cloud and edge shouldn’t just be technical—it must be strategic. Too many executives are making it without accounting for the full range of trade-offs.
The pressure I see businesses face most often involves short-term costs. Edge deployments require more upfront investment—in hardware, local processing capability and the infrastructure to support distributed systems. Cloud models have a lower barrier to entry and a monthly cost that’s easier to budget. The problem is that “easy to budget” and “right for the environment” aren’t always the same.
I recall one customer who immediately dismissed an on-premises deployment, assuming the cloud would be cheaper. We had to help them understand that “cloud” doesn’t always mean outsourcing—organizations can achieve centralized management by hosting the platform in their own environment while retaining full data control. The moment we shifted the conversation from device counts and monthly line items to security, compliance and long-term ownership, the decision became clear.
To get this right, executives need to ask the right questions. Don’t just ask, “What’s the cost?” Ask, “What happens when the connection fails?” Go beyond asking, “Can this scale centrally?” Ask, “Does the environment cooperate with that assumption?” I believe those who work through those trade-offs explicitly instead of defaulting to one model because it’s familiar can build systems that hold up.
Building Where The Work Happens
The edge isn’t a concept; it’s a factory floor, a remote pipeline, a campus with unreliable connectivity and a security team that needs real-time awareness. It’s every environment where constant, fast, reliable cloud access isn’t a safe assumption.
AI built for those environments looks different from AI built for a demo. It can function when conditions aren’t ideal, make decisions locally, handle intermittent connectivity and keep data where the risk of exposure is low. That kind of design can be the difference between a system that performs on its best day and one that performs on its worst.
The future of intelligent systems isn’t in one place. It’s distributed, resilient and built around where decisions need to happen. I believe the organizations that understand that, and build accordingly, will be the ones that are still running in five years.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?