Interview | Wanja Oberhof, CEO of Subsquid Labs
As artificial intelligence(AI) becomes the foundation of economic competitiveness, the debate over digital infrastructure is shifting beyond cloud computing to the resilience, sovereignty, and trustworthiness of the data that powers AI.
The October 2025 AWS outage exposed how deeply the global economy depends on a handful of centralized platforms, raising new questions about whether efficiency has come at the expense of systemic resilience.
In this exclusive written interview with Korea IT Times, Wanja Oberhof, CEO of Subsquid Labs, argues that the next generation of AI infrastructure must be built on verifiable, distributed, and resilient data systems rather than centralized architectures designed primarily for operational efficiency.
From digital sovereignty and decentralized infrastructure to Europe's MiCA framework and the future of AI-native data networks, Oberhof explains why "trust"—not simply access or speed—will become the defining strategic asset of the AI era.
The October 2025 AWS outage exposed how much of the digital economy depends on a handful of centralized cloud providers. Was that incident an operational failure, or did it reveal a deeper structural vulnerability in today's internet architecture?
Wanja Oberhof:
Both, though the operational failure is the less interesting aspect. Outages happen, and every provider has them. What made October notable was the blast radius, as the root cause was that core dependencies like authentication and routing were tied to a single region.
This is the predictable effect of a design the industry chose. We have spent two decades optimising for efficiency by concentrating workloads with as few providers as possible. Concentration is what ultimately produces correlated failures. Most organisations hedge by running in a second region of the same provider. That's like keeping your emergency cash in a second account at the same bank. Reassuring, until the bank itself is the thing that fails.
None of these choices were unreasonable. Every company that moved to a major cloud provider made a rational swap, a little independence for a great deal of efficiency. No servers to run, no data centre to staff, better uptime than most could manage alone, at lower cost. The problem is that the whole industry made the same trade at the same time. The risk did not disappear; it pooled.
We built an internet that is extraordinarily efficient but quietly fragile. We can’t prevent the next outage, but we can decide how much of the economy we leave exposed when it does happen.
As AI becomes the foundation of economic and national competitiveness, should digital infrastructure be regarded as strategic national infrastructure in the same way as energy, telecommunications, and transportation? Why?
Wanja Oberhof:
If digital infrastructure stops, does ordinary life stop along with it? Digital infrastructure has crossed this threshold some time ago, as services such as payments, logistics, healthcare records, and public services all sit on top of it.
AI raises the stakes completely. AI is a direct function of compute and data availability. A country's ability to compete in AI is bounded by its access to both, and increasingly, so is its ability to deliver basic services, as AI systems are embedded into everything from fraud detection to diagnostics to border control.
What differs this from energy or rails is jurisdiction. A power grid sits inside a border. Cloud regions, model weights and data pipelines do not. Much of the stack a nation depends on is owned and operated elsewhere, under another country's law and another company's commercial priorities. A government can regulate a domestic utility. It cannot easily regulate a design decision made in another hemisphere that determines whether a hospital can access its records on a Tuesday morning.
The question this raises is how much of the digital infrastructure a nation is comfortable with not controlling, and what it is prepared to build to change that.
Decentralization has long been associated with blockchain. Do you believe it is now evolving into a practical architecture for AI infrastructure rather than simply a cryptocurrency concept?
Wanja Oberhof:
It is, but the reason is less philosophical than people might expect. Decentralisation used to mean accepting a slower, more expensive system in exchange for resilience. It’s a trade any enterprise had no reason to make, but that has now changed. Distributed architectures compete on latency, uptime, throughput, and cost now, not ideology.
SQD runs close to 2,000 independently operated nodes worldwide, meaning no single node or operator controls the network. Losing one part of that doesn’t take down data availability for everyone else. This is resilient design, not a belief system. The fact is that enterprises aren’t choosing this architecture because they believe in the cypherpunk principles of decentralisation, but because it’s objectively faster and more consistent than the alternative. AI necessitates this. Autonomous systems making decisions in real-time can’t afford single-point-of-failure risks or blind trust in single parties. The workload needs to be defined by the architecture now, not the other way around.
Many organizations continue to prioritize efficiency and cost over resilience. What changes—in technology, regulation, or business incentives—will be necessary before resilience becomes a primary design principle?
Wanja Oberhof:
Regulation and technology will always have a part to play, but in the free market, incentives will always be the primary driver. Resilience is a hard sell because the payoff is invisible. Nobody gets credit for an outage that didn’t happen, while a cost cut shows up immediately on the balance sheet. This is why efficiency keeps winning budget conversations.
I think three things will shift this, however. Downtime needs a real cost attached to it, so resilience becomes a number in a business case. Buyers also need to start asking suppliers what else fails if they fail. Concentration risk stays invisible until somebody asks, and a 99.99% uptime figure means little if your provider's failure takes your whole stack down with it. Resilience, then, needs to stop being the expensive option, which is where distributed architectures earn their place in the conversation. Fragility gets designed out once it costs more than avoiding it.
As AI systems increasingly rely on massive amounts of real-time data, what characteristics will define the next generation of data infrastructure over the next decade?
Wanja Oberhof:
The next decade of data infrastructure will be defined by a formula. Correctness, freshness, and reliability, working together rather than any single feature working on its own, something which matters more as AI enters the picture. Autonomous agents can’t pause to sanity-check a bad response the way a human can, they need verified, real-time data by default. Blockchain data is no different.
Blockchain infrastructure was built to write data in, not serve it back out. Never mind serving it out at speed. Retrofitting that is the real engineering challenge, not reinventing the underlying architecture. Neutrality is also something that has to be considered here. Agentic systems will depend on this data, and data gated behind centralised single vendors will surface issues if these systems go down. The infrastructure that wins will be credibly neutral, built for machines as much as people.
Europe's MiCA framework reflects a growing emphasis on digital trust and governance. Looking ahead, do you expect regulation and decentralized infrastructure to evolve as complementary forces rather than competing approaches?
Wanja Oberhof:
I see them as complementary, not competing. Frameworks like MiCA exist because institutions need to trust the data and systems they're building on. That’s what decentralised infrastructure is designed to provide. Verifiable, tamper-proof data isn’t a workaround for regulation; it’s the substance regulators are asking for. From banks to payment providers, and large enterprises, the institutions moving on-chain now won’t accept a black box, regulated or not. They want data they can independently verify, and infrastructure that isn’t dependent on trust in one operator’s word. This is precisely what credible neutrality gives you: an audit trail regulation can actually point to.
Rather than regulation constraining decentralised infrastructure, I would argue it will accelerate adoption of it. The more scrutiny this space gets, the more valuable it becomes to have infrastructure that was built to be verifiable from day one.
If you were advising governments and enterprise leaders today, what is the single most important principle they should adopt when building AI-era digital infrastructure that will remain resilient over the next decade?
Wanja Oberhof:
Build for data you can trust, not just data you can access. Over the next decade, AI will make it trivial to spin up software. Anybody can use AI to generate complex software overnight. What AI can’t do is manufacture the underlying data itself, so the value and risk shift entirely to whether that data is accurate and verifiable.
Don’t build infrastructure that depends on trusting one party’s word; that will be the most important principle going forward in the world we’re entering into. Design it so correctness can be independently checked and not just assumed. Speed and scale alone don’t equate to resilience, but the ability to verify what you’re relying on before you act on it does. Governments and enterprises that get this right now will save themselves a far more expensive problem of retrofitting trust into systems later.
