At Computex this year, we ran a simple live poll at our booth. Among all the options we put in front of the thousands of engineers, architects, and decision-makers who stopped by, one answer pulled ahead of everything else:

"Defend Sovereignty: Say NO to data leaks and vendor lock-in!"

It wasn't the flashiest choice on the board. It wasn't a benchmark number or a shiny demo. It was a principle. And the fact that it won, across visitors from finance, healthcare, government, manufacturing, and pure-play tech — told us something we already suspected but hadn't heard stated so plainly by the market itself: the era of "just send it to someone else's cloud" is ending, and people know it.

This article is about why that vote matters, and why we believe the two forces behind it — AI for Sovereignty and an Open Source Foundation — are not two separate ideas but one architecture.

The two fears hiding behind one vote

Computex voting board: Attendees choose AI sovereignty to prevent vendor lock-in.

When someone chooses "say NO to data leaks and vendor lock-in," they're really naming two distinct anxieties that have finally converged.

The first is data leakage.

In the AI era this is no longer a hypothetical. An autonomous agent tasked with a routine query can, through excessive data access, pull and expose non-public information it was never meant to touch. In regulated industries this isn't an inconvenience. It's a direct violation of frameworks like GLBA, GDPR, and fair-lending rules, and it can end in fines, mandated disclosures, and lasting reputational damage. Worse, the pressure to move fast produces "Shadow AI": business units, tired of waiting on constrained IT teams, quietly adopt unvetted, ungoverned AI tools. Legacy technical debt doesn't just block safe AI adoption. It actively promotes the unsafe kind.

The second is vendor lock-in.

Different models are good at different things, and the landscape shifts monthly. Hardcoding a single provider into your applications doesn't just cost you flexibility. It hands a third party control over your economics, your data residency, and ultimately your strategic direction. Costs can spiral silently: a single runaway agent can burn through a monthly AI budget in hours because token consumption, unlike traditional infrastructure, is wildly unpredictable.

Put these together and you arrive at the real question every enterprise is now asking: Who actually controls my AI — my data, my costs, my compliance, my future?

That question is what we mean by sovereignty.

Sovereignty is not isolation — it's control

There's a common misreading of "sovereign AI" that treats it as a euphemism for "unplug from the world" or "build everything yourself." That's not what our Computex visitors were voting for, and it's not what we build for.

Sovereignty is the ability to make your own decisions — about where your data lives, which models you use, how much you spend, and how you prove compliance to a regulator. It is control, not isolation. A sovereign enterprise can still use the best commercial model on the market on Monday and swap it for a better or cheaper one on Tuesday, precisely because it never surrendered the controls that make switching possible.

Globally, this has become a survival-level concern rather than a preference. Executives increasingly view AI sovereignty as existential, and many organizations are actively seeking a "third option", an alternative to being locked into either dominant cloud bloc. Regulation is accelerating the shift: with the EU AI Act in force and data-localization laws spreading, high-sensitivity sectors like finance, healthcare, and government increasingly must deploy on their own terms to stay compliant. Sovereignty has moved from the "nice to have" column into the "non-negotiable" one.

But you cannot buy control back after you've architected it away. Sovereignty has to be built into the foundation. And that foundation has to be one you can inspect, own, and change.

Which is exactly why it has to be open.

Why open source is the foundation, not a feature

Here's the uncomfortable truth for anyone selling "sovereignty" as a proprietary black box: you cannot claim sovereignty over infrastructure you're not allowed to read.

If the control plane that governs your AI(the layer that decides how data flows, how models are routed, how costs are capped, how access is authorized)is itself a closed, opaque product from a single vendor, then you haven't escaped lock-in. You've just relocated it. You've made your defense against vendor dependence depend on a vendor. That's not sovereignty; it's sovereignty theater.

An open-source foundation breaks that trap in three concrete ways:

1. Transparency you can audit.

You can read exactly how your data is handled, how requests are routed, and where the boundaries are. For a CISO trying to certify controls under SOX, or a compliance team proving GLBA and GDPR adherence, "trust us" is not an answer. An auditable, inspectable code base is. Governing an autonomous agent means being able to trace what it did and why; a black box makes that impossible by definition.

2. Freedom that can't be revoked.

Open source means the exit door is always unlocked. No single company can change licensing terms, discontinue a product, or raise prices in a way that strands you. Your ability to run the software isn't contingent on anyone's business model surviving. That is the deepest form of anti-lock-in there is.

3. A community that compounds.

A foundation-backed, open project draws contributions, security review, and integrations from far beyond any one company's walls. The result is infrastructure that improves faster and hardens faster than a proprietary equivalent, and that belongs to its users, not to a quarterly roadmap.

Open source is not a "community edition" giveaway or a marketing checkbox. For sovereign AI, it is the mechanism. It's the only way the promise "your data stays yours, your choices stay yours" can be verified rather than merely asserted.

Where architecture meets principle: the AI-native gateway

digiRunner AI-native gateway architecture showing multi-LLM routing for sovereign AI infrastructure.

Principles need a place to live. Ours lives in the AI-native gateway, the control plane that sits between your AI agents and everything they touch. This is the layer where sovereignty is either won or quietly lost, and it's what our team builds with digiRunner, released as an open-source project.

Traditional API gateways were built for predictable, stateless, human-driven traffic. AI workloads break every one of those assumptions: token consumption is wildly variable, conversations are stateful, agents generate bursty and autonomous traffic, and tools are discovered dynamically at runtime through protocols like the Model Context Protocol (MCP). Bolting AI onto a legacy gateway doesn't work. You need a control plane that speaks AI natively.

Here's how an AI-native, open gateway turns the Computex vote into working architecture:

Against vendor lock-in: multi-model routing. Instead of hardcoding one provider into your applications, the gateway becomes an intelligent router across many models(commercial and internal alike), selecting by task, cost, or latency, with automatic fallback when a provider is down. Your applications talk to the gateway, not to any single vendor. Swapping models becomes a configuration change, not a rewrite.

Against runaway cost: centralized token governance. Token quotas, hard spend limits, AI-specific rate limiting, and real-time usage dashboards live at the gateway, not scattered across every application. Cost gets attributed per user, per department, per function, and budget ceilings are enforced before the bill shock, not after. This is how you keep economic control in your own hands.

Against data leakage: governance and AI-specific security at the edge. Fine-grained, context-aware authorization ensures agents can only reach the tools and data they're actually entitled to. The gateway adds AI-specific defenses that traditional security misses entirely(prompt-injection detection, anomaly detection, PII handling), enforced in one place instead of reimplemented (and forgotten) in every app.

Against opacity: full observability. A single user request can fan out into a dozen agent-to-model calls across multiple providers. The gateway traces the whole chain: which agent, which model, what cost, what latency, what outcome. That end-to-end auditability is what lets a regulated institution actually prove its controls rather than hope they held.

And crucially: legacy without rewrites. The gateway acts as a bidirectional translator, exposing your existing REST, SOAP, and gRPC services as MCP-consumable tools for AI agents without rewriting them. You modernize toward an agentic future while keeping — and governing — everything you already own.

Every one of these capabilities is a lever of control. Bundle them into an open, inspectable control plane, and "Defend Sovereignty" stops being a slogan on a poll board and becomes the actual shape of your infrastructure.

The third choice

There's a strategic dimension to all this that goes beyond any single enterprise. Much of the world does not want to choose between two dominant technology blocs. It wants a third option: an ecosystem where data autonomy is the default, where the underlying software is open and inspectable, and where control genuinely rests with the organizations and nations that adopt it.

That's the conviction behind the AISO(AI for Sovereignty) — direction we've been building toward: an ecosystem where open software and trustworthy hardware combine into a genuinely sovereign stack, and where "your data stays yours" is guaranteed by architecture and verifiable in code, not just promised in a contract. An open-source gateway like digiRunner is the natural control plane for that world, a neutral, transparent layer that no single vendor can capture.

What the vote was really telling us

When the crowd at Computex chose "Defend Sovereignty: Say NO to data leaks and vendor lock-in!" over everything else, they weren't rejecting AI. Just the opposite. They were saying they want to adopt it on their own terms — fast, but safely; powerfully, but under control; connected to the best the world has to offer, but never captured by it.

That's the whole thesis in one line: AI for Sovereignty, built on an Open Source Foundation. One is the goal, the other is the only foundation strong enough to reach it. You can't defend sovereignty with a black box, and you can't control your future with infrastructure you're not allowed to read.

So we're building the alternative in the open — and we'd love for you to build it with us.

→ Stay updated with the latest OpenTPI news and events: ►Visit our LinkedIn!

→ Explore digiRunner on GitHub: ►Explore our GitHub!

→ Learn more about the OpenTPI project: ►Read out Blog!

→ Subscribe to OpenTPI Newsletter: ►Click to Subscribe!

Defend Sovereignty: Why the Future of Enterprise AI Must Be Open