by 
“By contributing to an open-source, transnational effort that aligns with its own industrial priorities, Vietnam inserts itself not as a peripheral adopter but as a co-author of global AI infrastructure. This is asymmetry as strategy: composing relevance not by competing at the centre, but by accruing influence at the edge.”
I’ve now read this essay several times.
It’s compelling in its scope and language, inspiring and hopeful as well (which is not always the case when I read things about AI). It’s also troubling for what it misses, including a necessary discussion about politics. Whatever else we know about AI and AI policy, we know it’s all about politics. Technology always is.
With elegance and clarity, it sketches a “third way” approach to securing AI sovereignty. And it does so from the perspective of the “Global South”, whose interests, expertise and influence in much of the AI debate seems too often to be mute or missing.
All of which feels freighted with big implications for global as well as Australian policy and public leadership. Does the intent and architecture of Vietnam’s strategy offer any guide to possible Australian responses to the thin menu of options we seem to face – regulation or innovation, Europe or the US, the US or China?
What follows is a quick guide to the essay’s narrative and its interrogation of viable alternatives to the bleak choice between domination by Washington/California/Brussels/Beijing as the only viable architecture for AI’s global regulation and governance. Does this story offer an escape from the “dictatorship of no alternatives”?
There is an alternative
Roberto Unger would be pleased. Turns out there is an alternative. But it relies on a high sense of national strategic purpose as well as considerable political and bureaucratic capability to bend the arc of AI’s possibilities towards choice and independence.
At the heart of this approach is the story of what the essay calls “epistemic self-determination.” It’s the equivalent of pronouncing “we’ll decide what others know about us and the terms and conditions under which they can find out.”
To reach that point, the Vietnamese story (and they are not alone) combines clear strategic intent, a pragmatic approach to blending the necessary components of infrastructure and digital capability from different sources and an ability to navigate a relationship with the big players that ensures productive engagement without conceding control and domination. In other words, strategy committed to playing fully – to dance with giants – without getting trampled to death in the process.
What’s at stake?
Sovereignty, the essay declares, “is authorship of the perceptual field itself. What the development lexicon still dismisses as “local innovation” is, in truth, a claim to epistemic self-determination”. [Emphasis added].
How has Vietnam pulled off that trick?
One key component is the willing involvement of the business community, in this case FPT, a global technology and IT services provider which has becomes integrated into the strategy,
For example, it will open its “core tech stack” — language models, cloud infrastructure, even training data — to any domestic partner who wants to build with it. The idea is to open a national sandbox for controlled experimentation with the aim of creating a locally trained GPT-style model and support a state-backed push to teach AI in schools.
Three important signals right there: infrastructure, capability and education.
The essay explains what a “stack” is and why it is so crucial in this context:
In computing, a “stack” is simply the layered architecture that makes technology run: chips and circuits at the base, then operating systems, then applications, all the way up to the user interface. Each layer builds on the one below. Decisions made at one level cascade upward. Which is why choices about the stack are never just technical — they decide who holds power, and who must follow.
To be clear – this is not a tech decision. It’s about power and authority. Or, in other words, politics (although the essay doesn’t say as much, or at least not directly, which is a problem).
Along with “stack” design comes the commitment to “an open, comprehensive and state-regulated electronic ecosystem.” That matters to navigate the “familiar poles of AI politics” – Silicon Valley’s proprietary platforms and Beijing’s centralized infrastructure. They are never named, “but everyone in the room understands what is being contested: who gets to define the terms of intelligence itself.”
In a world where AI is changing the terms of engagement around the very notions of knowledge, intelligence and learning, Vietnam is responding with an instinct for a robust response that privileges a measure of local or indigenous control and authority.
On its own terms
The strategy is a claim to AI sovereignty: “the ability to build and govern infrastructures on Vietnam’s own terms while still enabling cross-border flows of data, talent and computation.”
The essay explains that AI sovereignty does not mean isolation, but “authorship — deciding which data, models and rules shape, and will shape, how machine intelligence is built and deployed.” Vietnam is not picking sides. Neither fully aligned with the U.S. nor China, “it is assembling a third stack that draws selectively from both sides while cultivating its own infrastructural sovereignty”. In a way of course it is picking sides. It’s picking Vietnam’s side. Which is inspiring and troubling at the same time.
It’s a strategic and policy choice of infrastructural nonalignment: “modular, adaptive and deeply attuned to the asymmetries of global AI.” In declaring its own stack, the essay continues, “Vietnam claims the right to decide how reality itself is translated into machine-readable form — what becomes visible, knowable and actionable to AI systems.”
You see what I mean about politics? The intent to “decide” reality is about as political as you can get. Who decides? On what terms are those choices made? Who can influence them? Who can challenge or change them? There’s no discussion here of that dimension nor how it is being wrangled in Vietnam.
Epistemic discretion and the risk of new exclusion
The essay then unpacks a little further some of the work that turns this framework into reality.
For example, the ambition is not just technical performance, but epistemic discretion — the ability to decide which models are retrained, how they are tuned and for whom they are made to speak.
The idea is to make sure the third stack is not sealed off but “selectively permeable: borrowing where it must but governing what it borrows — even if stitching together components from competing powers requires constant negotiation of technical standards and political constraints.”
So the story seems to be a national determination not to give into the big global players but to remain open to their capabilities and services (there’s little choice on that front anyway) that can be woven into the Vietnamese “stack”. But even if that happens, there’s still the risk that Vietnam won’t “transmute this influx of code and capital into sovereign capacity before the licenses and safety regimes around it harden into a new perimeter that encloses — or possibly imprisons — its third stack.”
The essay continues:
Many view AI geopolitics as a culture war between Silicon Valley’s libertarian individualism versus China’s communitarian authoritarianism. That familiar tableau of cowboy disruptors and state-backed titans still lingers in op-eds, but it obscures the quieter territorial redrawing that’s occurring along the infrastructural level.
The sharper fault line now runs not between nations but infrastructures — between the guarded logic of proprietary systems and the unruly emergence of open-weight models; between centralized command and distributed improvisation; between the doctrine of safety and the discipline of scrutiny.
What travels across borders aren’t values per se, but configurations of infrastructure: model weights, licensing schemes, data regimes, cloud dependencies and developer ecosystems. These are “the substrates through which AI systems are made legible, tractable and governable.” It is these substrates — rather than grand narratives about freedom or control — that shape how knowledge is produced, validated and operationalized.
The result “is a stack that is assembled, rather than monolithic, with domestic platforms, a sovereign cloud, high-performance compute and transnational research folded into its modular system.” Each of its layers carries different dependencies, but together “they allow Vietnam to hold authorship over the shape, orientation and reach of its AI infrastructure — one that’s stable enough to anchor public deployment and open enough to adapt or travel.”
What does this mean?
The story is “a quiet counterexample”, the essay argues, to the prevailing narrative of AI as an ideological battleground. The idea is that Vietnam, rather than choosing between spheres, is working with partners like FPT to build “connective tissue — embedding Vietnamese researchers within Mila’s lab, circulating knowledge across borders and “shaping governance standards from a position that is neither defensive nor derivative. “ Where too often openness is declared, this is what “infrastructural diplomacy” can look like.
Vietnam is not alone
The essay points out that this approach, which it describes as “asymmetry as strategy: composing relevance not by competing at the centre, but by accruing influence at the edge” is also fuelling similar work in other countries including Malaysia, Indonesia and the UAE. Taking the need for cultural and linguistic diversity as a distinct opportunity, these are national AI strategic responses that “point to a wider shift: not a rejection of global AI paradigms, but a refusal to be wholly contained by them.”
Their task is to “hold open the space between technical borrowing and epistemic capture.” The real question is not whether nations can build independently, but “whether they can stay in control of what their systems are allowed to know, remember and act upon.” The third stack holds an “exquisite contradiction” – it both “evades and entangles itself with old powers”.
Regularly through the essay, the essence of Vietnam’s strategy is reinforced through variations on this theme – a both/and approach, not an either/or, or perhaps more bluntly making every effort to have its cake and eat it too.
The emphasis is on technology transfer, ecosystem development and workforce capacity. The partnership with FPT ”exemplifies Vietnam’s strategy of courting foreign investment while cultivating domestic sovereignty.”
What is sovereignty?
A thread through the essay is AI sovereignty.
Whatever it is and outside of the “US-China vacuum”, it’s “not a banner‐waving claim to territorial control;”. Rather it manifests as “the quiet right to decide what counts as knowledge and how that knowledge shows up in the world.”
The essay explains that by refusing to license its perception of reality from OpenAI, AWS or Alibaba Cloud, Vietnam “reserves the right to set the horizon of what can be perceived, queried and disputed within its own techno‐social field. The third stack becomes a sovereign entity — a self-authored architecture of appearance.”
The tone of almost righteous defiance is stirring; the implications of the ambition for (presumably) the Vietnamese communist party to “set the horizon of what can be perceived, queried and disputed” might give at least some pause.
This is not, though, about “localization” or “self-reliance.” The question, the essay argues, is no longer whether Vietnam can train a Vietnamese GPT, but “whether it can dictate the contours of Vietnamese reality as machines come to perceive it”. In other words, “sovereignty is authorship of the perceptual field itself” or “a claim to epistemic self-determination.”
The implications of a strategy built around epistemic self-determination reverberate deeply and widely across society and the economy.
Sovereignty is “the power to decide what appears to machines — and, through them, to the humans and institutions that depend on their judgments.” The power, in other words, to decide who gets to know what and on what terms. The argument here is that nonaligned stacks “make that power visible precisely because they embody a third way, asserting authorship outside the U.S.-China duopoly”. This is a “design choice, shaping what becomes visible to machines and, by extension, to societies.”
What’s happening is a series of choices that is nothing less than deciding “how knowledge is structured, what data is treated as relevant and which forms of uncertainty are permitted or pre-emptively excluded.”
This is not simply a question of bias or inclusion. It is a structural matter: “what kinds of questions a model is designed to answer, what counts as a valid input and whose epistemologies are legible within its architecture.” Basically, who gets to decide who knows what. Big calls.
What gets excluded — “polysemy, dialect variation, historical opacity” — is just as crucial as what gets encoded, because “these choices determine whose worlds become legible to machines — and by extension, retrievable to future humans through them — and whose are consigned to obscurity.”
The problem with this kind of flexibility in “stack governance” is that the shift from closed to open weight or what the essay calls “semi open” stacks means missing out on some measure of consistency and interoperability in pursuit of ”epistemic pluralism”. The risk is that systems become unstable and “create frictions across borders, and open governance gaps precisely where alignment is most needed.”
Here’s how the essay explains the consequences:
An agent trained on a proprietary U.S. stack might assume individual agency, default to English-language documentation or prioritize efficiency over negotiation. An agent built on a localized Vietnamese or Indonesian model, by contrast, might be embedded with different priors — attuned to collective coordination, informal hierarchies or context-sensitive constraints. These are not just behavioural quirks. They are epistemic scripts — coded assumptions about what the world is, how it works and how action within it should unfold. [Emphasis in original]
An epistemic script or ‘coded assumptions’ implies structural shifts to define and determine choices and behaviour. Contesting those structural boundaries and barriers – in other words, politics – presumably is difficult, if not impossible.
The friction of adaptation
The essay tells a story of the attempt to defeat Unger’s dictatorship of no alternatives by showing there are other choices. Accepting there is only a choice between two competing forms of subjugation is a myth. It should actively be resisted not by assertion and pious hope but through infrastructural design and national strategic policy.
The essay concludes that Vietnam’s third stack “is more than a hedge against platform dependence: it is a rehearsal for a future”. It enacts a worldview “not defined by Silicon Valley or Beijing, but by the granular, situated logics of a sovereign digital ecology.”
A final aspiration:
The future of AI will not be charted by accelerationist slogans or neatly layered diagrams. It will surface — messy, uneven, tactical — from the friction of adaptation and the patient labour of coaxing disparate systems into dialogue. To read this terrain is to steer between hype and despair, tuning into the pulse of alignment: code splicing into cable, vision bending to vernacular, sovereignty assembled incrementally. Like a signal routed through stray relays, the coming architecture will glow with detours that seldom make headlines yet quietly redraw the map
Some of this, certainly its technical underpinnings, is beyond my paygrade and I suspect well outside the comfort zone of most competent policy makers and political decision makers.
And yet it feels heavy with both astonishing possibility and dire warnings about a world in which national decisions to curate a sense of sovereignty and national agency in the face of AI’s unsettling disruptions are not technical or transactional, but deeply structural, cultural and political.
Leave a Reply