HANOI, 4 AUG 2026 — Among the thirty products Vietnam has designated as strategically important, alongside vaccines, batteries, industrial robots and 5G equipment, sits an entry most states would not think to name: large-scale Vietnamese language models. The country has written a class of AI model into industrial policy, and attached the machinery of investment incentives to it.
Decision No. 21/2026/QD-TTg was signed on 30 April by Deputy Prime Minister Ho Quoc Dung and took effect on 1 July. It promulgates ten strategic technologies and thirty strategic technology products. A companion instrument, Decision No. 23/2026/QD-TTg, sets out seventy high technologies prioritised for investment and a hundred high-tech products encouraged for development.
What is on the list
The ten strategic technology groups are broad. Digital technologies come first — artificial intelligence, big data, digital twins, cloud and edge computing, the Internet of Things and blockchain — followed by next-generation mobile networks, robotics and automation, advanced biotechnology and biomedicine, advanced energy and materials, semiconductor chips, cybersecurity and quantum, marine and underground technologies, aerospace, and high-speed and urban railways.
The product list is more revealing, since it is split by how close each item is to market.
| Group 1 | Group 2 | |
|---|---|---|
| Products | 22 | 8 |
| Basis for inclusion | Established market, direct economic impact | Emerging technology and national-security priority |
| Examples | Vietnamese language models, edge-AI cameras, blockchain systems, 5G equipment, industrial robots, vaccines, batteries, UAVs | Specialised chips, quantum systems, rare-earth processing, satellite constellations, high-speed rail |
RECATOOLS summary of the two product groups in Decision 21/2026/QD-TTg. The 22 and 8 are as published and sum to the 30 the decision promulgates.
Group one is where a government says it expects revenue. By putting Vietnamese language models here, rather than with emerging technologies like quantum systems, Hanoi is claiming they are a product with a market now, not a research programme.
Why a list is a lever
The lists are not honorary; they are the eligibility gate for incentives. They replace their predecessors outright: Decision 21 supersedes Decision 1131/QD-TTg and Decision 23 supersedes Decision 38/2020/QD-TTg. From 1 July, a new project or application is assessed against the new lists rather than the old framework.
That cuts both ways, and the advice to investors is unusually direct about the downside. Where an activity was covered by one of the superseded decisions but is no longer expressly included, a company has to work out whether it still falls inside some broader category, or whether it needs an alternative basis for incentives altogether. Existing high-tech certificates are worth re-reading. An industrial policy that names winners is also, silently, dropping things it previously named.
The lists arrived inside a much larger day
1 July was not a quiet Wednesday in Hanoi. Twenty-nine laws, two ordinances and thirty-four implementing decrees — sixty-five instruments — entered into force simultaneously, spanning taxation, digital transformation, e-commerce, construction, cybersecurity, anti-corruption, customs and administrative procedure.
Several of them bear directly on anyone the technology lists are meant to attract. A new Law on E-commerce sets rules for online marketplaces, cross-border digital trade and platform responsibility. A new Law on Cybersecurity tightens data-governance and compliance duties for organisations running digital systems. A 0.1 per cent tax now applies to transfers of digital assets, gold bars, carbon credits and auctioned licence plates. And personal income tax is waived for qualified professionals working in semiconductors and artificial intelligence — a talent measure aimed at precisely the two sectors the strategic lists lean on hardest.
Two of those instruments point at the same target from opposite ends. Cybersecurity is one of the ten strategic technology groups, and a new cybersecurity law imposing data-governance duties took effect on the same day — the state promoting an industry and regulating its customers in a single morning.
Without that context, the lists look like a wish. Alongside a tax exemption for the engineers and a rewritten corporate income tax law taking effect the same day, they look like the index to a policy.
The capacity behind the list
Cloud computing platforms sit on the strategic product list, and the plant to run them is going up at a pace that suggests the list is describing something already under way rather than summoning it.
Viettel opened its An Khanh data centre at an investment of 17.5 trillion dong, about US$665 million, built to Uptime Tier III and running the operator's own AI systems. Its Tan Phu Trung facility is a larger proposition again — 140 MW, construction from April 2026, operations from early 2027 and completion targeted for 2030. CMC is building a chain of its own, including a hyperscale site in Hanoi's high-tech park, with about US$250 million committed and an initial 30 MW expandable to 120 MW.
The largest single commitment reported is a partnership of Kinh Bac City Development, AIC and VietinBank: roughly US$2 billion across ten hectares, a 200 MW IT load and capacity described as around 100,000 GPUs. A separate hyperscale project involving G42, Microsoft, FPT, VinaCapital and Viet Thai is put at a similar US$2 billion, and VNG's joint venture with ST Telemedia brought two AI data centres in Ho Chi Minh City into operation in the first half of 2026.
Those figures are announcements and commitments, not commissioned megawatts, and the gap between the two is where data-centre pipelines usually lose years. They establish direction, not delivery — which is also the right way to read a strategic list.
The sovereign-model question
The Vietnamese language model entry is the one to watch, because it is a bet with a visible failure mode. The case for a national model is straightforward: general models are trained overwhelmingly on English and a handful of other high-resource languages, and they cost more and perform worse per unit of text in languages that tokenise badly. A state that wants its own administration, courts and schools running on machine translation and summarisation has a real interest in models that handle its language efficiently.
The case against is that the frontier moves, and general models have repeatedly absorbed capabilities that specialists were built to own. A national model programme can be overtaken by a general model that was never aimed at the market at all. Nothing in the decision says how that risk is to be handled, which is not a criticism of a list — it is simply not what a list does.
The wider framing arrived three weeks later. On 25 July the same deputy prime minister who signed the technology lists approved a national digital transformation strategy for 2026 to 2030, with a vision running to 2045. Its objectives are qualitative rather than numeric — comprehensive transformation across sectors, national governance powered by data and AI, domestic capability in strategic digital technologies and platforms, and what the strategy calls firm protection of national digital sovereignty.
That last phrase frames the language-model entry. A domestic model trained for Vietnamese, named in an instrument that gates tax incentives, turns digital sovereignty from rhetoric into procurement policy. Whether it is achievable is a different question from what it is.
Vietnam has stated its ambition in an instrument with legal force, which is more than most of its neighbours have done. Indonesia's contribution to the same conversation has been a regional pact treating AI literacy as shared infrastructure. Vietnam has instead named the artefact it wants built and pointed the tax code at it.
