Vietnam Has Four AI Policies and No Agreed Measure of Return

Adoption Moved. Measurement Did Not.

On 29 September 2026 in Hanoi, Amazon Web Services released a study conducted by Strand Partners finding that 26 per cent of Vietnamese businesses now use artificial intelligence in some form, roughly 245,000 companies, against 18 per cent a year earlier. The same study found that 13 per cent of those adopters have a defined framework for measuring what the investment returns, and that 46 per cent have no reliable method of measuring returns at all.

Vietnam has assembled one of the most complete AI policy stacks in Southeast Asia in under two years, and almost every binding target in it counts an input: rankings, brands, research centres, open datasets, compute capacity, enterprise numbers, risk classifications. On the same day, Bain and Company published an arithmetic that cuts the other way. Its seventh global technology report argues the AI build-out needs about US$6 trillion in annual revenue by 2031 to pay for itself, roughly one additional percentage point of annual global GDP growth, and that existing applications get the industry only to US$1.2 trillion or US$1.8 trillion of it. The remainder is revenue nobody has invented yet. For a market whose instruments measure deployment and whose most-cited survey cannot say what deployment earned, that is the live question, and Vietnam currently has nothing pointed at it.

What the Policy Stack Actually Targets

The architecture went up in four pieces. Decision 127/QD-TTg of 26 January 2021 set the national AI strategy: into the top four ASEAN countries and top 50 globally by 2030, with 10 regionally recognised AI brands, three national centres for big data and high performance computing, and 50 linkable open datasets. It talks about reducing cost and improving productivity, and attaches a figure to neither.

Resolution 57-NQ/TW, issued by the Politburo on 22 December 2024, is the heavier instrument. It sets the digital economy at a minimum 30 per cent of GDP by 2030, R&D spending at 2 per cent of GDP with over 60 per cent from non-state sources, at least 3 per cent of annual budget expenditure to science, more than 40 per cent of businesses engaged in innovation activity, and at least five domestic digital technology enterprises on par with those in advanced economies.

The Law on Digital Technology Industry, issued 14 June 2025 and fully in force from 1 January 2026, designates AI system development and AI data centres as specially prioritised investment sectors, carrying preferential tax treatment, land use rights, customs advantages and the possibility of direct state support for infrastructure.

Law No. 134/2025/QH15 on Artificial Intelligence, passed on 10 December 2025 and in force from 1 March 2026, completes it. The law sorts systems into high, medium and low risk, requires providers of medium and high risk systems to notify the Ministry of Science and Technology through a one-stop portal, requires conformity assessment before high-risk deployment, mandates labelling of AI-generated content, and creates a National AI Development Fund. Systems already running on 1 March 2026 have 12 to 18 months to comply.

Read across the four, the pattern holds. Vietnam has built reporting rails and pointed them at risk. Nothing in the stack asks a company what the spending earned, and the incentives are granted at the point of investment and are not assessed against an outcome afterwards.

The Capacity Is Arriving Faster Than the Measurement

AWS opened a Local Zone in Hanoi on 19 June 2026, bringing single-digit millisecond latency and in-country data residency. The customer list the company published includes Techcombank, VPBank, VIB, Vinamilk, MOMO, Sovico Group and VTV, alongside smaller firms such as Trusting Social, Green SM and AI Hay. Pham Duc Long, Deputy Minister of Science and Technology, said the facility would "provide a low-latency, secure cloud computing platform and contribute to supporting Vietnamese businesses' technology needs". AWS says it has trained more than 157,000 people in Vietnam in cloud and AI skills.

At the startup layer the figures are strong. Strand Partners put average annual revenue growth among AI-native Vietnamese startups at 147 per cent and found them 4.8 times more likely to earn over US$1 million a year. NoteX, a voice intelligence platform, reports 1 million users across more than 20 countries in its first year and a 30 per cent reduction in processing cost and latency. Diaflow, an agent orchestration service built on Amazon Bedrock, reports more than 40,000 users, 75 per cent of them in the United States. Techcom Securities says 460 staff now use the AWS coding agent Kiro, cutting development time by nearly 25 per cent, with 70 per cent of generated code passing first review. OmiGroup says its OmiKG research tool identified 15 of 17 known molecular mechanisms for type 2 diabetes in a controlled evaluation.

Each of those figures was supplied by the company to a vendor that sells it infrastructure, and reported in that vendor's own release. None can be verified from outside. That is no reason to discard them, since firm-level deployment data in Vietnam is otherwise close to absent and the direction they point is consistent with the policy record. It is a reason to read them as company claims, and to notice what the companies were not asked.

Where the Argument Runs Out

Two things complicate it. Resolution 57 does carry a return target, and a demanding one: total factor productivity contributing more than 55 per cent of economic growth by 2030. That is an outcome measure, and it is quantified. And 71 per cent of Vietnamese adopters told Strand that their AI investments had broken even or delivered positive returns, while 64 per cent reported revenue up by an average of 15 per cent.

The thesis narrows accordingly. Vietnam is not indifferent to return. What it lacks is a measure operating at the level where the decisions are made. The TFP target is a national accounting measure, calculated after the fact by the statistics office and revised for years; it arrives as a verdict long after the decisions it would have informed. And the 71 per cent and the 46 per cent came from the same respondents in the same survey.

Nor is the 61 per cent of Vietnamese adopters still in an experimentation phase obviously abnormal. Bain's own framing treats absorption speed, the rate at which organisations move deployments into production, as the competitive variable, which implies most of the market is mid-absorption everywhere. The infrastructure for outcome reporting now exists in Vietnam even though it was built for something else: from March 2027 the AI Law's portal and national database will hold a registry of notified systems by risk class and sector, the first dataset that could be joined to firm-level financials.

What Would Settle It

The real test is the second cohort. Adoption moved from 18 to 26 per cent in a year, which means a large share of current adopters are inside their first 12 months and have nothing to measure yet. The question is whether the share with a defined ROI framework moves off 13 per cent as that cohort matures, or stays flat while the headline adoption number keeps climbing. Nobody can answer that now, on a single year of comparable history.

Three things will settle it, each checkable against a named source. Whether the National AI Development Fund, once its operating rules are issued under Law 134/2025/QH15, conditions disbursement on outcome reporting or on project category alone. Whether the share of adopters with a defined ROI framework moves in the next Strand Partners round. And whether the General Statistics Office's TFP contribution series, the one Resolution 57 commits to lifting above 55 per cent, moves in the 2027 and 2028 readings.

One limit should be stated plainly. The Strand Partners survey of 1,000 business leaders and 1,000 members of the public is commissioned by AWS and its sampling and weighting methodology is not public, so the adoption rate and the 245,000 figure derived from it cannot be independently reconstructed, and the company figures are self-reported. UNESCO's 2025 AI readiness assessment of Vietnam, an independent exercise, reaches a compatible conclusion on measurement, noting the absence of a unified coordinating body for AI governance and the lack of impact assessment on AI's workforce effects, but it predates the AI Law and does not survey enterprise returns. No independent survey of Vietnamese enterprise AI returns was identified. Where this article states a figure, it names whose figure it is.

Sources. AWS and Strand Partners, "Unlocking Vietnam's AI Potential 2026", released Hanoi, 29 September 2026; Bain and Company, 7th Global Technology Report, 29 September 2026; Decision 127/QD-TTg, 26 January 2021; Resolution 57-NQ/TW, 22 December 2024; Law on Digital Technology Industry, 14 June 2025; Law No. 134/2025/QH15 on Artificial Intelligence, 10 December 2025; AWS Hanoi Local Zone announcement, 19 June 2026; UNESCO Viet Nam AI Readiness Assessment Report, 2025.

Vietnamese names and terms appear without diacritics in this edition for international readability.