Vietnam AI Development 2026 — Talent, Incentives, Rules

The first question Japanese manufacturers face in Vietnam AI development is rarely technical. It is who should build it — an in-house team inside the local subsidiary, a Vietnamese software service firm, or an integrator that understands how a factory actually runs. Four inputs decide the answer — the price and depth of local talent, an incentive regime designed nothing like Thailand’s BOI, data transfer rules that dictate your cloud architecture, and the new AI Law that took effect in March 2026 and pushes real operational load onto vendors. This article works through all four using primary sources, and ends with a practical way to screen the firms you might hire.

Three delivery models for building AI in Vietnam

When head office in Japan sends down the instruction that the Vietnam site should be using AI too, the options you will actually weigh on the ground boil down to three. Which one is right has little to do with budget size or technical sophistication. It is decided by who will look after the result, and for how many years.

Hire an in-house team inside the local subsidiary

You employ developers directly at the Vietnamese entity and build the system internally. The big advantage is that domain knowledge stays inside the company, and you do not have to request a fresh quotation every time a requirement changes. The trade-off is that hiring and retention are hard, and until you have a second and third engineer the work is unavoidably tied to one person. Data engineers and people who can run MLOps are being fought over in the job market, and salary pressure in that band is strong. If you are dealing with production management or equipment data from a factory, it is easy to overlook that you need someone who can build the data foundation before anyone builds AI on top of it. Remote management from the information systems department in Japan works more easily to the extent that the time difference is small, but it tends to leave one thing vague — who exactly is responsible for gathering local business requirements.

Outsource to a Vietnamese AI development company

Vietnam has a deep bench of software service firms with a track record in offshore work for Japanese clients, represented by companies such as FPT Software, TMA Solutions, CMC Global, NashTech, Rikkeisoft and VNEXT. Some, like VNEXT, are headquartered in Hanoi with offices in Tokyo and Fukuoka, which makes it easier to secure a counterpart who is used to running requirements definition in Japanese. Bear in mind, though, that these names are examples of scale as offshore development companies. They are not a guarantee of any individual firm’s capability in AI specifically. At the selection stage, what you need to look at is not headcount but the maturity of operations and governance discussed later in this article.

Partner with an integrator who understands the operational side

The third model is to hand the work, AI portion included, to a partner who understands the factory-side systems — production management, energy management, equipment control — rather than one who understands only AI. Most manufacturing AI projects stall for unglamorous reasons that have nothing to do with model accuracy. The shop floor data is not being captured, or the new system cannot be connected to the production management platform already in place. This model is also the realistic choice when you want to roll out the same mechanism across sites throughout ASEAN.

The sequencing of adoption itself, and the wider regulatory picture including the AI Law, are covered in Vietnam AI implementation follows a different order. This article deliberately narrows the discussion to one question — how you secure the people who will build it.

Vietnam AI talent — what the rates and the depth actually look like

Before you settle on a delivery model, get the salary benchmarks in front of you. Both the in-house decision and any sanity check on an outsourcing quotation rest on these numbers.

Median monthly pay by role

Pulling from itviec’s Vietnam IT Salary and Recruitment Market Report for 2025-2026, and from a VietnamNet article citing TopCV’s sixth recruitment report, the medians for roles relevant to AI development line up as follows. The two sources differ and the currencies differ, so do not add them together and treat the result as a single market rate.

There are two things worth reading out of this table. The first is the ordering — data engineers command more than backend developers. The people who handle the data preparation that comes before any AI work are scarcer, and the market prices them accordingly. The second is that an entry-level backend developer and one with eight or more years of experience are more than four times apart. When an offshore quotation simply says “n developers”, the total can easily double or more depending on where inside that range those people actually sit.

A quotation without an experience breakdown cannot be compared

Lining up man-month rates side by side tells you nothing on its own. If a vendor is quoting a low rate on the back of a junior-heavy team, the review effort and the rework land on your side of the table. The practical way to verify a bid is to insist on a staffing table that states years of experience and role for every person, check it against the medians above, and compare the proposals including their senior ratio. While you are at it, confirm whether each person is dedicated to your project or shared with others. In offshore development it is common for man-month figures to be presented with no stated assumption about utilisation, and cases where the actual working time turned out to be half of what was implied are not rare. It also helps to keep the management band in mind. The median for CTO, CIO and VPoE level is VND 101.25 million a month, more than twice the developer level, so the share of architects and tech leads inside a proposed team moves the total substantially. Always ask for the organisation chart and the rate card together.

Talent supply has a different character in each city

Hanoi and Ho Chi Minh City remain the concentration points, as they always have been, but Da Nang has been moving noticeably in recent years. According to VietnamPlus, part of the Vietnam News Agency group, the Da Nang authorities put the city’s digital technology workforce at 53,000, and in July 2026 a framework was announced under which AWS and Renova Cloud will work with the city on AI infrastructure. At Da Nang Hi-Tech Park, a USD 1 billion AI data centre with 100 megawatts of power capacity moved to the groundbreaking stage in April 2026. When you are weighing candidate locations for a development base, the concentration of infrastructure investment belongs alongside rent and payroll in the decision.

Comparing with AI work at a Thai site — what Japanese manufacturers get wrong

Companies that also have a site in Thailand will inevitably ask which country should build the AI. Comparing headline labour rates is the wrong way to answer that. What you should be comparing is which country the relevant factory data sits in, whether that data can legally leave the country, and who will maintain the finished system, and during which hours of the day. The talent, incentive and data transfer picture on the Thai side is set out in How to start AI development in Thailand in 2026, so read the two together if you are evaluating both countries at once.

Investment incentives behind AI adoption at an overseas subsidiary

Vietnam has no single window equivalent to Thailand’s BOI. That absence is often misread as an absence of incentives. In reality several schemes work in combination, and AI and data are explicitly named as priority fields.

The Investment Support Fund under Decree 182/2024/ND-CP is the practical BOI equivalent

Decree 182/2024/ND-CP, promulgated on 31 December 2024, establishes the Investment Support Fund. For semiconductor and AI research and development projects, up to 50% of the initial investment amount is eligible for support, and high-tech enterprises can access annual cost subsidies covering research and development spending, workforce training and fixed assets. Note the design difference — this is a subsidy against the costs themselves rather than a tax reduction, which gives it a different character from the corporate income tax privileges granted by the BOI. Through a research and development phase that runs at a loss for an extended period, a cash subsidy bites harder than a tax break.

The Investment Law 2025 widened the incentive list to 23 fields

The Investment Law 2025 (No. 143/2025/QH15), effective from 1 March 2026, together with Decree 96/2026/ND-CP promulgated on 31 March 2026, expanded the list of incentivised fields to 23 categories. AI, semiconductors and data are named on that list. Because the framework is new there is little accumulated administrative practice behind it, and eligibility in any particular case has to be confirmed project by project. Still, it is worth recognising the direction of travel — the regime is designed to welcome investment in AI development.

NIC, the hi-tech parks and the national AI strategy

On the location side, the National Innovation Center (NIC) plus the three hi-tech parks in Hanoi, Ho Chi Minh City and Da Nang serve as the receiving points. Seed funding sits with the Ministry of Science and Technology while NIC’s programmes sit with the Ministry of Planning and Investment, and it pays to understand that split before you start knocking on doors. At national policy level, the National AI Strategy approved in January 2021 (Decision 127/QD-TTg) sets the goal of placing Vietnam in the top four in ASEAN and within the world’s top 50 for AI research and development by 2030. Secondary write-ups sometimes cite different rankings, so check the wording of the official document before you copy any of this into an internal deck.

Setting the two countries side by side makes the structural difference easier to see.

Because there is no single window, structuring an application in Vietnam takes more work. Read the other way round, if you separate out the eligible cost categories in your books from the very beginning, this is a regime that acts directly on cash flow during the development phase. It is worth aligning with your finance team on which costs could qualify for support before you sign anything.

Data regulation decides where the AI system can live

Alongside the delivery model and the budget, one decision should be made as early as possible — where the data will sit. Vietnam’s rules feed straight into any cloud-based AI architecture.

Cross-border transfer under Decree 13 reaches cloud usage

Decree 13/2023/ND-CP on personal data protection, effective from 1 July 2023, is known for its broad definition of cross-border transfer. Where personal data of Vietnamese citizens is processed automatically by a system located outside Vietnam, that in itself is understood to constitute a cross-border transfer. In other words, even without physically exporting anything, running AI training or inference on a cloud region abroad brings you inside the scope of the rule. On top of that, an impact assessment dossier must be filed with the Department of Cyber Security under the Ministry of Public Security within 60 days of processing starting. Any design that places an AI platform holding employee data or the contact details of business-partner personnel offshore needs to be revisited at this point.

Data localisation under the Cybersecurity Law

The 2018 Cybersecurity Law, together with Decree 53/2022/ND-CP promulgated on 15 August 2022 and in force from 1 October the same year, requires domestic data storage under Article 26. It applies unconditionally to domestic enterprises, while for foreign enterprises it is conditional. The obligation is triggered only where the company provides one of the specified services — telecommunications, cloud storage, e-commerce, online payment, social networks and others — and has received a warning from the Department of Cyber Security that its service has been used in violation of the law, and has failed to take remedial action. In that case the company must store the data domestically and establish a branch or representative office within 12 months of receiving the ministerial decision, with a minimum storage period of 24 months. An internal system running your own factory will not often fall within this, but if the deployment takes the form of a service open to external users, legal review is mandatory.

Domestic region options keep expanding

Infrastructure supply is catching up with what the regulations demand. In June 2026 an AWS Local Zone in Hanoi became generally available and was positioned explicitly as a way of meeting Vietnamese data residency requirements. The fact that global cloud providers are now installing facilities inside the country also means it is harder to use compliance as an excuse for architectural compromises. Together with the Da Nang data centre project mentioned earlier, the assumption of a few years ago — that keeping data in country leaves you with no options — has already collapsed. Building domestic storage into your design no longer forces the concessions it once did.

Four questions that separate building in-house from outsourcing

With that material in hand, the decision comes down to four questions. The point is to cut along the axis of time horizon and accountability, not technical difficulty.

First, will the target business process keep changing over the next three years? Areas where requirements refuse to settle lean towards in-house work, while areas with stable specifications lean towards outsourcing. Second, does the data involved include personal data? If it does, the constraints on where it can live are settled first, and your vendor options narrow with them. Third, who operates the system once it is finished? Building in-house when the Vietnam site has not one person assigned to operations is an architecture that stops the day its one developer resigns. Fourth, do you plan to roll the system out to other sites? If deployment across several ASEAN countries is the premise, using a different vendor in each country makes the maintenance burden jump sharply.

What to ask vendors in light of the AI Law

The AI Law (Law on Artificial Intelligence No. 134/2025/QH15), effective from 1 March 2026, affects how Vietnamese software service firms operate as well. DTS Software Vietnam, a local IT company, argued in a February 2026 article that outsourcing firms need to maintain model cards, records of data provenance, risk logs and human oversight checkpoints, and framed maturity in exactly these practices as the thing that earns trust from Japanese and global clients. From the buyer’s side of the table, that converts directly into a vendor selection checklist.

The items in this table look like a defensive AI Law compliance exercise, but in practice they work as a proxy for development quality. A partner who does not record where data came from cannot reproduce the same work when the model has to be rebuilt. A partner who has not designed a route for detecting bad outputs and escalating them to a human has not pictured the operational phase after go-live. Conversely, a company that answers these questions immediately and looks used to being asked is very likely already working with clients who impose the same standard. As a way of gauging a vendor’s real capability in a short meeting, this beats asking about accuracy figures or the number of projects delivered. Contract terms and how to read a quotation in detail are covered in How to choose an AI development company in 2026.

Fitting this into a Southeast Asia AI adoption plan

Optimising for Vietnam alone does not produce an optimum for ASEAN. In companies with sites in Thailand, Vietnam and Indonesia, letting each country handle data its own way regularly ends with head office unable to see a single group-wide indicator. The realistic approach is to complete storage and processing within each country, then collect only the aggregated indicators centrally. That structure minimises the cross-border transfer question while still allowing sites to be compared. Far more indicators than people expect — equipment utilisation, defect rates and the like — need aggregated values rather than raw data, and once the figures are shaped so that they contain no personal data, most of the operational constraints disappear. Vietnam’s talent costs and incentive regime pay off best when the country is used as the place where things are built inside that structure. Separating where you use AI from where you build it clears up a surprising amount of the decision.

For context on local demand, the VietnamNet article citing the TopCV report states that 40.7% of surveyed companies have already incorporated AI into their strategic plans. That figure rests on more than 3,000 responses and roughly 300,000 job postings, using data from the third quarter of 2025. Interest on the Vietnamese corporate side has clearly moved past the evaluation stage.

FAQ

How much do AI engineers cost in Vietnam?

In itviec’s 2025-2026 report, the median monthly salary for a data engineer is VND 41.3 million, for a data analyst or data scientist VND 40.65 million, and for a backend developer VND 37.8 million. The VietnamNet article citing TopCV’s sixth report puts the median for AI engineers and data engineers at over USD 1,580 per month, with solution architects with five or more years of experience at about USD 2,320. The sources and the currencies differ, so do not combine the two into a single market rate.

How should we estimate the cost of outsourcing AI development in Vietnam?

There is no reliable public statistic for vendor man-month rates, so no universal benchmark can be given. The practical method is to test a quotation against a structure built up from the salary medians above plus overhead and margin. Require a staffing table broken down by years of experience and role, and check whether a low rate is simply the result of a junior-heavy team. Claims in sales material about being a certain percentage cheaper than other regions usually come without any stated basis, and are safer left out of the decision.

How far does Vietnamese data regulation affect AI adoption at an overseas subsidiary?

Decree 13/2023/ND-CP treats automatic processing of Vietnamese citizens’ personal data by an overseas system as a cross-border transfer. Simply running AI in a foreign cloud region brings you within scope, so the region has to be decided at the earliest design stage. Where the rule applies, an impact assessment dossier must be filed with the Department of Cyber Security under the Ministry of Public Security within 60 days of processing starting. An AWS Local Zone in Hanoi became generally available in June 2026, which makes an architecture built around domestic storage considerably more feasible than before.

For a Japanese manufacturer with both Thai and Vietnamese sites, which should go first?

Do not decide on labour cost alone. The starting points are which country holds the production data in question, whether that data can leave the country, and who will run the system after handover. Thailand’s regime is built around BOI tax privileges, while Vietnam’s centres on cost subsidies from the Investment Support Fund, and the two act very differently on cash flow during a research and development phase. Where you have sites in both countries, designing around a split between where AI is used and where it is built tends to move faster in the end.

Summary

• AI development in Vietnam resolves into three options — an in-house local team, outsourcing to a software service firm, or partnering with an integrator who understands the operations — and the deciding factor is who will run the result and for how many years
• itviec’s report puts the median monthly pay for a data engineer at VND 41.3 million, above backend developers, showing that data preparation talent is scarcer and priced higher
• Backend developer pay ranges from VND 12.4 million at entry level to VND 54.9 million at eight or more years of experience, so a quotation without an experience breakdown is not a basis for comparison
• Vietnam has no single BOI-style window, but the Investment Support Fund under Decree 182/2024/ND-CP supports up to 50% of initial investment in semiconductor and AI research and development, and the Investment Law 2025 widened the incentive list to 23 fields
• Decree 13/2023/ND-CP treats automatic processing by an overseas system as a cross-border transfer, so cloud region selection has to be locked down early, in parallel with requirements definition
• The AWS Local Zone in Hanoi that went live in June 2026 and the large data centre project in Da Nang make designs premised on domestic storage far more workable
• Following the AI Law that took effect in March 2026, the state of a vendor’s model cards, data provenance, risk logs and human oversight works as a practical quality indicator during selection
• Across ASEAN, completing data processing within each country and sending only aggregated indicators to head office keeps the cross-border transfer question small while still allowing sites to be compared

Decisions about delivery model and data location are exactly the ones that cost least to revisit when they are discussed before requirements harden. We are happy to talk at the exploratory stage — how much your Vietnam site could realistically build in-house, or how a new system would connect to the production management platform you already run. We field a lot of enquiries from companies with sites in both Thailand and Vietnam, and can help untangle an architecture that spans them. Tell us where things stand and we will map out the options with you. Get in touch through our Contact Us page.

References

• Vietnam IT Salary and Recruitment Market Report 2025-2026 – itviec
• AI reshapes Vietnam’s job market but IT still dominates on salary and demand – VietnamNet
• Global technology firms partner with Da Nang to develop AI infrastructure – VietnamPlus
• Decree 182 on Investment Support Fund – KPMG Vietnam
• Vietnam’s new Law on Investment 2025 and Decree 96/2026/ND-CP – Lexology
• Vietnam Decree 13 and the new regulations on personal data protection – DLA Piper
• Decree 53 Provides Long-Awaited Guidance on Implementation of Vietnam’s Cybersecurity Law – Tilleke and Gibbins
• AWS Local Zones now available in Hanoi, Vietnam – AWS
• Vietnam’s AI Law takes effect on March 1, 2026 – what IT outsourcing firms should do next – DTS Software Vietnam