AI Development Thailand 2026 — Talent, BOI and Data Rules

“We have looked at generative AI in Thailand from every angle. The next step is building AI that fits our own operations. What we cannot settle is where the development should happen, who should do it, and what the team structure should be.” AI development in Thailand is a question we hear more and more from Japanese manufacturers with operations here. The option tends to be discussed purely in terms of low cost, but the factors that actually drive the decision are three others — whether you can hire the engineers, whether you can use BOI incentives, and whether you are allowed to move data out of the country. This article sets out the practical constraints of placing a development function in Thailand, and how to choose between three delivery models — offshore outsourcing, nearshore collaboration, and in-house development inside your Thai entity.

Why building AI in Thailand has become a real option

The digital economy is forecast to grow at 2.2 times the pace of overall GDP

Start with how much momentum Thailand’s digital sector actually has. Wetang Phuangsup, Secretary-General of Thailand’s National Board of Digital Economy and Society Office (ONDE, also reported as BDE), has projected that Thailand’s digital GDP will reach 5.7 trillion baht in 2026, roughly USD 171.8 billion. Growth in this sector is put at 5.6 percent, about 2.2 times the 2.5 percent growth rate of the Thai economy as a whole. Breaking it down further, the software segment is expected to grow 9.4 percent, one of the strongest rates within the digital sector. This outlook is reported by The Nation Thailand — Thailand digital economy outlook for 2026.

These figures carry two implications for Japanese companies. The first is that demand for software development inside Thailand is itself growing. Where demand grows, the number of firms and engineers able to deliver tends to follow. The second is that growing demand also means domestic and foreign companies competing for the same people. That double edge feeds directly into the talent question discussed below.

BOI investment applications have surged, with digital leading the growth

Policy is moving in the same direction. Investment applications to the Thailand Board of Investment (BOI) grew roughly 2.4 times year on year in the first quarter of 2026, and the digital sector is cited as one of the areas driving that growth. This trend is also noted in VOIX — Data Section secures BOI approval for an AI infrastructure business in Thailand.

What stands out in the pipeline is large-scale investment in data centres and GPU infrastructure. In the data centre segment, six projects worth a combined 95.8 billion baht — including a TikTok project — were approved, as reported by Bangkok Shuho — Report on Thailand data centre investment approvals. Separately, the Japanese firm Data Section obtained investment approval for an AI infrastructure business in Thailand, as reported by VOIX — Data Section secures BOI approval for an AI infrastructure business in Thailand. In other words, investment in compute is coming not only from large foreign players but from Japanese companies as well.

The point worth noting is that the investment is going into the compute needed to run AI, not into the buildings needed to use it. More data centres and GPUs inside Thailand means a higher chance of being able to keep training and inference environments within the country. That becomes a premise for the personal data discussion later in this article.

Distance and time difference from Japan — unglamorous, but it tells

Thailand, with Bangkok at its centre, has steadily built a presence as an offshore software development base. Cost competitiveness in outsourced development across AI, IoT and VR, together with geographic proximity to Japan, are the reasons usually given. This is noted in sources such as Time Doctor — Outsourcing to Thailand.

The time difference between Japan and Thailand is two hours. 9 a.m. in Japan is 7 a.m. in Thailand; 6 p.m. in Japan is 4 p.m. in Thailand. Working hours on both sides overlap almost entirely. Direct flights from Narita or Haneda to Bangkok take roughly six to seven hours, so a same-day return trip is unrealistic, but weekly travel between the two sides stays within a practical range.

Compared with outsourcing development to Europe or Latin America, this is a decisive difference. AI development frequently starts before the specification is fully settled, and the pattern of “look at today’s results, decide tomorrow’s direction” comes up constantly. When you are running that cycle, the number of hours per day in which both sides can talk in real time becomes a productivity gap that never shows up on a quotation.

Adopting AI and developing AI are two different decisions

Before getting into the substance, it is worth defining what this article covers.

When Japanese companies in Thailand consider AI, three decisions of quite different character often get mixed together. The first is whether to use off-the-shelf generative AI services in day-to-day operations. The second is how to select a company to build AI for you. The third — the subject of this article — is where to place the development itself and how to structure the team.

The first question, the “using” side, covers where Japanese assumptions stop working when you deploy generative AI in Thailand and how to think about the cost. We have set that out in Thailand AI Implementation 2026. The second question, the country-agnostic criteria for choosing a development company such as differences in contract structure and how to read a quotation, is covered in How to Choose an AI Development Company in 2026.

This article deals with the third. Before comparing the merits of individual development companies, we look at the constraints that apply simply because the work is being run in Thailand. Specifically, these three.

• Talent — can you hire or otherwise secure engineers capable of AI development in Thailand
• BOI incentives — can investment in a development function qualify for BOI privileges
• Data — are you permitted to move training data and customer data out of Thailand

These are matters your own organisation has to settle before deciding which development company to engage. Put the other way round, if you start collecting quotations while these remain vague, the assumptions collapse later and you go back to the drawing board.

Question 1 — can you hire the people to build AI in Thailand

Multiple studies point to a developer shortage

Talent is the first wall you hit when developing AI in Thailand. Several studies and industry reports indicate that the supply of developers in the Thai IT talent market continues to fall short of demand. Even within AI specifically, the view is that supply will not catch up for several years. This is covered in Aree Technology — Navigating the 2026 tech talent shortage and Asia News Network — Thailand AI ambition and skills deficit.

Some material does put a number on the size of the shortfall, but the counting scope and job definitions are not consistent, and figures range from tens of thousands to well over a hundred thousand depending on the source. For that reason this article does not adopt any single figure. If you quote a shortfall number in an internal approval document, go back to the source and confirm what it actually counted before using it. What matters for practical planning is not the number itself but the consequence — hiring takes longer and costs more than initially assumed.

AI talent commands higher pay than general IT roles

The shortage shows up most plainly in salaries. Where supply fails to keep pace with demand, pay levels generally rise, and since supply shortfalls are documented in Thailand’s AI and data field as described above, you should budget on the assumption that people with AI and data skills command higher pay than general-purpose IT generalists. The size of the premium varies with how the role is defined and with years of experience, so no single uplift figure applies across the board, but you should at least assume you cannot hire at the same market rate as a general IT position.

The implication is that the assumption “labour is cheap because it is Thailand” simply does not hold in the AI field. If you build a budget benchmarked against clerical roles or general application development, you will post the vacancy and receive no applications. At the budgeting stage, confirm AI and data market rates individually with recruitment agencies rather than relying on general IT rates.

Even once you hire, Japanese expatriates alone cannot run it

There is a second point that is easily missed in team design. A handful of Japanese expatriates trying to control AI development will not make the work function in practice. There are three reasons.

• The bulk of the development capacity will be Thai engineers, so technical discussion happens in English or Thai
• Interpreting shop-floor data requires interviewing Thai operators directly
• Expatriate postings end after a few years, but AI models need continuous care once they go into operation

The third is especially important. AI development does not end at delivery. When the data distribution shifts, accuracy drops; when the process changes, the inputs themselves change. If that context is lost when expatriates rotate out, a model that was working ends up with nobody able to maintain it. From the earliest stage of development, decide whether to place someone on the Thai side who can take over operation, or to secure a partner who will act as the handover destination.

The realistic options for securing talent

Taken together, there are three routes to securing AI development talent in Thailand. The first is hiring directly. The second is outsourcing to a Thai development company. The third is partnering with a Japanese-affiliated systems company based in Thailand and drawing on the engineers it already employs. The strengths and weaknesses of each are covered in the delivery model comparison later in this article.

Question 2 — can BOI incentives be used for AI development in Thailand

Digital industry is among BOI’s promoted activities

Any discussion of investing in Thailand inevitably raises the BOI, the Thailand Board of Investment, and its incentives. The BOI recognises digital industry as one of its promoted categories, and businesses relating to software development and digital services can fall within scope. The available incentives include basic measures such as exemption from import duty on machinery and corporate income tax exemption depending on the activity category. The overall framework is set out by JETRO — Investment incentive schemes for foreign companies in Thailand.

The thing to understand here is that BOI incentives are not something that comes automatically once you set up a company in Thailand. They require that the business activity falls within a promoted category, that the application is approved, and that you continue to meet the post-approval conditions.

Additional incentives come in three directions

On top of the base incentives, there are mechanisms that add further privileges when certain conditions are met. Three directions are relevant when considering an AI development base.

• Where research and development spending meets the threshold, the corporate income tax exemption period is extended. The extension is set in tiers according to the scale of spending, with an upper limit of around five additional years indicated
• Where advanced technology training is delivered, additional incentives are available
• Where the site is located in specified areas, additional years of corporate income tax exemption are added. Some areas indicate an addition of around three years

AI development sits comfortably with all three. Model research and development can itself qualify as R&D spending; technology transfer to Thai engineers naturally takes the form of advanced technology training; and a development base is far less constrained by location than production equipment, so it is easier to choose an area with richer incentives.

That said, exactly which expenditures are recognised as R&D, how the delivery of training is evidenced, and which tier the extension falls into all depend on the content of the application and on the current administration of the scheme. Figures such as year counts and percentages can change with scheme revisions, so treat the latest BOI publications and confirmation with specialists who handle BOI applications as prerequisites for any assessment. The figures in this article are indicative starting points only.

Do not miss the timing of the BOI application

The most common failure in practice is timing. As a rule, BOI incentives apply to investment and business activity that takes place after approval. If you have already imported the machinery, or already incorporated and started trading, realising afterwards that BOI was available makes retroactive application difficult.

For an AI development base, the bulk of the investment is usually personnel and cloud cost rather than hardware, which invites the judgement that “there is no major capital expenditure, so BOI is irrelevant.” But corporate income tax exemption bites after you become profitable, and that is a separate matter from whether equipment is involved. At the point you decide to establish a development base, or to add an AI development function to an existing entity, put the question of whether to apply for BOI on the table at least once.

The option of placing GPUs inside Thailand

As noted above, investment in data centres and GPU infrastructure inside Thailand has been approved. That means the compute needed for training and inference can increasingly be procured domestically. The choice between using an overseas cloud and using domestic compute is not a simple cost comparison — it ties directly into the cross-border data question below.

Question 3 — can training data and customer data leave Thailand

The PDPA has been fully in force since 2022

Thailand’s personal data law, the PDPA (Personal Data Protection Act), has been fully in force since 2022. Like Japan’s Act on the Protection of Personal Information and the EU’s GDPR, it sets rules on the collection, use and third-party disclosure of personal data.

In an AI development context, this law bites when the data you handle contains personal data. Even shop-floor data from a manufacturing plant can qualify if it includes operator IDs, working times, images showing faces, or attendance records, so you cannot dismiss it with “this is factory data, so it does not apply to us.”

Cross-border transfer sub-regulations were finalised in April 2025

The significant change in PDPA practice as of 2026 concerns the sub-regulations on cross-border data transfers. These were finalised and published on 7 April 2025. The development is explained in PwC — Trends in data-related regulation in Thailand and Lexology — Thailand cross-border data transfer rules.

The practical point is that a data recipient outside Thailand must be shown to have a level of protection equivalent to what the PDPA requires. The frameworks contemplated for demonstrating this include entering into Standard Contractual Clauses and establishing Binding Corporate Rules within a group.

Brought back to AI development, this affects situations such as the following.

• Gathering operating data from a Thai plant onto head office servers in Japan for training
• Sending Thai customer data to an AI service on a cloud in Japan or a third country for processing
• Handing data collected in Thailand to an overseas development contractor to build a model

All of these constitute transfers of data out of Thailand. The question is not whether it is technically possible, but whether the legal groundwork is in place.

Three things to settle at the design stage

In practice, we recommend settling these three points before development starts.

• Produce a single list of which data is stored where and processed where
• Identify the items that qualify as personal data and consider whether anonymisation or pseudonymisation can take them out of scope for cross-border transfer
• For data where cross-border transfer is unavoidable, decide who will put the contractual arrangements in place and when

The second has the biggest practical effect. What training genuinely needs is usually the time series of operations and the state of equipment, not information that identifies individuals. If you can design the data so that personal data never enters the pipeline, you may avoid the cross-border transfer procedure entirely. Conversely, once you have built a data lake that contains personal data, separating it out later is expensive.

Thailand’s AI legislation is still at draft stage

One further point deserves stating plainly because it is easily misunderstood. As of August 2026, no law regulating AI itself has been enacted in Thailand. The draft AI Act published by Thailand’s Electronic Transactions Development Agency (ETDA) on 9 July 2026 is at the stage of a public consultation with a deadline of 14 August 2026, and the authorities are targeting finalisation of the bill within the year. The timing of enactment and entry into force is not fixed. The issues under discussion include securing transparency, accountability for automated decision-making, and sector-specific rules for high-risk uses. This is explained in Norton Rose Fulbright — Thailand’s draft AI law.

Take care not to confuse Thailand with other Southeast Asian countries here. Vietnam already has AI legislation in force, and Thailand’s situation is different. Commentary articles that treat the two together are in circulation, so when you prepare internal material, write on the premise that Thailand’s AI law is at the review and draft stage.

The practical meaning is not “there are no rules right now, so do as you please.” Quite the opposite — since it is clear that rules will settle in the near future, you need a design that will not require rebuilding when requirements are added later. Concretely, that means keeping records of what data a model was trained on, keeping model decisions explainable, and preserving a step where a human makes the final judgement. Even where none of this is legally required today, building it in from the start is cheaper than adding it after implementation.

Choosing a delivery model — offshore outsourcing, nearshore collaboration, in-house development

With those three questions in mind, we can compare the structural options. Delivery models for AI development in Thailand fall broadly into three — offshore outsourcing, nearshore collaboration, and in-house development within your local entity.

Offshore outsourcing means contracting the development itself to a local Thai development company. Nearshore collaboration means partnering with a Japanese-affiliated systems company based in Thailand, combining requirements definition in Japanese with implementation on the ground in Thailand. In-house development means giving your Thai entity an AI development function, employing engineers and building internally.

Comparing the three on the dimensions that actually drive the decision gives the following.

The most important row in this table is where know-how accumulates. In AI development, the business knowledge of which data means what and how to interpret it is a far harder asset to reproduce than the work of building the model. Choose a model that leaves that knowledge outside your organisation and you will be explaining everything from scratch to an external party again on your second and third AI project.

The way cost arises also differs by model. Because price levels vary enormously with requirements, the table below organises not amounts but the structure of when and how cost occurs.

Because in-house development carries fixed monthly personnel cost, it works out expensive for a one-off project. On the other hand, where several AI themes come up each year, the total can end up lower than outsourcing project by project. The dividing line is your outlook — is there one AI you want to build, or will they keep coming.

Note that automating production equipment and developing AI call for different qualities in a partner. We have set out how to choose on the equipment side in How to Choose an Automation Company in Bangkok in 2026, which is worth referring to if you are considering capital investment alongside this.

The realistic answer combines all three

In practice there is no need to treat the three as mutually exclusive. A common landing point is a phased shift — launch the first project through nearshore collaboration to establish the working pattern, in parallel recruit one or two engineers into the local entity to take over operation, and carve out only the parts that can be standardised to offshore.

The advantage of this sequence is that development does not stall if recruitment goes badly. In a field where securing talent is uncertain, putting team build-up and development progress on the same schedule means a delay in either one stops everything.

Five common ways AI development in Thailand goes wrong

Importing Japanese specification assumptions unchanged

This is the most frequent failure. Hand over a specification premised on the forms, process boundaries and judgement criteria used at your Japanese plant, and it will not mesh with the shop floor in Thailand. Even when the same product is being made, how the process is divided, when records are taken, and how exceptions are handled differ site by site. Because AI learns from real shop-floor data, that gap feeds straight into accuracy. Before starting, build in a step to review several weeks of actual data from the Thai site and identify what differs from Japan.

Deferring PDPA compliance

The approach of “build something that works first, then consult legal” is dangerous in AI development. The composition of training data is hard to change later, and if the permissibility of cross-border transfer only becomes clear afterwards, you go back to redesigning data collection. At the data design stage, settle whether personal data is included and where processing will happen.

Missing the BOI application window

As noted above, incentives apply as a rule to activity after approval. At the point you decide to establish a development base or add a function, insert a BOI review at least once. If the conclusion is not to apply, that is fine. The problem is losing the option because you never considered it.

Estimating salaries at general IT market rates

Pay levels in the AI and data field run higher than general IT roles. Build a budget at general IT rates and you will post the vacancy, receive no applications, and see the hiring plan slip by six months. Confirm this field’s market rates separately at the budgeting stage.

Starting development without deciding who owns the operation phase

AI accuracy degrades after delivery, because the data distribution shifts. Start development without deciding who notices that shift and who performs retraining, and the ending six months later is “accuracy dropped, so we stopped using it.” Decide the owner of the operation phase at the same time as you decide the development structure.

Frequently asked questions

What are the advantages of doing AI development in Thailand

There are three main ones. First, proximity to Japan — a two-hour time difference and roughly six to seven hours by direct flight. In AI development, which proceeds before specifications are fully settled, the length of the window in which both sides can talk in real time translates directly into productivity. Second, the possibility of using BOI incentives for digital industry. Third, investment in data centres and GPU infrastructure inside Thailand is advancing, widening the option of keeping compute resources within the country.

What is the typical cost of AI development in Thailand

A single market rate cannot be given. The amount varies greatly with the type of AI being built, the state of existing data, and the number of sites in scope. When you build a cost outlook, however, keep two points in mind. First, pay levels for AI and data talent are noted to be clearly higher than for general-purpose IT generalists, so benchmarking personnel cost against general IT market rates will understate the budget. Second, AI development incurs continuing cost for retraining and data upkeep in the operation phase, not only the initial build cost. When you compare quotations, check whether that operational element is included.

Can BOI incentives be used for small-scale AI development

It depends on the business activity falling within a promoted category and the application being approved. Even where the investment scale is small, it may still qualify. That said, the requirements and conditions such as minimum investment amounts are set by activity category and can change with scheme revisions. To determine whether your own case qualifies, review the latest BOI publications and consult a specialist who handles applications in practice.

Does Thailand have regulation comparable to the EU AI Act

As of August 2026, no law regulating AI itself has been enacted in Thailand. The governance framework is at draft stage, with review and a public consultation under way aimed at finalising the bill, and the timing of enactment and entry into force is not fixed. The issues under discussion are transparency, accountability for automated decision-making, and sector-specific rules for high-risk uses. Note that Vietnam already has AI-related legislation in force, so its situation differs. Commentary treating the two countries as equivalent is in circulation, so distinguish between them when preparing internal material.

Is it acceptable to send data from our Thai plant to head office in Japan for training

If the data contains personal data, it falls within the PDPA’s cross-border transfer rules. Under the sub-regulations finalised on 7 April 2025, the recipient outside Thailand must be shown to have a level of protection equivalent to the PDPA. Frameworks such as Standard Contractual Clauses may be required. Because operator IDs, facial images and attendance records can qualify as personal data, we recommend first taking an inventory of the data in scope and considering whether anonymisation or pseudonymisation can remove it from cross-border scope. Confirm case-specific judgements with a legal specialist.

Summary

Here are the key points of this article.

Building AI in Thailand has become a realistic option, carried by two tailwinds — growth in the digital sector and the trend in BOI investment. Thailand’s digital GDP is forecast to reach 5.7 trillion baht in 2026, roughly USD 171.8 billion, and its 5.6 percent growth rate is about 2.2 times the 2.5 percent for the economy as a whole. The software segment is put at 9.4 percent growth. Investment applications to the BOI grew roughly 2.4 times year on year in the first quarter of 2026, with digital cited as one of the areas driving that growth. Large investments in data centres and GPU infrastructure continue to follow.

Against that, three constraints bite once you actually try to run development. The first is talent. Multiple studies indicate that developer supply continues to fall short of demand, and pay levels in the AI and data field run higher than general IT roles. Budget at general IT rates and hiring will not progress. The second is BOI. Digital industry is a promoted category, with additional incentives available in three directions — R&D spending, advanced technology training, and area of location. But because incentives apply as a rule to activity after approval, missing the application window removes the option. The third is data. The PDPA has been fully in force since 2022, and cross-border transfer sub-regulations were finalised on 7 April 2025. A recipient outside Thailand must be shown to have an equivalent level of protection.

Thailand’s AI law itself has not been enacted, and review aimed at finalising the bill is still under way. The situation differs from Vietnam’s, so distinguish between them in internal material. This does not mean freedom because there is no regulation — the practical preparation is to build recording of training data and explainability into the design from the outset, on the premise that requirements will settle in the near future.

Delivery models fall into three — offshore outsourcing, nearshore collaboration, and in-house development. The dividing lines are whether the AI you want is a one-off or the start of a continuing stream, and whether the know-how needs to stay inside your company. In practice we find the realistic answer is often a phased shift — launch the first project nearshore, secure people in the local entity in parallel, and carve out the standardisable parts to offshore.

Talk to us at the early stage of your assessment

AI development in Thailand is an area where a great deal has to be settled on your own side before you start comparing development companies. Which data will you use, where will it be processed, will you hold the talent in-house, will you pursue BOI. Start collecting quotations before that is organised, and every change in assumptions sends you back to redo the comparison.

TOMAS TECH supports Japanese manufacturers in Thailand from the shop floor upward — from production management systems and IoT-based collection of shop-floor data through to assessing AI use. It is entirely fine if neither budget nor timing is settled, or if you are not yet sure whether in-house or outsourced suits your organisation. We can start simply by checking together how usable your current data really is. Please get in touch through our contact page.

References

• The Nation Thailand — Thailand digital economy outlook for 2026
• JETRO — Investment incentive schemes for foreign companies in Thailand
• VOIX — Data Section secures BOI approval for an AI infrastructure business in Thailand
• Bangkok Shuho — Report on Thailand data centre investment approvals
• PwC — Trends in data-related regulation in Thailand
• Lexology — Thailand cross-border data transfer rules
• Aree Technology — Navigating the 2026 tech talent shortage
• Asia News Network — Thailand AI ambition and skills deficit
• Time Doctor — Outsourcing to Thailand
• Norton Rose Fulbright — Thailand’s draft AI law