Türkiye's AI Action Plan 2026-2030 and Public Budgets: Using the 2 Percent Share

What the AI Action Plan 2026-2030 means for a public agency

The Türkiye Yapay Zekâ Eylem Planı 2026-2030 (Türkiye Artificial Intelligence Action Plan 2026-2030) was announced publicly on 13 June 2026 at the Türkiye Artificial Intelligence Summit. The plan was prepared by the Sanayi ve Teknoloji Bakanlığı (Ministry of Industry and Technology) and announced the same day by the Cumhurbaşkanlığı İletişim Başkanlığı (Presidency Directorate of Communications). For public agencies the plan cuts two ways: it commits a minimum share of public investment programmes to AI projects, and it positions the public sector as the first buyer of domestic AI solutions.

The announcement of 13 June 2026 describes the plan's frame in these words: "Eylem planımız 'fark et, istifade et, üret ve yönet' olmak üzere 4 temel eksen" (our action plan rests on four core axes: detect, benefit, produce and govern). The same announcement states the public sector's role as "Kamu sektörümüz başarılı ve yerli yapay zekâ çözümlerinin ilk alıcısı" (our public sector is the first buyer of successful domestic AI solutions). These two sentences are the policy ground a strategy department head can stand on when seeking internal approval.

The plan has a predecessor. The Ulusal Yapay Zekâ Stratejisi 2021-2025 (National Artificial Intelligence Strategy 2021-2025) was published through Presidential Circular No. 2021/18 in the Resmî Gazete (Official Gazette) dated 20 August 2021, issue 31574, and its period closed at the end of 2025. The 2026-2030 plan is the successor policy document, with quantified targets.

The targets below are taken verbatim from the Presidency Directorate of Communications announcement dated 13 June 2026. Presenting them with their source inside an internal briefing measurably strengthens the funding case.

• Added value: "Türkiye Yapay Zekâ Eylem Planı ile harekete geçireceğimiz kaynakların üreteceği katma değerin 1 trilyon lirayı aşmasını bekliyoruz" (added value expected to exceed 1 trillion Turkish lira).
• Public investment share: "Kamu yatırım programlarımızdan yapay zekâ projelerine en az yüzde 2 pay ayıracağız" (at least 2 percent of public investment programmes allocated to AI projects).
• Data centre capacity: "2030 yılına kadar ülkemizin veri merkezi kurulu gücünü en az 1 gigavata (GW) çıkaracağız" (installed data centre capacity raised to at least 1 GW by 2030).
• Infrastructure resources: "Veri merkezi, bulut ve yapay zekâ altyapılarında en az 10 milyar dolarlık özel sektör ağırlıklı kaynağı harekete geçireceğiz" (at least 10 billion US dollars, predominantly private sector, to be mobilised). This is not a government fund; the announcement describes it as predominantly private sector resources.
• Open data: "En az 2 bin kamu veri setini Ulusal Veri Kütüphanesi üzerinden milletimizin istifadesine sunacağız" (at least 2,000 public datasets opened through the National Data Library).
• Human capital: "10 bin ileri düzey yapay zekâ uzmanı ve 100 bin yapay zekâ uygulama profesyoneli yetiştireceğiz" (10,000 advanced AI specialists and 100,000 AI application professionals), plus "2 yılda 5 milyon vatandaşımıza eğitim vereceğiz" (5 million citizens trained within two years).
• Governance: the announcement states "Ulusal Yapay Zekâ Kurulu ise bu sürecin yönetişim zeminini teşkil edecektir" (the National AI Council will form the governance basis of this process).

At least 2 percent of public investment to AI: what the clause means in a budget

The commitment to allocate at least 2 percent of public investment programmes to AI projects raises two budget questions for an agency: what base the share is calculated against, and which spending lines count toward it. In the Presidency Directorate of Communications announcement dated 13 June 2026 the commitment reads "Kamu yatırım programlarımızdan yapay zekâ projelerine en az yüzde 2 pay ayıracağız"; the announcement text does not describe a calculation method.

As of 13 August 2026, when this article was prepared, no secondary regulation defining the calculation base, the classification criteria or the reporting calendar for the 2 percent share could be verified in publicly available sources. The most durable preparation an agency can make today is therefore not to guess the method but to tag its existing and planned AI-related spending so that it is traceable. When the implementing rules arrive, an agency with a ready inventory is at least one budget cycle ahead of one starting from zero.

In practice, AI-related items sit scattered across an agency's investment files. Some appear under hardware, some under service procurement, some under staff training. The list below shows the typical lines to sweep when preparing a funding case. Final classification is a decision for the agency's strategy development and financial services units.

A practical note on terminology: in Turkish public procurement these engagements are usually written as "yapay zeka yazılımı kiralama" (AI software rental) or "yapay zeka hizmet alımı" (AI service procurement); the terms technical teams use, such as API, token and model call, rarely appear in tender documents. Defining both vocabularies against each other inside the funding case removes the interpretation gap between procurement and IT from the start.

• Hardware: accelerator cards, server and storage investment, data centre capacity expansion
• Hosting and infrastructure: cloud resources, private cloud, on-premise hosting costs
• Software and services: AI software rental, model access services, API consumption charges, licence renewals
• Data: data preparation, cleaning, labelling, data quality and data governance work
• Integration: connecting to existing enterprise applications, authentication, logging and monitoring work
• People: staff training, certification, technical consultancy and project management support
• Evaluation: pilot deployments, comparative benchmarking and acceptance criteria validation

The plan's four axes, the agency equivalent, and the first step

The Türkiye Artificial Intelligence Action Plan rests on four axes: fark et (detect), istifade et (benefit), üret (produce) and yönet (govern). These four axes offer a natural skeleton for an agency-level action plan, because when the agency's own document speaks the national plan's language, mapping it into correspondence and budget requests becomes straightforward.

The table below connects the plan's axes to agency-level equivalents and a first concrete step. The axis names come from the announcement text; the agency equivalents and first steps are interpretations agencies can follow in practice, not obligations stated in the plan.

No AI-specific law is yet in force: how to proceed under regulatory uncertainty

Türkiye has no single AI-specific law in force. Three separate legislative proposals sit in the TBMM (Grand National Assembly of Türkiye) process, and all three remain at proposal stage: a standalone AI law proposal dated 25 June 2024, a proposal dated 10 November 2025 amending existing laws, and a proposal dated 3 December 2025 focused on Law No. 5651 (Internet Law). None has been enacted and none is yet in force.

The TBMM Yapay Zekâ Araştırma Komisyonu (Parliamentary Research Commission on Artificial Intelligence) published its report on 30 March 2026. Among urgent measures, the report recommends establishing an independent AI coordination and oversight body and becoming a party to the Council of Europe Framework Convention on Artificial Intelligence. The report is an advisory document; it does not by itself create binding obligations. Separately, the plan announcement of 13 June 2026 commits to "kullanıcıların haklarını koruyan ve yatırımcılara öngörülebilirlik sağlayan bir düzenleyici çerçeve oluşturacağız" (a regulatory framework protecting user rights and giving investors predictability) and refers to the Ulusal Yapay Zekâ Kurulu (National AI Council) as the governance basis.

For an agency this means the following: waiting for regulation before starting an AI project is costly against the plan's 2 percent commitment and the TÜBİTAK call calendar, yet locking into a single provider without accounting for the likely direction of regulation is also a risk. The reasonable path between the two is to start the project but build the architecture so it can be changed.

Portable architecture is a risk management decision here, not a marketing point. If new regulation introduces obligations around data residency, auditability or specific use categories, being able to change the model or the hosting location without rewriting the system is direct budget protection. Technically this means the application talks over a standard, widely implemented API surface, and the model identifier is a value held in a configuration file.

This content is informational and does not constitute legal advice. Final assessment rests with the agency's own compliance and legal units.

• Put a data residency clause in the contract, stating where data is processed and stored; do not settle for verbal assurance
• Define an exit clause: retrieval of agency data in machine-readable form and deletion on the provider side when the contract ends
• Write in the right to change models: switching to a different model under the same contract without a new tender
• Require usage reporting: consumption broken down by unit, date and model, in a format the financial services unit can verify
• Require subcontractor and sub-processor transparency: which part of the service is delivered by which party must be documented
• Write acceptance criteria numerically: accuracy, response time and availability thresholds must be measurable

TÜBİTAK Public AI Ecosystem 2026 Call: support through the 1007 programme

TÜBİTAK opened the Public Artificial Intelligence Ecosystem 2026 Call under the 1007 Public Institutions Research and Development Projects Support Programme. The client institution for the call is the Presidency Cybersecurity Directorate (Cumhurbaşkanlığı Siber Güvenlik Başkanlığı), and the call budget is set at 15,000,000 TL. The call calendar includes 2 April 2026 for needs submissions from public institutions and 6 August 2026 for project proposals from executing organisations.

The programme structure carries an important distinction for agency budgets: in the 1007 model, the public institution that defines the need and the organisation that executes the project are different parties. Private sector entities active in AI or public R&D units serve as project executing organisations, and universities may participate as project executing organisations depending on the technology required. This means an agency can route an AI need into an R&D channel without drawing on its own investment budget.

The call covers six priority thematic areas. If an agency's need overlaps with one of them, the funding case can present both the investment programme share and the R&D support channel together; this dual presentation prevents the budget request from appearing dependent on a single source.

Describing the support channel correctly inside the funding case matters. TÜBİTAK support does not replace the share an agency allocates to AI from its investment programme; it is typically used alongside that share, for a need at a different maturity level. An operational and repeatable service need is met through service procurement, while a need that is predominantly research in character is met through the R&D channel.

• Financial technologies
• Intelligent production systems
• Smart agriculture, food and livestock
• Climate change and sustainability
• E-commerce technologies
• Smart education technologies

How LLMTR answers this budget and uncertainty picture

LLMTR is an access layer for teams that need AI spending to be traceable and the architecture built without locking into a single provider. It maps onto the three concrete needs in this article at three points: making the 2 percent share reportable, being able to change models under regulatory uncertainty, and having a Türkiye-hosted option available behind the same interface.

On budget reporting, an agency can issue a separate API key per unit; each key carries its own rate limit, spending ceiling and usage report. That lets the financial services unit answer the question of which unit spent how much on which model in which month from a single statement. Credit top-ups carry an 8% platform margin; no margin is added on top of model prices, and the unit prices in the catalogue are provider prices.

On portability, LLMTR exposes an OpenAI-compatible /v1 surface. On the application side, migration amounts to changing the base URL and the model identifier; business logic, prompt templates and integration code stay in place. When regulation changes or the agency decides to move to a different model, that is the extent of the surface that changes.

On data residency, the catalogue includes models hosted in Türkiye; llmtr/gemma-4, llmtr/qwen3-6-35b and llmtr/trendyol-7b are among them. Because models from global providers are callable with the same key, an agency can route sensitive workloads to a Türkiye-hosted model and non-sensitive workloads to whichever model is cost-appropriate. User prompts and model response bodies are not written to the usage and billing database; customer API keys are stored as SHA-256 hashes, never in plain text.

LLMTR provides no compliance guarantee and makes no claim of official approval, accreditation or certification. Assessing regulatory conformity remains the responsibility of the agency's own legal and compliance units.

Migrating an existing OpenAI client: only the base URL and model identifier change

import os

from openai import OpenAI

client = OpenAI(
    base_url="https://llmtr.com/v1",
    api_key=os.environ["LLMTR_API_KEY"],
)

# The model id comes from configuration; the model can change without code changes.
MODEL = os.environ.get("AGENCY_MODEL", "llmtr/gemma-4")

response = client.chat.completions.create(
    model=MODEL,
    messages=[
        {"role": "system", "content": "You summarise internal agency documents."},
        {"role": "user", "content": "Summarise the attached investment justification report."},
    ],
)

print(response.choices[0].message.content)

Sources

The legislative and policy references in this article were verified against the primary sources below. Last checked: 13 August 2026.

This content is informational and does not constitute legal advice. Final assessment rests with the agency's own compliance and legal units.

• President Erdoğan announces the Türkiye Artificial Intelligence Action Plan, 13 June 2026 — iletisim.gov.tr
• Public Artificial Intelligence Ecosystem 2026 Call, 1007 programme, call budget 15,000,000 TL — tubitak.gov.tr
• Presidential Circular No. 2021/18 on the Ulusal Yapay Zekâ Stratejisi 2021-2025, Resmî Gazete dated 20 August 2021, issue 31574 — resmigazete.gov.tr
• Ulusal Yapay Zekâ Stratejisi 2021-2025 document page — cbddo.gov.tr
• TBMM Yapay Zekâ Araştırma Komisyonu Raporu (Parliamentary AI Research Commission report), 30 March 2026 — tbmm.gov.tr
• Analysis of the Türkiye Artificial Intelligence Action Plan 2026-2030 — setav.org

Preparing an internal AI budget justification in a public agency

Steps for building an internal funding case for an AI project, grounded in the 2 percent commitment of the Türkiye Artificial Intelligence Action Plan 2026-2030.

1. Document the policy ground. Place the 2 percent commitment and the statement that the public sector is the first buyer of domestic AI solutions, both from the Presidency Directorate of Communications announcement dated 13 June 2026, on the first page of the funding case with their source. Do not include figures you cannot verify.
2. Inventory existing AI-related spending. Sweep the last two years of investment and operating expenditure and list the hardware, hosting, software rental, data preparation, integration and training items that can be associated with AI. This lets you show numerically what percentage of the investment budget the agency currently directs to this area.
3. Choose one narrowly scoped pilot with numeric acceptance criteria. From the per-unit use case inventory, pick the highest-volume and most repetitive workload. Write at least three measurable before-and-after criteria for the pilot: time per transaction, accuracy rate and unit cost. Do not launch a pilot without them; a pilot without criteria produces no evidence for a budget request.
4. Model the cost in both vocabularies. Calculate expected consumption both in the technical team's terms (monthly request count, average input and output length, model unit price) and in the procurement unit's terms (an AI software rental or service procurement line, estimated annual value, ceiling amount). Show that both calculations produce the same figure.
5. Add support channels to the justification. If your need overlaps with one of the six thematic areas of the TÜBİTAK Public AI Ecosystem 2026 Call, present the R&D channel as a second source alongside the investment programme share. Match operational and repeatable needs to service procurement, and predominantly research-oriented needs to the R&D channel.
6. Write the risk and exit plan. State plainly in the funding case that no AI-specific law is yet in force, that three legislative proposals remain at proposal stage, and that a regulatory framework commitment appears in the plan. State the mitigation too: a portable API surface, exit and data portability clauses in the contract, the right to change models, and per-unit usage reporting. Submit the file for the opinion of the agency's legal and compliance units.

Frequently asked questions

When was the Türkiye Artificial Intelligence Action Plan 2026-2030 announced and who prepared it?

The Türkiye Yapay Zekâ Eylem Planı 2026-2030 (Türkiye Artificial Intelligence Action Plan 2026-2030) was announced publicly on 13 June 2026 at the Türkiye Artificial Intelligence Summit. It was prepared by the Sanayi ve Teknoloji Bakanlığı (Ministry of Industry and Technology) and announced the same day by the Presidency Directorate of Communications. The plan rests on four axes: fark et (detect), istifade et (benefit), üret (produce) and yönet (govern).

What does allocating 2 percent of public investment to AI actually mean?

In the Presidency Directorate of Communications announcement dated 13 June 2026 the commitment reads "Kamu yatırım programlarımızdan yapay zekâ projelerine en az yüzde 2 pay ayıracağız" (we will allocate at least 2 percent of our public investment programmes to AI projects). It means agency budgets will carry a minimum share for AI projects. As of 13 August 2026, no secondary regulation defining the calculation base or reporting method for that share could be verified in publicly available sources.

Does Türkiye have an AI-specific law?

No. Türkiye has no single AI-specific law in force. Three legislative proposals sit in the TBMM process and all three remain at proposal stage; none has been enacted and none is yet in force. In its report dated 30 March 2026, the TBMM Yapay Zekâ Araştırma Komisyonu (Parliamentary AI Research Commission) recommended establishing an independent AI coordination and oversight body and becoming a party to the Council of Europe Framework Convention on Artificial Intelligence; the report is advisory in nature.

Is an agency required to prepare its own AI action plan?

As of 13 August 2026, no regulation requiring agency-level action plans could be verified. Even so, preparing an agency document mapped to the national plan's four axes gives a practical advantage: when budget requests and internal correspondence use the national plan's language, mapping and justification become easier.

Who can apply to the TÜBİTAK Public AI Ecosystem 2026 Call?

The call was opened under the TÜBİTAK 1007 Public Institutions Research and Development Projects Support Programme, with the Presidency Cybersecurity Directorate as client institution. Private sector entities active in AI or public R&D units serve as project executing organisations; universities may participate as project executing organisations depending on the technology required. The call budget is 15,000,000 TL.

Is it risky to start an AI project while regulation is pending?

The risk is determined not by starting the project but by how the architecture is built. If the application talks over a standard, widely implemented API surface and the model identifier is held in configuration, the model or hosting location can be changed when new regulation arrives. In an application embedded in one provider's proprietary interface, the same change carries redevelopment cost. Adding exit, data portability and model-change clauses to the contract covers the same risk at contract level.

Related posts

• Planning an AI pilot budget within Turkey's 2026 direct procurement limit
• Are Turkey's domestic LLMs enough? BILGE, KUMRU, MAIN and multi-model strategy

• Pricing
• Usage tracking
• API keys
• Model catalog