Türkiye’s AI Action Plan, Explained for Business

What Is on the Agenda?

What landed on the table is not a statement of intent but a budgeted, scheduled, monitorable implementation plan. Four traits set it apart.

Bound by circular, watched by portal

The plan was written into every public institution’s duties by a circular published in the Official Gazette; its progress will be published action by action on a public Transparency Portal — responsible institution, budget realisation and a green-amber-red status light for each. The promise’s report card is open to the public: unusual for policy documents anywhere, and the first exhibit in the plan’s claim to seriousness.

Targets tied to numbers, numbers tied to dates

The document carries more figures than adjectives: a 1-trillion-lira value target, 10,000 advanced experts plus 100,000 applied professionals, AI literacy for 5 million citizens, 20 million GPU-hours of annual allocation, 2,000 public datasets, at least 1,000 SME vouchers. Every action closes with an italic “Our target” paragraph, most with 12-24-month interim thresholds. Numbers and dates are accountability’s raw material — and this document produces it in bulk.

The money is architected as a public-private blend

Funding is not expected from one pocket: public budget, private investment, international funds and public-private partnerships together, with each action’s annual plan naming its source. The largest line — the infrastructure build — is explicitly labelled “predominantly private”; the state prepares land, energy, permits and demand. The role the state chose is enabler, not doer — an allocation that writes the address of value creation into the market from page one.

The public sector committed to being first customer

The plan’s most concrete business-facing line is a purchasing commitment: at least 2 percent of public investment programmes for AI projects, a digital badge for solutions that succeed, and a fast procurement route through the State Supply Office catalogue for badge holders. A public-demand answer to the oldest trap in technology entrepreneurship — no sale without a reference, no reference without a sale. For producers, this is the plan’s most valuable page.

Why Is the Question Being Asked Now?

The plan was not born in a vacuum; it was born at the crossing of four forces.

The first five years’ lessons were on the table

This is a second act, not an opening: it builds on the 2021-2025 national strategy, a parliamentary commission report and policy-board documents, drafted through workshops with some five hundred professionals and over two thousand citizen proposals. The first period’s scattered gains — curriculum, research infrastructure, an AI institute, incentive programmes — are being bolted to a single spine. A second attempt with experience deserves a different reading than a first attempt with hope.

The global race raised the price of indifference

Data dominance and compute advantage are widening the gap between economies; major powers are publishing national plans in sequence. Türkiye’s plan answers that race — and for a business the real message is in the race itself: in a field where states pay billions, “wait and see” means arriving late at a table where someone else set the prices.

Tool abundance created a demand for order

AI has stopped being an access problem and become an order problem: everyone holds tools; few hold sequence, measurement and rules. The plan’s axis architecture — a maturity ladder — is a line drawn through exactly that chaos: awareness and data first, use next, production after, governance throughout. The state wrote itself a sequence; it recommends the same to every organisation.

Sovereignty became an economic category

The plan’s spine principle is digital sovereignty: the country that protects its data, develops its compute and can build its own models sits at the decision table. The principle is not ideological decoration; it is dependency cost calculated at state scale — and it has a business-sized counterpart: cleanly titled data, tools with exit doors, diversified access. The state builds it with gigawatts; a business builds it with rules.

What Is Wrong?

Public conversation about the plan circles four maxims, each pulling the reading off its axis.

“It will stay on paper” — wholesale distrust

Fed by past disappointments, this reflex misses two things: the monitoring architecture (an open portal, an annual report with an independent component) and the first period’s actually completed work. The healthy stance is neither faith nor denial; it is this: treat the lights as signal and the outcomes as proof — did the call open, was the credit allocated, did the catalogue publish? Wholesale distrust’s real cost is commercial, not emotional: the business that sulks walks past the doors as they open.

“A game for giants” — the scale fallacy

Gigawatt and billion headlines dress the plan as a hyperscaler’s document. The text says otherwise: separate SME access windows in GPU-for-All, at least 40 percent of credits reserved for tech startups, an SME voucher programme, small-and-medium contract priority in pilots. The plan did not merely leave a door open for small business; it cut the door to small-business size. Not entering will be the visitor’s choice, not the door’s.

“Our job is to wait” — the passive reading

“The state builds, we use” sounds reasonable and contains two errors. Timing first: being ready when programmes open requires data order, capability and files today — who does not prepare while the state builds cannot apply when the state opens. Scope second: half the plan describes homework only a business can do for itself — it is a frame to accompany the doer, not a package to await.

“AI is global; national plans are noise” — misplaced cosmopolitanism

Models may be global; data regimes, procurement rules, incentive windows and sector regulation are national — and the layer that touches a business’s daily life is precisely the second one. Using global tools while ignoring national rules is sailing international waters without reading harbour law: free on the open sea, fined at every berth.

The Real Mechanism

The plan’s machine has four parts; they run interlocked, not side by side.

Part 1: the four-axis spine

Be aware (awareness, talent, data, rules) → Utilise (infrastructure, public and sector adoption) → Produce (funding, models, physical AI) → Govern (investment attraction, diplomacy, safety). Four actions per axis; sixteen in all. The axes divide the labour and sequence it at once: each lays the next one’s floor. The sequence the state wrote for its own climb doubles as the reader’s key: which action speaks to which maturity is visible from the axis it sits on.

Part 2: five layers, three crossing components

The plan’s architectural diagram is a building, not a pie: energy at the base, compute hardware above it, then infrastructure, then models-and-data, applications at the top; talent, funding and regulation-and-trust are three columns cutting through every floor. The diagram’s message is structural: weakness on any floor constrains everything above it. The business translation is identical — an application floor built over a disordered data floor is a roof without foundations.

Part 3: governance with four functions

Decision and coordination sit with a National AI Board chaired at the highest level; implementation tracking with a Programme Office; ethical guidance with a multi-stakeholder Ethics Board; technical security evaluation under cybersecurity coordination with national research-institute capacity. The Board will publish an annual Policy Letter — a direction signal to investors and the market. What matters to a business is not the organs’ names but their outputs: letter, report, portal — the three items of a three-channel watch list.

Part 4: the scan-pilot-scale loop

The plan’s operational heartbeat is a three-stroke cycle: scan use cases systematically, launch fast pilots, scale what passes the threshold. Every ministry will run the loop; projects clearing the bar scale within 6-12 months. The loop meshes with the phased calendar — build and pilot in the early years, scale and export later, with early maturers jumping the queue. It is also the plan’s implicit advice to companies: the road from demo to production runs through the same three strokes at every scale.

Who Is Affected, and How?

The same plan reads like four different documents at four tables.

The micro business

The direct shelf is short but valuable: free literacy programmes, a national competency assessment, SME vouchers and ready-made solution packages. The plan’s real contribution to this profile is psychological — it pulls the floor from under “this is beyond our size” with voucher and package mechanisms. The first step costs nothing: one learning hour, one competency check.

The growing SME

The widest shelf sits here: GPU credits, sector pilot calls, the data library, prototyping at regional excellence centres. The homework is also clearest here: every one of these doors opens to the business with a defined project — the queue admits files, not curiosity. The SME’s period strategy is one sentence: finish the preparation phase on your own resources; accelerate the scale phase on public ones.

The technology producer and startup

The plan’s most generous reader: a two-fund financing ladder (research plus growth), the procurement window, the badge-catalogue-export chain, growth-zone infrastructure. The thresholds are equally explicit: a tech-startup certificate, model cards, impact assessments, security testing. For producers the period is one of aligning the product roadmap with public specification language — the auditable, portable product is tomorrow’s tender dialect. For international firms, a parallel shelf: growth-zone investment packages, cloud partnership frameworks and mutual-recognition mechanisms are written for cross-border players.

The agency, consultant and integrator

An invisible but large market: literacy for five million people, certification for a hundred thousand professionals, a roadmap for every ministry, voucher projects for thousands of SMEs — all of it training, consulting and implementation work. The winner will be sorted by a familiar line: the hour-seller melts in voucher budgets; the partner who moves clients up the ladder becomes the period’s natural ally.

Decision Order

A business’s response compresses into four steps; the order is deliberate — each step feeds the next.

One: read the document through a business lens

A ninety-minute sitting: convert the four axes into four questions, hunt each action’s target and access sentences, leave with a three-column note page — direct, indirect, far. The reading guide covers the whole method. Without this page, the next three steps float.

Two: locate your rung

The four rung tests: does the team know the tools, are gains measured, does the customer notice, are the rules written? The first “no” is your rung — and the rung tells you which programmes speak to you today and which tomorrow. Reading the plan without a rung is ordering a shirt without a measurement.

Three: mesh the two calendars

Against the state’s build and scale phases, sort your own work onto three shelves: NOW (prerequisites that wait for no call — data order, learning, written rules), TIME-IT (work public money will cheapen), LATER (work that rides the scale phase’s products). The sorting answers “why now / why not” permanently.

Four: install and connect the watch

Three official channels — portal, Policy Letter, annual report — watched with three quarterly questions, under one named owner, reporting one line to the management agenda. The routine keeps the first three steps’ map alive: a green light shifts a shelf, an open call pulls a file. After setup the yearly cost is two hours; like unused data, an unwatched plan lies fallow.

Where to Start?

The opening month, four finishable weeks.

Week one: the reading session and the note page

A ninety-minute calendar block; the output, one three-column page. A half-hour team share fits the same week — one person reads the plan, everyone knows it.

Week two: the rung test and one target

The four tests run with the team, the first “no” is found, a one-rung target gets written: a sentence, a date, an owner. That is the whole week — and more clarity than most businesses have produced in years of “we should look into AI.”

Week three: the shelves and the first NOW job

Planned AI work goes onto three shelves; the first NOW item — typically a data inventory or the weekly learning hour — actually starts. So the shelf does not stay paper, the first job’s start date is written the same day as the shelf.

Week four: the watch setup and the file skeleton

Three-channel list, quarterly questions, owner — one hour. The rest of the week builds the application skeleton: company profile, data summary, project template, and where relevant the tech-startup certificate application. The month ends holding not an opinion but a running arrangement.

What Not to Do?

The period’s four traps, in two symmetric pairs.

Believing the plan — or dismissing it

The two extremes are the same laziness: trusting without thinking and rejecting without thinking. The plan is neither scripture nor scrap paper; it is a list of commitments to be monitored. The right distance is a cautious partnership that commits resources as results appear — as in any commercial relationship.

Mistaking subsidy for strategy

Public money subsidises work worth doing; it does not justify work. The project that begins with “a voucher came out, let’s do something” ends when the voucher does. The order never changes: the business need first, then the support that happens to match it — reversed, it is the price list of free cheese.

Deferring preparation to call day

Calls arrive in waves and close through short windows; whoever starts writing the file when the window opens either misses it or files poorly. Preparation — data, capability, documents — belongs to the calendar, not the call. And evaluators have long memories: a rushed bad application shadows the next good one.

Ignoring the compliance side

The plan carries rules as well as carrots: a proportionate-risk framework, impact assessments and model cards for high-impact systems, strengthened data protection. Every business processing customer data with AI today is already a neighbour of tomorrow’s rulebook. Whoever learns rules on publication learns them by fine; basic hygiene like human-approval flows gets built before regulation asks.

What to Watch?

The period’s pulse reads on four gauges.

Portal lights — especially your column

Your note page’s “direct” actions get a quarterly check: anything turned green, anything fallen red? Green is a file-refresh signal; red sends the dependent work back to your own budget.

The first progress report

The first public report, due within twelve months, is the plan’s character test: are slippages written honestly, are corrective steps concrete, is the independent component real? That report sets the trust coefficient for the remaining four years.

Call volume and allocation realisation

The field measure of a promise is a call: GPU credit allocations, voucher waves, pilot programmes, dataset publications. Count realisations, not announcements — how many businesses drew credit, how many pilots scaled. Outcomes are the proof behind the lights.

The annual Policy Letter

The Board’s yearly letter announces the coming year’s emphases — which axis moves forward, which sectors get priority. It is the annual revision input for your shelves: when the state’s wind changes direction, the first sail to turn gains the most.

The rule calendar’s maturing

The proportionate-risk guide, sector annexes, ethics-board guidance and sandbox calls — the plan’s rule side flows on its own calendar. For any business handling customer data, that flow deserves the same watch as the opportunity flow: a rule is cheap before publication and mandatory after.

How Does This Period End?

Three separations will be legible at the period’s close.

The gap between prepared and unprepared widens

Whether the plan’s targets land in full or in part, one group of businesses will exit this period with ordered data, capable teams and public references; the other with “we’d heard about it.” What opens the gap will not be the state but the routine — public money is wind, and wind only pushes an open sail.

The market’s language changes

GPU-hours, model cards, impact assessments, data spaces — terms that sound foreign today will settle into specifications and contracts by period’s end. The business that learns the language early negotiates without an interpreter; the late learner pays the interpreter’s commission.

The plan gives way to habit

In the good scenario, the plan’s ending is its own forgetting: the watch routine, the rung climb and the calendar mesh dissolve into a business’s ordinary annual rhythm — the way nobody now waits for a “budgeting movement” to draw up a budget. The document’s success will be measured in making itself unnecessary, not indispensable: a map goes into the pocket once the road is learned by heart.

A Solid Digital Foundation

Four stones turn the period to a business’s favour — each detailed in the cluster’s working pieces, each one page long.

The three-column note page

Direct, indirect, far — the plan mapped to the business. The ninety-minute reading’s durable output and every other stone’s input.

The rung map and the one-rung target

Four tests, one diagnosis, a one-sentence target. The business’s maturity photograph — retaken quarterly; the difference between photographs is climbing speed.

The three-shelf calendar

Now, time-it, later. The hinge connecting the public calendar to the business calendar — and the institutional antidote to “we’re waiting for the state.”

The three-channel watch list

Portal, letter, report; three quarterly questions; one owner. The period’s cheapest intelligence unit — two hours a year, and one correct application repays a decade of it.

Frequently Asked Questions