$10 Billion of Investment, 1 GW of Power
Two numbers, one sentence: ten billion dollars of investment, one gigawatt of power. Türkiye’s AI Action Plan targets raising the country’s AI and data-centre installed capacity to at least 1 GW by 2030, mobilising 10 billion US dollars of predominantly private investment to get there. The supporting cast of targets is not small either: 10 TWh of low-carbon power purchase agreements a year, an exabyte of storage, 100 Tb/s of backbone capacity — and a contractual floor of 60 percent low-carbon electricity.
Struggling to picture a gigawatt? Try this: take the entire output of a nuclear reactor and dedicate it to machines that do nothing but compute. That is the scale of the equation. And the equation’s most interesting line is who the power opens to — not just the giants. You.
What the Number Says
BU BÖLÜMÜN ÖZETİ
- Compute has been declared a national utility
- The door says “everyone” on it
- The energy clause defines the investment’s character
The infrastructure target reads in three layers.
Compute has been declared a national utility
The plan treats computing capacity the way states treat electricity and roads: a Compute Portfolio will pool public reserve capacity, domestic data centres and international cloud partnerships under one roof — and even public institutions are directed to use the shared pool before buying their own hardware. The message is unambiguous: GPU is no longer a luxury good but a grid utility, and the connection rules are being written now.
The door says “everyone” on it
The target’s business-facing side is the GPU-for-All programme: startups, SMEs and researchers will access world-class compute through GPU-hour credits. The first phase targets 2 million GPU-hours a year, rising to 20 million annually by period’s end — with at least 40 percent of credits reserved for tech startups. One of the gigawatt plant’s sockets, in other words, is being wired to the desk of whichever small business files the right application.
The energy clause defines the investment’s character
The 60-percent low-carbon requirement is not an environmental gesture; it is an export strategy. Global clients and funds are cooling on models trained in carbon-heavy infrastructure, so Türkiye writes the “clean compute” label into the investment from day one. For any business paying cloud bills, the measuring stick will drift the same way: where you compute will be discussed alongside what you compute.
What the Headline Misses
BU BÖLÜMÜN ÖZETİ
- The investor is not the state
- GPU-hours are arriving as a budget line
- Access has a queue, and the queue has an order
The gigawatt headline shades three details.
The investor is not the state
The 10 billion dollars carries a carefully chosen adjective: “predominantly private.” The state prepares land, energy and permits — AI Growth Zones, accelerated licensing, a single-window office; the money comes mostly from investors risking their own capital. That is a good sign for the target’s seriousness: an investor spending his own money does not build a facility to leave it empty. It also matters for dependency management — as capacity diversifies, captivity to any single provider shrinks.
GPU-hours are arriving as a budget line
The plan measures compute in GPU-hours and distributes credit in the same unit. Most businesses have never heard the term; within a few years it will enter grant applications, project budgets and possibly accounting entries. You read your electricity bill in kilowatt-hours; you will read your AI bill in GPU-hours — and whoever learns the unit early, learns the negotiation early.
Access has a queue, and the queue has an order
Capacity opens to everyone — through accreditation criteria, access windows and allocation quotas: an orderly queue. Its front belongs to those with defined projects and ready data. “Let’s try some AI” gets no credit; “we will train this model on this data to automate this decision” does. The queue admits projects, not curiosity.
What a Business Should Do
BU BÖLÜMÜN ÖZETİ
- Write your compute needs down today
- Reread your cloud contract for portability
- Put the access programmes on a calendar
The number translates into three moves.
Write your compute needs down today
A one-page inventory: which of your work needs model training, which runs inference on ready models, roughly how many hours a month would you consume? If that page does not exist today, neither does your application when the credit call opens. The business with written needs catches the programme’s first wave; the one without waits in the second wave’s crowd.
Reread your cloud contract for portability
As national capacity comes online, options multiply — and when options multiply, bargaining power flows to whoever can move. What does your current contract say about data egress costs, formats and exit terms? Half a day of reading today makes tomorrow’s cheaper capacity actually usable — this is precisely the infrastructure chapter of the rent-versus-title distinction.
Put the access programmes on a calendar
GPU-for-All calls and growth-zone announcements will arrive in waves across the period. Assign the watch to a name: once a quarter, the call pages get checked and the standing file gets refreshed for anything that fits. Opportunity tracking is not a matter of inspiration; it is a matter of calendar.
The gigawatt is the state’s job; how many hours you draw from it is yours — and the hours flow to the ready project.