Türkiye has launched EVREN, a national artificial intelligence platform developed within its Presidency of Defence Industries (SSB) to bring data, models and high-performance computing together on infrastructure located in the country. The launch turns a defense-sector initiative into a service for companies, universities, entrepreneurs and students who can authenticate through the government's e-Devlet system.

The official portal describes EVREN as an end-to-end platform. Users can manage and version datasets, label text, audio and images, train models across distributed infrastructure, deploy inference and track experiments. It also offers SAM-2-assisted pre-labeling and YOLO-based object-detection tools. EVREN is not a new Turkish foundation model: it is an infrastructure and services layer that brings together open models, data and computing resources.

The published hardware specifications are specific. EVREN says it operates a cluster of 64 NVIDIA H200 GPUs connected by 400 Gbps InfiniBand NDR for training. For visual inference, it has eight A6000 GPUs and offers camera, image and video processing. Its language-inference interface provides an OpenAI-compatible API and, at launch, lists 12 open models online for conversation, coding, reasoning and visual understanding.

The location of the infrastructure is central to the proposal. According to the official information and Anadolu Agency, prompts, responses and usage metrics are processed on high-performance servers located in Türkiye. The stated aim is to keep sensitive data from being sent to foreign cloud services. This describes the architecture announced by the operator; it is not, on its own, an independent audit of security, isolation or data sovereignty.

EVREN uses a credit system rather than direct paid access. Participants can earn credits by contributing datasets, labeling information or sharing trained models, then use those credits for GPU hours. The website lists enterprise controls including single sign-on, role-based access, data isolation and audit logs. It also says API calls will not incur credit deductions until November 1, 2026.

Anadolu Agency reported that the platform had reached 7,500 active users. At the same time, the public site displayed some general counters at zero while other figures — including 163.1 million tokens generated and 12 models online — showed values. That discrepancy may reflect how the site's metrics load, but the user figure should be treated as a claim attributed to the SSB rather than an independently verified count.

The immediate scope is shared infrastructure, not a demonstrated operational military deployment. The SSB presents EVREN as a foundation for expanding the national AI ecosystem and anticipates adding locally developed language, image and audio models, as well as distributed GPU management. The announcement did not include comparative evaluations, details about sensitive datasets or evidence of use in weapons systems.

The significance of the launch lies in its combination of sovereign computing, access through a state identity system and a contribution mechanism for building data and models. For the defense sector, that design could reduce dependence on foreign cloud services and enable controlled experimentation. The decisive test will be whether the platform can sustain capacity, security controls and data quality as its user base and projects grow — and whether it publishes verifiable results beyond its initial specifications.

Sources:
- Official EVREN platform (SSB): EVREN (SSB)
- Anadolu Agency, published by Daily Sabah: Daily Sabah