Turkish Airlines runs more than 60 AI models built in house
Turkish Airlines says it has more than 60 AI models live, built in house and run on its own servers with Red Hat's OpenShift AI, covering pricing, fraud prevention, customer chatbots, aircraft assignment and operations forecasting. At least 40 more are in development. The airline has published no results for individual models.
The operating picture
AI kept on the airline's own servers
Turkish Airlines, Airbus A330-300, TC-JOF - NRT. Photograph: Masakatsu Ukon / CC BY-SA 2.0 (cropped to fit). Illustrates the airline, not an AI deployment.
Turkish Airlines flies to 353 destinations with 479 aircraft, according to Red Hat's May 2025 announcement. It has built its AI development platform on Red Hat OpenShift AI, which keeps the airline's data on its own premises.
More than 200 employees work on AI development, and the airline says deployment speed has doubled on the platform. The models span the commercial side, fraud checks on card payments and Miles&Smiles loyalty activity, and the operation itself.
Selected implementations
The work behind the evidence
01 / Deployed
Matching aircraft to routes and forecasting the day's delays
The operational problem
Which aircraft flies which route affects fuel burn, and delays are cheaper to manage when the operations team sees them coming.
What the evidence reports
Red Hat said on 20 May 2025 that Turkish Airlines runs more than 60 live models on OpenShift AI, with at least 40 more in development. They include aircraft assignment models that weigh routes, aircraft details and fuel consumption; operations models that predict ground times, block times and punctuality; dynamic pricing; fraud screening; and customer service chatbots built on open source large language models. Serdar Gürbüz, general manager of Turkish Technology, said the platform lets the airline "keep data on-premises while accelerating model development".
Operational significance
Building in house on its own infrastructure gives the airline control of its data and models. The assignment and delay models are aimed at operating decisions, where their effect can be measured in fuel and punctuality.
Evidence still needed
The published figures describe the platform, not outcomes. No fuel saving, punctuality change or pricing gain has been published for any model, and the announcement does not say how operations teams use the forecasts.
Other records are shown separately, including non-AI technology and programmes at different stages of deployment. A register entry does not establish a measured AI benefit.
No other publicly listed records.
Editorial assessment
Turkish Airlines has built its AI in house and at scale, on its own servers and with its own development team.
The next useful evidence is outcome data for the operational models: fuel saved through aircraft assignment, and delays avoided through earlier warning.
Live data from the Aerostack register. Status reflects published evidence only. See Our approach for how records are graded.
Aerostack register · 1 tracked deployment
Aerostack tracks AI deployments across airports and airlines, with the published evidence behind each record. Photography is real aviation imagery. Record thumbnails may show the named airport or airline; other photographs do not.
