Summary

• Deutsche Telekom targets about €2.5 billion of indirect cost savings from AI and automation by 2030 compared with 2023.
• Non-US AI revenue from business customers is expected to rise from about €250 million in 2026 to €800 million by 2030.
• The group is applying AI across networks, customer service, software development and administration while reinvesting some savings in fibre deployment.

Deutsche Telekom has attached unusually specific financial targets to its artificial intelligence programme, expecting AI and automation to cut indirect costs by about €2.5 billion by 2030 while increasing non-US enterprise AI revenue to roughly €800 million.

Deutsche Telekom presented the targets at an AI investor day in Bonn, placing revenue, cost and service metrics around technology that large companies have often discussed through experimentation and productivity potential rather than group level financial outcomes.

The operator expects around €1.1 billion of gross savings outside the US by 2027 compared with its 2023 cost base, rising to approximately €2.5 billion of indirect cost savings by 2030. Some of the earlier gains will be reinvested in digital transformation and Germany’s fibre network rather than flowing directly into profit.

AI related revenue from business customers outside the US is expected to reach about €250 million in 2026 and grow to approximately €800 million by 2030. Telekom is building services for large organisations alongside a platform intended to help smaller companies automate recurring work in areas including customer service and logistics.

The figures provide a clearer test for enterprise AI than the number of pilots, licences or employees with access to assistants because Telekom is asking investors to judge the technology by whether it alters operating economics and produces services customers will pay for.

Automation has to show up in operating work

Much of the cost case comes from processes where Telekom already handles large volumes of repeatable activity. Network operations, customer service, software development and administration all contain work that can be standardised sufficiently for automation while employees remain responsible for exceptions and final decisions.

The company says its Frag Magenta chatbot handled about 2.6 million customer service calls during the first half of 2026. At T-Mobile US, AI agents now handle 40% of customer contacts and the number of customer service calls has fallen by 55%.

Network operations provide another measurable example because Telekom’s RAN Guardian Agent monitors the mobile network for emerging capacity problems, including unusual demand around major events, and assists teams in responding. The company says reaction time has fallen from several hours to around one minute.

Customer service tools meanwhile provide employees with relevant information during calls and generate documentation afterwards. Early deployment around new connections has produced a reported 30% reduction in complaints, according to Telekom.

Those figures do not establish how much of the improvement can be attributed to AI alone, and the 2030 financial numbers remain management targets rather than realised savings. They nevertheless create a basis on which deployment can eventually be judged.

Sovereign AI becomes a commercial product

Part of the revenue ambition rests on Telekom turning European demand for greater control over infrastructure and data into a commercial AI proposition rather than leaving sovereignty as a compliance characteristic.

The company is combining its networks and European infrastructure with technology from external partners so businesses and public institutions can operate sensitive workloads with greater control over where systems run and how data is handled. Its Industrial AI Cloud in Munich forms one part of that strategy.

Capacity at the Munich AI cloud had already filled rapidly earlier this year, while Telekom is also considering participation in Europe’s planned AI Gigafactory programme. The latest targets attach a revenue expectation to that infrastructure rather than presenting European control as an end in itself.

Europe is spending heavily on sovereign compute without any guarantee that enterprises will shift enough workloads onto it to justify the investment. Providers therefore need applications, managed services and credible migration paths around the infrastructure if local control is to become a sustainable business rather than mainly a policy objective.

Telekom’s SME platform extends the proposition to companies that may lack the engineering teams required to assemble models, orchestration tools and governance systems themselves, making packaged automation one possible route to broader adoption.

Savings do not translate directly into headcount

Telekom presents automation as a redesign of work in which employees set objectives, check results and remain accountable for decisions. More than 100,000 employees have received AI training through its AI for All programme, while staff have access to tools including AskT, ChatGPT Enterprise and Microsoft Copilot.

The economic outcome will depend on whether productivity improvements reduce the resources needed for existing work, allow employees to handle greater volumes or free capacity for other activities.

Headline savings can therefore overstate the amount that ultimately reaches profit because automation programmes also require spending on technology, training, integration and new infrastructure. Telekom explicitly expects some gains to support faster fibre deployment, showing how operational efficiency can be recycled into capital intensive investment.

There is also a considerable step between a successful individual application and group wide financial impact. A network agent that responds faster or a customer service tool that reduces documentation effort can demonstrate technical value without automatically scaling across business units, countries and legacy systems.

Telekom acknowledges that its projections are forward looking and exposed to technological, competitive and operational risks. The next four years will test whether applications that work inside particular teams can be standardised sufficiently to shift consolidated cost and revenue figures.

Enterprise AI is therefore entering a more demanding phase of adoption. The technology no longer has to demonstrate that it can answer questions, write software or assist employees; it has to survive integration with operating processes and become visible in the numbers companies already use to measure performance.