Deutsche Telekom told investors on 5 October that artificial intelligence and automation will take about €2.5B out of its indirect costs by 2030, measured against where those costs stood in 2023. The German operator set the figure out at an AI investor day in Bonn, alongside a firmer near-term number: savings outside the United States should reach €1.1B by 2027, up from the €800M it had guided before.
Most of that will not show up as profit. The company intends to put the additional 2027 savings back into accelerating its digital transformation and extending fibre in Germany, which is why its overall guidance did not move when the cost targets did. For a group expecting adjusted earnings before interest, tax, depreciation and amortisation of roughly €47.5B this year and free cash flow near €19.8B, a €2.5B reduction spread across seven years is meaningful without being transformative, and the company has been careful not to present it as more than it is.
"AI is fundamentally transforming Deutsche Telekom," chief executive Tim Höttges said. "It makes our networks smarter, service more personalized, and opens new business models."
Where The Savings Come From, And What Gets Sold Back
The company named four areas where AI is doing the work: network operations, customer service, software development and administration. Customer service is where it has published the most, and the figures are specific enough to be checked later.
Its Frag Magenta chatbot handled roughly 2.6M calls in the first half of 2026. In the United States, AI agents now take about 40% of customer contacts, and the volume of calls coming in has fallen by 55%. Deutsche Telekom also says complaints dropped around 30% in early deployments and that the time it takes to respond to strain on the network has come down from hours to roughly a minute. Underneath those numbers sits a training programme the company puts at more than 100,000 employees who have been taught to use AI tools, which is over half a workforce it counted at about 197,000 in March.
Selling AI is the other half of the plan, and it is the smaller half by some distance. Deutsche Telekom expects around €250M in AI revenue this year from business customers outside the United States, raised from an earlier €200M, growing to €800M by 2030. Set beside €2.5B of cost reduction, the revenue line is roughly a third the size, which makes this primarily a plan to run the existing business with less overhead rather than a plan to enter a new market.
The product work points at a customer base the operator already has. Deutsche Telekom is building a platform for small and medium-sized businesses to automate recurring work such as customer service and logistics, a segment where European operators hold billing relationships and local support arrangements that cloud providers have never built. Whether those relationships convert into software revenue is an old question for telecoms companies, and one they have more often answered badly than well.
Sovereign AI Is The Larger Commercial Bet
Behind the efficiency story sits a bigger wager on European companies wanting AI infrastructure that stays in Europe. Deutsche Telekom opened an Industrial AI Cloud in Munich with Nvidia in the first quarter of this year, a €1B facility built around more than 1,000 DGX B200 systems holding up to 10,000 Blackwell processors, in a renovated data centre at Tucherpark.
The early customer list reads as a sketch of German industry with a few outsiders attached: SAP, Siemens, the robotics companies Agile Robots and Wandelbots, the drone maker Quantum Systems, the simulation firm PhysicsX and the search company Perplexity. Sovereignty in this context is operational rather than philosophical. The hardware sits in Germany, the people running it are employed in Germany, and the data stays under European law, which matters to manufacturers whose process data is also their competitive position.
That is also where the savings are heading. Deutsche Telekom plans to pass another 2.5M German homes with fibre, having already reached 99% population coverage with 5G, and the company has framed the AI programme partly as a way of funding that build without raising capital expenditure. A cost programme that pays for a network programme is an easier story to tell shareholders than a cost programme that simply lifts margins, and it fits a German political environment where network investment carries weight.
The Industry Was Shedding Staff Before AI Arrived
The part that deserves care is the connection between automation and employment, because telecoms has been reducing headcount for years without AI involved. Analysis published by MTN Consulting in May found no reliable link between headcount cuts and margin expansion at telecoms operators, even allowing for a lag of several quarters, with labour savings frequently offset by rising depreciation on the systems that replaced the people.
The numbers around it are substantial. Telefónica's headcount fell by 18,700 over the year to the end of 2025 and BT's by 7,700, while global telecoms employment declined 1.9% to 4.34M. In a detail that captures the shift in the industry's standing, hyperscaler headcount now exceeds telecoms employment by 3.5%, reversing a position in 2011 when operators employed nearly four times as many people. "Automation has been a central part of headcount cuts at these and similar companies for many years," the firm's Matt Walker wrote. "AI is only an after-thought."
Deutsche Telekom published no net headcount figure alongside its targets. A company reporting that AI agents handle 40% of customer contacts in its largest market, and that call volumes there have more than halved, is describing work that people used to do, and the presentation did not say how many positions that represents or what becomes of them.
Two other gaps sit in the arithmetic. The €2.5B is described as indirect cost savings without a stated geography, while the €1.1B for 2027 explicitly excludes the United States, so the two headline numbers are not measured on the same basis. And the operating metrics the company published, however real, stop short of conversion: calls handled, complaints reduced and response times shortened are not the same as euros, and nothing in the release shows the path from one to the other.
A Cost Programme That Pays For A Network
What Deutsche Telekom presented in Bonn is a plan to run the same company with less manual process, using AI as the instrument and fibre as the destination for the proceeds. That is a coherent position, and it is more honest than the version some operators have offered, where AI appears as a growth story with no corresponding account of what shrinks.
The 2027 figure is now firm enough to be checked against results, and the 2030 figure sits far enough away that it will be revised more than once before anyone can test it. Neither will be easy to verify from the outside, because indirect costs are exactly the category where a reorganisation, a disposal or a change in accounting can move a number as readily as a chatbot can.
The part that will produce evidence sooner is Munich. The Industrial AI Cloud has hardware, named customers and a quarterly cost attached to it, and German manufacturers either buy sovereign AI capacity at the scale Deutsche Telekom has built or they do not. That answer arrives well before 2030, and it will say more about whether the operator has found a new business than any figure in the efficiency column.
