Deutsche Telekom’s Uplink Reality Check: Why AI Traffic Won’t Inflate RAN Capex

Deutsche Telekom’s Uplink Reality Check: Why AI Traffic Won’t Inflate RAN Capex

For the past two years, the telecom vendor community has been selling a compelling story: AI is coming, and it will break the mobile network’s fundamental asymmetry. Smart glasses, autonomous robots, agentic AI assistants, and real-time video analytics will flood the uplink, forcing operators to re-architect their radio access networks (RAN) with more uplink capacity, more compute, and more intelligence. The implication: a new capex supercycle, with Open RAN and AI-RAN as the primary beneficiaries.

Deutsche Telekom (DT) just threw cold water on that narrative. In a candid assessment reported by Light Reading on October 7, 2026, DT’s senior VP for group controlling—and incoming CFO—Dhananjay Mirchandani said he could not recall a single internal conversation where the CTO or anyone else requested an incremental investment to prepare for additional uplink or AI-related traffic. That’s not a minor footnote. It’s a direct challenge to the assumption that AI will automatically translate into higher RAN spending.

For engineers and planners, this is a signal to separate the physics from the financials. The uplink problem is real, but the solution may be absorbed within existing roadmaps rather than funded by new money. And that has profound implications for how Open RAN and AI-RAN capabilities get deployed.

Mobile networks have always been downlink-heavy. Typical 5G NR configurations allocate far more time and frequency resources to the downlink than the uplink. In TDD bands, the pattern is often 3:1 or 4:1 downlink-to-uplink slots. This asymmetry is baked into everything from frame structure to RF front-end design, and it reflects decades of consumer behavior: we download more than we upload.

AI workloads threaten to flip that. Consider a pair of smart glasses streaming 1080p video to a cloud AI for object recognition. The uplink demand is continuous and can be tens of megabits per second. Multiply that by thousands of users in a dense urban cell, and the uplink becomes the bottleneck. Agentic AI—where software agents act on behalf of users—could generate bursty but high-volume uplink traffic from devices that are constantly sensing and reporting.

Ericsson has been particularly vocal about this, pinning its 6G hopes on AI’s rising uplink demands. The company’s argument is that 6G will need to be designed from the start with uplink-centric use cases in mind, potentially requiring new spectrum, new duplexing schemes, and more compute at the edge.

But here’s the engineering reality: you can’t just add uplink capacity without cost. More uplink slots mean fewer downlink slots, which degrades the consumer experience that still drives revenue. You can add more carriers, more antennas, or more sites—all of which cost money. You can deploy supplementary uplink (SUL) in lower bands, but that requires spectrum and new radio units. You can use AI to schedule uplink traffic more efficiently, but that’s a software upgrade, not a forklift replacement.

DT’s Position: Confidence in Existing Capex Plans

Mirchandani’s statement, as reported by Light Reading, was blunt: “I cannot recall a single conversation in which Alex Jenbar [CTO of Telekom Deutschland] … or somebody else approached us and said, ‘in addition to whatever we currently have on our roadmaps for RAN modernization, that there is an incremental requirement for an investment to be able to prepare ourselves for additional uplink/AI-related traffic.’”

He added that this reflects “the degree of confidence we have in terms of our own capex planning, specifically related to mobile network capacity.”

That confidence is notable because DT is not a laggard. It is in the middle of a major RAN modernization: replacing Huawei equipment with open RAN technology from Nokia at 3,000 sites by 2027. It had 1,000 Nokia sites live in 2025. This is one of the largest Open RAN deployments in Europe, and it’s happening in a brownfield environment with strict performance requirements.

If any operator were positioned to see an AI-driven uplink crunch, it would be DT. Yet its CFO-in-waiting says the capex plan is sufficient. That suggests DT believes it can handle AI uplink growth through a combination of existing spectrum, software upgrades, and the inherent flexibility of its Open RAN architecture—without a step-change in spending.

The Open RAN Angle: Flexibility Without Extra Capex

DT’s stance is actually a vote of confidence in Open RAN’s ability to absorb new requirements through software. In a traditional proprietary RAN, adding a new feature like AI-based uplink scheduling might require a hardware refresh or a costly license. In an Open RAN architecture with disaggregated CU, DU, and RU, and with open interfaces like the E2, O1, and A1, operators can introduce new xApps and rApps that optimize uplink performance without touching the hardware.

For example, a near-real-time RIC (near-RT RIC) could host an xApp that dynamically adjusts the uplink-downlink slot ratio based on traffic patterns. During periods of high uplink demand—say, a factory floor with many AI cameras—the xApp could shift the ratio to favor uplink. During consumer-heavy periods, it could shift back. This kind of dynamic adaptation is a core promise of Open RAN, and it doesn’t require new capex if the underlying hardware has enough headroom.

Similarly, AI-powered beamforming and interference management can improve uplink spectral efficiency, effectively squeezing more capacity out of existing spectrum. Nokia’s recent AI-RAN field trials with eight operators, including DT, are exploring exactly these kinds of gains. If successful, they could delay or eliminate the need for additional uplink spectrum.

The catch is that these software-based improvements have limits. You can’t create capacity out of thin air. If AI uplink traffic grows exponentially, at some point you need more physical resources. DT’s bet is that the growth will be gradual enough—and the software gains large enough—that the existing capex plan can keep up.

The Vendor Disconnect: Selling a Problem Operators Don’t Yet See

The vendor community has been aggressively marketing AI-RAN as the solution to the coming uplink apocalypse. Ericsson, Nokia, Samsung, and others have all announced AI-RAN platforms, GPU-accelerated RAN compute, and AI-powered scheduling. The message is clear: AI will change everything, and you need to buy our new stuff.

DT’s comments suggest a disconnect. Operators are not yet seeing the traffic patterns that would justify a new spending cycle. Agentic AI and physical AI are still nascent. Smart glasses remain a niche product. Industrial robots with high uplink needs are confined to specific verticals and often use private 5G networks, not macro RAN.

This doesn’t mean AI-RAN is useless. It means the business case is different. Instead of selling AI-RAN as a capacity solution for a flood of uplink traffic, vendors should be selling it as an efficiency solution: lower cost per bit, better energy efficiency, automated operations. Those are the metrics that operators like DT are actually tracking. Mirchandani’s comments imply that DT’s capex is driven by coverage, capacity, and modernization needs that already exist—not by speculative AI traffic.

DT’s position is about the medium term—the next 3 to 5 years. In the longer term, 6G may change the calculus. Ericsson and others are already arguing that 6G must be designed with uplink-centric use cases in mind. That could mean new spectrum bands, new duplexing schemes (like flexible duplex or full-duplex), and new network architectures that put more compute at the edge.

If 6G standardization embraces uplink-heavy AI workloads, then operators will eventually need to invest. But that’s a 2030s problem. For now, DT is signaling that it will not pre-emptively spend on 6G-scale uplink capacity when the demand isn’t there. This is a rational, cash-flow-driven approach. It also puts pressure on vendors to prove that AI-RAN can deliver tangible benefits today—not just promise future capacity.

What This Means for Open RAN Engineers

For engineers working on Open RAN and AI-RAN, DT’s stance has several practical implications:

  1. Focus on software efficiency, not hardware expansion. The near-RT RIC and its xApps are the primary tools for adapting to new traffic patterns. If you can demonstrate uplink throughput improvements or latency reductions through intelligent scheduling, you have a compelling story that doesn’t require new capex.

  2. Energy efficiency is the new capex. Operators are under pressure to reduce opex, and energy is a huge component. AI-RAN features that can power down carriers or MIMO layers during low traffic, or optimize beamforming to reduce RF power, are directly monetizable. DT has been a leader in energy-saving initiatives, and it will prioritize features that reduce opex.

  3. Interoperability and testing remain critical. As RCR Wireless reported from the Intelligent RAN Forum 2026, integration timelines for active antenna units have fallen from months to weeks, but testing AI applications is still a challenge. If operators are going to rely on software to handle AI uplink traffic, they need confidence that multi-vendor xApps will work reliably. That means investment in testbeds and validation—not necessarily in new radios.

  4. The uplink problem may be solved by network slicing. DT’s peers, like Reliance Jio, are deploying 5G network slicing to create dedicated uplink capacity for specific services. If slicing can isolate AI uplink traffic and allocate resources dynamically, it could avoid the need for blanket uplink upgrades. Open RAN’s disaggregated architecture is well-suited to slicing, but it requires tight integration between the RAN and the core.

  5. Don’t over-index on AI-RAN hype. The blog’s own archives are full of AI-RAN announcements. DT’s comments are a reminder that operators move at the speed of their budgets. Engineers should be skeptical of vendor claims that AI traffic will force a rip-and-replace. The more likely path is incremental software upgrades within existing hardware.

TelecomLead reported on October 6 that global telecom capex rose 5% in H1 2026, with operators shifting spending to capacity, AI, and network modernization. That sounds bullish, but the details matter. Much of that spending is going to fiber, data centers, and core network functions—not necessarily RAN uplink capacity. DT’s comments suggest that RAN capex is being driven by modernization (like the Huawei rip-and-replace) and coverage, not by AI uplink.

This aligns with Dell’Oro’s projections of $35 billion cumulative AI RAN revenue by 2030. That’s a big number, but it’s spread over many years and includes software, services, and compute. It doesn’t imply a sudden doubling of RAN capex. Instead, it suggests a gradual reallocation of existing budgets toward AI-capable platforms.

The Skeptical Engineer’s Takeaway

Deutsche Telekom’s message is clear: show me the traffic, then we’ll talk about capex. The company is confident that its Open RAN deployment with Nokia, combined with software upgrades and AI-driven optimization, can handle the AI uplink traffic it expects in the medium term. If that confidence is misplaced, DT can always add capacity later—but it won’t spend ahead of demand.

For the Open RAN ecosystem, this is both a challenge and an opportunity. The challenge is that the AI uplink story alone won’t unlock new budgets. The opportunity is that Open RAN’s software-centric architecture is exactly what operators need to adapt without massive new spending. If vendors can demonstrate real, measurable gains in spectral efficiency, energy efficiency, and automation, they can win share within existing capex envelopes.

The AI era will indeed put new demands on the RAN. But as DT’s CFO-in-waiting suggests, those demands may be met with smarter software, not bigger budgets. Engineers should design accordingly.

Sources

Related Posts

Fujitsu's 1Finity Bets Big on India: 200 Engineers, AI-RAN and the Open RAN Waiting Game

Fujitsu's 1Finity expands its India team to 200+ engineers, targeting AI-RAN, 6G and optical networking as it waits for the Open RAN market to finally take off.

WiSig Opens Commercial Orders for Made-in-India O-RAN Massive MIMO Radio: What It Means for Open RAN Economics

WiSig Networks opens commercial orders for its Made-in-India O-RAN 32TR Massive MIMO ULPI radio, with volume shipments planned for 2027. A decision memo for Open RAN operators and vendors.