Most 5G networks aren’t ready for conversational agentic AI: Ookla

Details: John Tanner; Category: Operators; 07 July 2026; 709 views

Conversational agentic AI is heralded as the next big opportunity for 5G – provided 5G operators have sufficient uplink speeds to support its latency requirements. But a new report from Ookla finds that most don’t – and the gap is getting worse.

The new report – “Beyond Download Speed: Benchmarking 5G Mobile Networks Against AI Workloads” – points out that most 5G networks deployed so far have been engineered for peak download speeds. The problem is that AI apps depend on faster uplink speeds with lower latencies. Text-based AI requires uplink latencies under 50ms, while voice-based AI requires uplink latencies under 40ms, and video-based AI apps like AR vision require latencies under between 10-30ms.

According to the report – which surveyed 86 operators in 22 markets across North America, Europe, Asia Pacific, the Middle East, and Latin America – 18 of those markets have at least one 5G operator that’s able to support text-based AI, while 5G in only 13 markets can support conversational voice AI. Only Singapore’s 5G networks are able to support AR vision (and even then, only barely).

For developing markets, 5G telcos in Malaysia, Indonesia, Thailand, the UAE and Brazil are among those with enough uplink latency to support text-based and conversational voice AI, while telcos in the Philippines can support text-based AI. India is one of four markets (which include South Korea, Spain and the US) whose 5G networks can’t meet even text-based AI latency requirements.

AI reshapes traffic loads

At issue, says Ookla industry analyst Affandy Johan, is the fact that AI traffic is very dynamic, depending on the application. Text chat, conversational voice, multimodal and AR vision, generated video, and agentic activity each load the network differently. That means AI isn’t about how much downlink capacity you have, but the shape of the traffic itself, which is heavier on upload, always on, and bursty, rather than download-led and session-based.

“The metrics that decide AI performance – upload capacity, latency under load, and the path to the cloud – follow a different order, and the gap widens as adoption shifts toward heavier use cases like conversational voice and multimodal AI,” Johan wrote in a blog post. “In aggregate, the most demanding modality remains beyond what current 5G delivers, even in the most advanced deployments.”

Another way of looking at it is that most 5G networks were deployed on the sensible notion that users download far more than they upload. Ookla says the typical operator devotes only around 10% of throughput to the uplink, and that share has actually dropped since 2023 in more than half of markets surveyed, even though absolute upload speeds have increased.

That's a problem, writes Johan, because text-based AI traffic already runs at roughly a 29/71 uplink-to-downlink split by volume, while conversational and agentic AI workloads require more of a 50/50 split.

Another issue is that while most networks clear the AI baseline at rest, that changes drastically when the network is fully utilised, Johan says. “Degradation ratios, the multiple by which loaded latency exceeds the baseline, stretch from 3.7x in the U.K. to 11.4x in Thailand, where median loaded latency reaches 960.3 ms.”

The prospects for mobile AI apps become even more grim when looking at the latency from the 5G network edge to the cloud where the various LLMs reside. Latencies vary from one cloud provider to another, and in developing markets surveyed, they can be anywhere from 63ms (for AWS in Indonesia) to over 164ms (for Google Cloud in Brazil), Ookla says.

5G-A and 6G will fix it: GSMA

If there’s a bright spot, it’s that the mobile sector overall is aware of 5G’s current uplink limitations in supporting agentic AI, and is banking on evolution to 5G-Advanced and 6G to address that.

At last month’s MWC Shanghai event, the GSMA showcased China Telecom, China Mobile and China Unicom (none of whom are covered in the Ookla report) as harbingers of things to come. All three have deployed 5G-A, and all three have launched token-based AI plans that package access to AI usage in a similar way to traditional mobile data bundles.

However, it’s early days for 5G-A, which has only been commercially launched by 11 operators around the world as of April 2026, according to the Global mobile Suppliers Association (GSA). Meanwhile, 6G is expected to reach its initial commercial stage by 2030.

On the other hand, going by the discussions at last month’s Digital Transformation World event in Copenhagen, agentic AI adoption in the telco space is being slowed down by regulatory concerns about the impact of AI, as well as issues like trust and AI sovereignty. That could buy extra time for 5G operators to upgrade their uplinks accordingly.