A new study by Ookla concludes that upload capacity, latency and connection to the cloud will be decisive in supporting the next generation of artificial intelligence applications. Europe starts with an advantage, while Spain combines strengths with some pending challenges.
The emergence of AI is transforming the way we understand mobile networks. For years, download speed has been the main indicator to measure the quality of 5G, but the arrival of applications such as conversational assistants, intelligent agents or future augmented reality experiences forces us to rethink that model. Now, factors such as upload speed, latency or connection to the data centers where AI models run will be as important as traditional megabits per second.
This is the main conclusion of the report Beyond Download Speed: Benchmarking 5G Mobile Networks Against AI Workloads, prepared by Ookla, which analyzes the performance of 5G networks in 22 markets and 86 operators around the world. The study concludes that most current networks can already support text-based AI applications, but still present important limitations to address the next generation of intelligent services, characterized by the use of voice, video and multimodal content in real time.
The change responds to the very nature of these applications. Unlike video or social media, where download traffic predominates, AI needs to continually send information from the device to the cloud to process it and generate a response. For this reason, Ookla considers that the readiness of a network should be measured by five parameters: upload capacity, latency, latency under load, latency to the cloud infrastructure and connection stability or jitter.
Europe, well positioned for the adoption of AI
The analysis places Europe in a favorable position thanks to the quality of its connectivity with large cloud service providers. Germany, the United Kingdom, France and Ireland record some of the shortest access times to platforms such as AWS, Azure or Google Cloud, while Finland, Norway and Sweden stand out for offering one of the best latencies towards Azure. This proximity to data centers allows us to reduce the total time it takes an AI application to respond, a key aspect for voice assistants or intelligent agents.
Furthermore, Germany stands out as one of the most balanced European markets by combining high upload capacity with low latency, becoming one of the continent’s references in preparation for AI.
Spain, good growth figures but with room for improvement
Spain obtains mixed results. On the one hand, it occupies seventh position among the 22 markets analyzed in percentage of capacity dedicated to upload traffic, with 11.48%, ahead of countries such as the United Kingdom, Italy, France or the United States. A relevant fact if one takes into account that AI applications will significantly increase the volume of information that users send to the cloud.
However, the report also identifies a less positive trend. Between 2023 and 2025, Spain reduced the weight of upload traffic by 2.9 percentage points, the second largest decrease in the study after Indonesia, precisely when this parameter gains importance to support new artificial intelligence workloads.
Latency is another challenge. Spain registers an average of 50.2 milliseconds, sufficient for current text-based applications, but slightly above the goal set by Ookla to offer an optimal experience in conversational assistants. Along with the United States, India and South Korea, it is one of four markets that do not meet that recommended threshold.
Prepare for the next generation of AI
Although current networks respond adequately to the most widespread AI applications, the report warns that none of the 22 economies analyzed still meets the necessary requirements to safely support multimodal or augmented reality applications, which are much more demanding in terms of latency and upload capacity.
AI applications will significantly increase the volume of information that users send to the cloud
To face this scenario, Ookla recommends accelerating the deployment of 5G Stand Alone, strengthening upload capacity, optimizing direct connections with large cloud providers and incorporating technologies such as network slicing, which will allow offering more stable connections with lower latency for AI applications.
The conclusion of the study is clear: the 5G race will no longer be decided solely by download speed. In the era of artificial intelligence, it will be network responsiveness, upstream efficiency and proximity to the cloud that will make the difference between infrastructures prepared for the next generation of intelligent services and those that will need to continue to evolve.
