With worldwide artificial intelligence (AI) spending forecast to hit $2.59tn in 2026 and tech giants committing roughly $2tn to long-term capacity, it’s no wonder that digital infrastructure is under immense strain, and research from independent global network service provider RETN has found that while businesses are “addicted” to AI, half don’t want infrastructure built near them, and real-time processing will reverse decades of internet traffic patterns.
The survey, How AI is reshaping infrastructure, energy and the networks that connect it, was conducted in September 2026, polling 2,000 adults aged 18-64 across 11 markets: the US (600); the UK, Germany, the Netherlands and Spain (800); and China, Hong Kong, Taiwan, Singapore, Malaysia and the Philippines (600).
The survey findings were supplemented by insights from senior network experts across RETN’s global team to expose the critical infrastructure bottlenecks, shifting traffic patterns and soaring security risks behind the global AI boom.
Among the key founding assumptions was that AI was reshaping daily lives faster than almost any technology before it, helping us create, build and digest information at an unprecedented rate. The result was that businesses were fundamentally rethinking how they operate, roles are being redesigned, and workers are under pressure to adapt.
Such a dependence was found to be fuelling what was described as unprecedented investment in datacentres, and 53% of those in the survey named AI tools as the biggest driver of new datacentre construction. However, the survey stressed that technology doesn’t rely on datacentres alone – it depends on four connected parts: the datacentres where models are trained and run, the power that feeds them, the carriers whose networks move the traffic, and the people and businesses using it.
Growing power and cooling demands were seen as pushing datacentres into more remote locations, where land and energy are available, but fibre connectivity may be limited. RETN warned that developers increasingly need to involve network carriers two to five years before site turn-up, compared with the previous one to two years.
Contrary to claims that AI is overloading public IP transit, the survey revealed that AI traffic accounted for only a minor percentage of open internet growth. Some 80-90% of RETN’s AI-related requests were for private point-to-point links, dark fibre and dedicated interconnects used to move petabytes of data directly between training clusters.
Overall, as many as 84% of those surveyed used AI, and 54% said losing access would disrupt their lives within days. However, the public was split down the middle: half were prioritising faster AI performance, while the other half would accept slower AI if it means fewer datacentres, and less data energy and water use where they live.
Quality of service when dealing with AI workloads was becoming crucial. Three-quarters of respondents experienced issues with their preferred AI service in the past year. And when disruption occurred, 53% of users in Asia switched to another service, compared with 31% in Europe and 17% in the US. Across all markets, 18% began using an alternative AI provider regularly.
Another interesting finding was that the current levels of real-time processing were set to reverse decades of internet traffic patterns. That is, as AI moves from centralised training to running real-time tasks fed by billions of connected devices, data will increasingly flow from end users back into the network. This shift was said to be set to flip the download-heavy asymmetry of the current internet and push compute power out to the physical edge.
Assessing the findings of the survey and the trends that can be drawn from it, RETN chief executive Tony O’Sullivan stressed how the industry was really only at the very beginning of the changes AI is bringing to the network.
“The AI race is often described as a race for compute, chips, power and data,” he said. “But it is the networks that bring all of these together. Those best placed for the future will not simply be those that build the most capacity, but those that build infrastructure capable of adapting when geopolitics changes and continuing to operate when individual parts fail.”
