Ookla, a network performance measurement company, recently shared research which found network responsiveness for AI applications varies significantly by location.
Geography, fibre routing and hyperscaler infrastructure were found to play a bigger role than population size or the presence of local data centres.
The research by lead analyst Mike Dano compared cloud latency, the round-trip time between a user’s device and major cloud data centres run by Amazon Web Services (AWS), Google Cloud, Oracle and Microsoft Azure, across a mix of large and small U.S. cities.
Dano noted the gap between the fastest median latency of 104 milliseconds (ms) recorded in Providence, Rhode Island, and the slowest rate of 146ms in Roswell, New Mexico, was just 42ms, which highlights the depth of the fibre backbone in the U.S.
The report singled out Chicago and Roswell as case studies showing how location shapes performance.
Chicago, a major internet hub with 208 data centres, held a clear latency advantage over Roswell for eastbound traffic, recording round-trip times of 27ms to 44ms to endpoints in Ohio and Virginia compared with 85ms to 95ms from Roswell.
But the results are reversed for westbound traffic. Roswell, despite having no data centres of its own, benefited from its position along a southern fibre corridor, reaching Los Angeles, California in 62ms versus Chicago, Illinois’ 75ms, and Las Vegas, Nevada in 63ms versus Chicago’s 78ms.
He also found meaningful differences between mobile operators as well. In Q2 2026, Verizon generally delivered the fastest median cloud latency from Roswell to western U.S. endpoints, while AT&T and T-Mobile U.S. showed stronger eastbound connections.
In Chicago, T-Mobile U.S. was often the fastest operator in both directions. Dano attributed the differences to variation in each carrier’s traffic steering, peering and interconnect policies.
International routing
In addition, Chicago’s proximity to transatlantic cable infrastructure gave it lower latency to European endpoints, while Roswell’s position benefited from connections to Asia-Pacific destinations via West Coast subsea cable landing points.
For Latin America, Roswell’s geographic proximity did not translate into a routing advantage, since transoceanic cables to Brazil largely depart from East Coast and Gulf Coast hubs. Roswell recorded 195ms to 203ms to Sao Paulo, Brazil, compared with Chicago’s 160ms to 161ms.
Dano’s research also found Oracle’s cloud endpoints consistently registered a few milliseconds higher latency than AWS, Google Cloud Platform and Azure across the U.S., which he attributed to differences in peering relationships and connectivity endpoints.
The report noted emerging technologies including edge computing, local traffic breakout and L4S could help flatten geographic disparities over time by processing data closer to users rather than routing it to distant, centralised data centres.
On-device AI processing was flagged as a further alternative, though Dano stated it remains unlikely such processing will be capable of handling complex AI requests in the near term.
“This all speaks to the ongoing AI-powered data centre boom, which is expected to surpass a whopping $700 billion in spending this year alone”, Dano stated. “It’s possible that a renewed focus on network latency will follow”.
Source: Mobile World Live
Image Credit: ShutterStock
