Does Photoshop's Neural Filter latency spike at a specific concurrent request threshold?

Adobe Photoshop's Neural Filters rely on Adobe Sensei cloud infrastructure, where each filter request incurs round‑trip latency that grows with server load. Observations indicate latency remains between 200 ms and 800 ms while concurrent request volume stays below roughly 10 requests per second, but begins to exceed 2 seconds once volume rises above 20–30 requests per second. Adobe does not publish an SLA or explicit throttling limits for Neural Filters, and reported thresholds vary by region and Creative Cloud plan.

The goal is to determine whether a consistent concurrency threshold exists where latency predictably degrades, or whether the observed spikes stem from shared resource contention that fluctuates with overall platform demand. Current evidence is anecdotal and limited to informal benchmarks, making the exact inflection point an unresolved behavior.

Can a reproducible latency inflection point be identified across different Creative Cloud tiers? Does the threshold shift when requests originate from different geographic regions? Which filter categories exhibit the steepest latency increase under concurrent load?