QumulusAI , a neocloud infrastructure provider purpose-built for the AI computing era, today announced the release of a new Futurum Research report, sponsored by QumulusAI, which finds that agentic AI can increase token consumption per task by 10 to 100 times compared with a simple inference call.
Announced in Atlanta, the report examines how that increase can expose organizations using per-token services to unpredictable and rising costs as applications move into production and scale across the enterprise.
The report, The Off Ramp From Per-Token Pricing: How Enterprises Regain AI Cost Control With Reserved Bare Metal, examines how enterprises can control rapidly growing AI inference costs by adopting a hybrid strategy that supports moving mature production workloads from per-token pricing to reserved infrastructure. It draws on Futurum Research market forecasts and a survey of 824 AI decision-makers conducted during the first half of 2026.
The report describes a typical progression from serverless APIs for initial experimentation, to on-demand GPUs, reserved infrastructure and ultimately a hybrid AI architecture. In this model, organizations match each workload to the right environment based on utilization, cost, latency, data governance and the level of control required.
"Per-token pricing makes sense when companies are experimenting, but the economics change quickly when AI applications move into sustained production," said Michael Maniscalco, CEO of QumulusAI. "This research shows why enterprises need a more deliberate infrastructure strategy, using flexible services for experimentation and reserved capacity for predictable, high-volume workloads. The goal is not to replace the public cloud, but to put each workload where it makes the most economic and operational sense."
Futurum Research recommends considering reserved bare-metal infrastructure for workloads with sustained and predictable utilization, high token volumes, specialized model-serving requirements, or data privacy and residency constraints. Bursty or experimental workloads may remain better suited to serverless or on-demand services.
"Enterprises are not choosing a single infrastructure model for AI," said Brendan Burke, Research Director, Semiconductors, Supply Chain and Emerging Technology at Futurum Research, and author of the report. "They are building hybrid environments that retain on-demand flexibility while moving predictable production workloads to reserved capacity. As agentic AI raises token volumes, knowing when to make that move will become an important part of controlling costs."
QumulusAI provides dedicated, reserved bare-metal and virtualized GPU infrastructure for AI inference and training. Its infrastructure gives customers direct control over their compute environment and replaces variable per-token charges with costs based on reserved hardware capacity and utilization.
About QumulusAI
QumulusAI is a distributed AI cloud platform that delivers accelerated access to high-performance GPU compute. Through an inference-first, demand-led deployment model across a network of data center sites, QumulusAI brings compute closer to customer demand, helping AI teams and enterprises scale production AI workloads with speed, flexibility and control. By combining rapid deployment with flexible private cloud infrastructure, QumulusAI gives customers a faster, more adaptable path beyond the capacity constraints of traditional centralized and hyperscale cloud models.
