“A CSP-controlled layer that can be put in place now is the more defensible path.” – John Abraham

This Appledore Research Note examines TM Forum’s Model-as-a-Service (MODaaS), one of the first projects in the Forum’s AI-Native Blueprint. As CSPs push AI into production, most now run a patchwork of foundation, domain and vendor-embedded models, sourced from different providers and governed inconsistently. MODaaS proposes a single operator-controlled layer between AI agents and the models they call.

We assess the case for that layer against the two obvious alternatives: consolidating on one hyperscaler, or waiting for the wider AI market to standardise. We set out the capabilities MODaaS is designed to provide, and draw lessons from earlier telecom and payments platforms about what makes this kind of layer endure. We also give our candid view of the headwinds it faces as an early-stage, consensus-driven blueprint competing with faster-moving vendor and open-source options.

This note is for CSP AI, IT and strategy leaders, and for the vendors and hyperscalers selling into them, who want a grounded read on whether an AI control plane belongs in their architecture.

Companies mentioned in this report: OpenAI.