Gladia moves from transcription to voice stack

The bigger prize here is moving from selling cheap audio processing to owning a slice of the full call handling budget. Gladia already turns live speech into text, but a production voice agent also needs to know exactly when the caller has stopped, interrupt itself when the caller starts again, look up order or account data in business systems, and speak back fast enough that the conversation feels natural. That is a workflow product, not just a better speech model.1 2 3 4 5

• Gladia today is sold like infrastructure. Developers stream or upload audio, get back transcripts, summaries, translations, and redaction, then wire that output into their own apps. Pricing is per audio hour, with live transcription at €0.68 per hour, which fits a component sale more than an outcome priced against replacing agent labor.1 2
• Competitors moving into voice agents show what must be added. Deepgram built Flux around end of turn detection and barge in control, because the hard part is deciding when to listen and when to answer. AssemblyAI then packaged recognition, reasoning, and speech into a single WebSocket, shifting from transcript budget to customer service automation budget.3 4 5
• OVH gives Gladia a path to bundle those layers with cloud, AI, and enterprise procurement. OVH said the acquisition lets it internalize Gladia speech technology and offer new voice AI services through OVHcloud and OVHai. That matters because large buyers often want one vendor that can host the model, connect to internal systems, and meet compliance requirements.1 5

The next step is likely a packaged voice stack that sits above raw transcription, aimed at support lines, booking flows, and regulated service calls. If OVH and Gladia can combine speech, orchestration, and enterprise integrations into one reliable service, they can compete for budgets that are many times larger than standalone transcription spend.1 3 4 5