Deepgram Shifts From API to Applications

Deepgram is turning a component business into an application business, which gives it more revenue per customer but puts it into direct conflict with the builders that used to sit above it. The shift is visible in three moves at once, text to speech, full voice agents, and restaurant ordering, where Deepgram is no longer just selling speech recognition pipes, it is packaging a working outcome that handles calls, menus, and actions inside customer systems.1 2 3 4 5

• Deepgram now describes restaurants as a direct solution area for drive thru, phone, and kiosk ordering, built for large chains and tuned to each brand’s menu and POS stack. That is much closer to selling an AI employee than selling an API, and it overlaps with the platforms that already use Deepgram underneath for similar voice workflows.4 5
• ElevenLabs has already made the same climb up the stack. It moved from synthetic voice generation into ElevenAgents, a platform to build, launch, and monitor agents across phone, web, apps, SMS, and chat, with workflow logic and enterprise connectors. Deepgram is now chasing a rival that started from TTS and reached the application layer faster.2 6 7 8
• The scale gap explains why this matters. Deepgram is tracked at about $100M annualized revenue as of July 31, 2026, while ElevenLabs is at about $600M as of June 30, 2026. Once speech models become swappable building blocks, the bigger prize shifts to owning the full workflow, the customer relationship, and the distribution surface.1 3 9

The next phase of voice AI looks more vertically integrated. Infrastructure vendors will keep moving toward packaged agents in categories like support, healthcare, and restaurants, while application companies will keep adding their own model routing and voice layers. The winners are likely to be the companies that control both the conversation quality and the business action at the end of the call.2 4 6 8