Voice AI: infrastructure, platforms, and vertical applications

1) Market thesis

Voice AI is being restructured from a speech-recognition feature into an end-to-end software stack: foundation models and telephony are commoditizing, while value is concentrating in agent platforms and vertical applications that own workflows, proprietary context, and measurable business outcomes.

2) Market map

Segment 1: Speech and voice-model infrastructure

Companies:

• ElevenLabs — expressive text-to-speech, voice cloning, dubbing, and increasingly broader voice-agent infrastructure
• Deepgram — speech-to-text, text-to-speech, and real-time voice APIs
• AssemblyAI — speech recognition, transcription, and audio intelligence APIs
• Cartesia — low-latency speech generation and voice models
• OpenAI — multimodal models and real-time voice capabilities
• Hume AI — expressive voice generation and emotion-aware interaction models

Dynamics: Model quality, latency, multilingual coverage, and inference cost are improving quickly; differentiation is shifting from raw transcription/TTS accuracy toward real-time interaction, controllability, voice consistency, and distribution into applications. Hyperscalers and model companies may compress standalone infrastructure margins.

---

Segment 2: Voice-agent developer infrastructure and orchestration

Companies:

• LiveKit — real-time communications infrastructure and voice-agent framework
• Vapi — developer platform for building and deploying voice agents
• Retell AI — APIs and tooling for conversational voice agents
• Bland AI — programmable phone agents for outbound and inbound calls
• Synthflow — low-code voice-agent builder for business workflows
• Cognigy — enterprise conversational AI and agent orchestration

Dynamics: These companies abstract telephony, turn-taking, tool calling, interruption handling, observability, evaluation, and model selection. The segment is growing rapidly but faces platform risk: model providers, CPaaS vendors, and contact-center incumbents can absorb many orchestration features.

---

Segment 3: Telephony, communications, and contact-center infrastructure

Companies:

• Twilio — programmable voice, messaging, identity, and communications APIs
• Telnyx — programmable voice, numbers, SIP, and communications infrastructure
• Vonage — communications APIs and contact-center products
• Plivo — programmable voice and messaging APIs
• Five9 — cloud contact-center infrastructure with AI capabilities
• Genesys — enterprise contact-center platform and AI agent tooling

Dynamics: Voice AI depends on reliable numbers, routing, carrier connectivity, recording, compliance, and CRM/contact-center integration. Incumbents possess distribution and enterprise trust, while newer vendors can compete on AI-native architecture and lower-latency experiences.

---

Segment 4: Horizontal enterprise voice-agent platforms

Companies:

• PolyAI — enterprise-grade customer-service voice assistants
• Parloa — AI agents for contact centers and customer service
• Kore.ai — enterprise conversational AI and agent platform
• Cognigy — customer-service automation and contact-center orchestration
• Sierra — AI agents for customer-service workflows
• Ada — automated customer service across digital and voice channels

Dynamics: The winning products increasingly need to handle complete interactions rather than simple FAQ calls: authentication, policy interpretation, system actions, escalation, and compliance. Sales cycles are long, but successful deployments can have high retention because agents become embedded in customer-service operations.

---

Segment 5: Vertical voice applications

Companies:

• Infinitus — healthcare phone automation, including payer and provider workflows
• Hippocratic AI — healthcare-focused AI agents for administrative and patient-support tasks
• Hyro — conversational AI for healthcare organizations
• EliseAI — AI communication and workflow automation for property management and housing
• Smith.ai — AI-assisted and human-backed phone answering for small businesses
• Suki — healthcare voice assistant and clinical documentation platform

Dynamics: Vertical applications can capture more value by owning domain-specific workflows, integrations, compliance, and outcome data. The main constraints are liability, procurement, integration complexity, and the need for human escalation. Some listed companies are voice-led but also use chat, documentation, or human-in-the-loop workflows.

3) Segment likely to win the most value

Vertical applications are most likely to capture the greatest durable value, particularly in healthcare, financial services, insurance, logistics, and property management.

Why:

1. They sell outcomes rather than minutes or API calls — appointments completed, claims processed, leads converted, or support costs reduced.
2. They own workflow context — domain rules, customer records, permissions, and integrations make the product harder to replace than a speech API.
3. They can monetize labor substitution — pricing can be tied to resolved cases, handled calls, or revenue generated.
4. Regulation and operational complexity create barriers — vertical expertise and auditability matter more than model novelty.
5. Proprietary interaction data compounds — real call outcomes, failure modes, and escalation patterns improve the product.

The caveat is that horizontal enterprise platforms may capture more aggregate revenue in the near term because of broader distribution, while vertical applications should have the stronger long-term value capture if they successfully become systems of action.

4) White-space opportunities

A. Independent trust, compliance, and control infrastructure for voice agents

Opportunity: A vendor-neutral control layer that provides identity verification, consent capture, disclosure management, call recording, audit trails, policy enforcement, real-time risk scoring, agent evaluation, and liability-oriented monitoring across voice-agent providers.

This is increasingly necessary for regulated interactions, but capabilities are fragmented among telephony vendors, contact-center platforms, identity providers, and application companies. I am not aware of a scaled independent player that covers this full stack across vendors; companies such as Twilio, Persona, Pindrop, and CallMiner cover portions of the problem.

B. AI-native voice operations for complex field and back-office work

Opportunity: Voice agents designed for workers operating in noisy, mobile, or low-connectivity environments—such as insurance adjusters, home-health workers, technicians, construction supervisors, and logistics operators—with offline support, multilingual dialect handling, photo/document capture, system-of-record updates, and authenticated execution of multi-step workflows.

Most current products focus on customer-service calls, scheduling, or transcription. The opportunity is underserved rather than completely empty—companies such as Augment AI, Abridge, Commure, and industry-specific workflow vendors address adjacent use cases—but there is no obvious scaled category leader for cross-industry, voice-first operational work.