Cartesia

Roadmap & Position in Voice AI

Real-time multimodal voice AI built on State Space Model foundation architecture.

Company Overview

Cartesia builds real-time voice AI on State Space Models, an architecture its founders invented at Stanford, delivered as a developer-first stack of text-to-speech (Sonic), speech-to-text (Ink), a voice agent platform (Line), and on-device deployment (Edge). Customers reach the product through SDKs, APIs, and a web playground, with private cloud and on-premise options for enterprise buyers.

What They're Building

The company's public product roadmap & what they're committed to building.

Sonic (TTS)

Real-time text-to-speech at 40ms latency on the Turbo variant, 90ms on Full, with 42+ languages, instant voice cloning, and fine-grained emotion control.

Ink (STT)

Streaming speech-to-text built for real-time use, with multi-speaker separation and diarization.

Line

Voice agent development platform, the product that turns Cartesia from a model provider into the platform layer competitors build on top of.

Edge

Open-source on-device model ecosystem with Apple Metal support, a bet that meaningful voice AI workloads move off the cloud entirely.

Developer API and SDKs

Python and JavaScript SDKs, WebSocket streaming, and a REST API, the primary distribution surface for developers embedding voice in their own products.

Voice Library

Curated catalog of voices across Tier 0 and Tier 1 languages, tagged with detailed metadata covering pitch, pace, and emotion range.

Latest Intelligence

Zeitgeist tracks private signals to determine where the company is heading strategically.

No Signals Yet

Founder and Key Execs

Karan Goel, Co-Founder and CEO

Co-inventor of State Space Models, Stanford AI Lab

Albert Gu, Co-Founder and Chief Scientist

Stanford AI Lab PhD, primary author of foundational SSM research

Founder Force Multiplier

Karan Goel and Albert Gu co-invented State Space Models at the Stanford AI Lab, where Gu authored the foundational research. That gives Cartesia a depth of architectural understanding that competitors adopting SSMs second-hand cannot replicate quickly, and a recruiting magnet for the small number of researchers who work at that layer.

Funding History

2023 | Founded at Stanford AI Lab
2023 | $27M Seed led by Index
2025 | $64M Series A led by Kleiner Perkins

Notable Open Roles

Tech Lead Manager, Data Infrastructure

Research role tied to Foundation model in strategic role description.

Transcription Operations Manager (Contract)

Product role tied to Foundation model in strategic role description.

Technical Account Manager

GTM role tied to Enterprise readiness: technical account manager + enterprise.

Competitors

Direct: ElevenLabs, Deepgram, OpenAI, Google, Microsoft Azure Speech, Amazon Polly

Indirect: Retell AI, Vapi, LiveKit, PlayHT

Cartesia

's Moat:

Architectural ownership of State Space Models, the compute primitive Cartesia's founders invented, which gives the company a multi-year research lead at the layer where latency and scaling advantages compound.

How They're Leveraging AI

Building AI safety as a product pillar, not just compliance

Enterprise-grade voice agent infrastructure that runs inside customer-managed environments (private cloud, on-premise) to unlock regulated industries. This is the main wedge for moving upmarket and the clearest point of direct competition with Deepgram, which already has production on-prem deployments in federal and restaurant environments.

Revenue tripled from $200M to $330M+ ARR in 13 months

Cartesia's founding bet: advancing State Space Model and hybrid architectures to deliver real-time voice AI that beats transformer-based competitors on speed, cost, and on-device deployment. SSMs offer linear scaling and constant-time inference, structural advantages over transformers that compound as models get larger and latency requirements get tighter.

Pivoting from audio-only to full multimodal creative suite

Cartesia is industrializing its multilingual data operation, transcription across 30+ languages, voice actor sourcing, annotation quality control, and automated data governance. This directly attacks ElevenLabs' 500,000+ hours of proprietary audio training data advantage. The bet is that a well-engineered data pipeline can close a training data gap faster than brute-force audio collection.

State Space Voice Models

Ultra-low-latency voice AI built on State Space Models for speech generation, transcription, and on-device conversational agents.

AI Use Overview:

Cartesia runs State Space Model architectures across the full voice stack (TTS, STT, voice agents, on-device inference, multilingual generation, emotion control), trading the quadratic cost of transformer attention for linear scaling and sub-100ms latency.

More Similar Companies

ElevenLabs

Building human-like AI voices that speak, clone, dub, and converse in 70+ languages

Having established defensible voice quality and market share through its API, ElevenLabs is now becoming a multimodal generation platform with an enterprise go-to-market engine.

Deepgram

Voice AI infrastructure for real-time speech-to-text, text-to-speech, and voice agents.

Deepgram controls the full vertical stack from bare-metal training hardware to a Rust inference runtime, a cost and latency moat that API competitors riding hyperscaler infrastructure cannot replicate without years of capex.

AssemblyAI

Speech-to-text and audio intelligence APIs for developers building voice-powered applications.

Voice is the next API primitive after text, and AssemblyAI has an accuracy and developer-experience lead over cloud incumbents with better margins than full-stack voice agent startups carry.

PolyAI

Voice AI platform building conversational agents for customer service call centers

Voice agents are one of the clearest enterprise LLM use cases with measurable ROI, and PolyAI has real logos, real deployments, and Cambridge speech research DNA.

Back To All Companies >