Tavus

Tavus has positioned itself as the infrastructure layer for the next interface shift: from chat windows to face-to-face computing. Its November 2025 $40M Series B — led by CRV with Scale Venture Partners, Sequoia, Y Combinator, HubSpot Ventures, and Flex Capital participating, bringing total funding to roughly $64M — capitalizes the only API-first company with a full proprietary stack for real-time conversational video: Phoenix-4 for lifelike rendering at conversational latency, Sparrow-1 for turn-taking, tone, and timing, and Raven-1 for contextual visual perception. That trinity matters because real-time is a different physics problem from Synthesia-style rendered video — sub-second latency with natural interruption handling cannot be assembled from off-the-shelf parts — and Tavus sells it as a Conversational Video Interface that is deliberately LLM-agnostic, letting developers bring their own brain while Tavus owns the human-presence layer, the same wedge Twilio and Stripe used to own their primitives. The customer surface is broad and budget-backed: AI interviewing, patient intake, sales, coaching, and customer support all convert measurably better with a face, and every enterprise now piloting voice agents is one step from wanting embodied ones. The PALs launch — persistent AI humans with memory and agency — extends the platform up the stack without abandoning the API business. Founded in 2020 by Hassaan Raza and Quinn Favret, the company has already survived one full pivot (from personalized outreach video to real-time) with investor conviction intact; in a category where HeyGen and Synthesia are enterprise apps, Tavus is the arms dealer to everyone building anything else.

Tavus is a $64M-funded research lab competing on a frontier where its adjacent giants treat its core product as a feature. HeyGen (at $100M+ ARR) ships interactive avatars, Synthesia (at $4B and $150M ARR) is moving the same direction, D-ID and Soul Machines have sold conversational faces for years, cheaper API upstarts like Anam undercut on price, and the deepest risk sits below: OpenAI, Google, and open-source realtime stacks are collapsing the latency and quality barriers that make Tavus's orchestration valuable, and a frontier lab shipping native audio-visual conversation would commoditize the entire middle layer overnight. The unit economics of real-time rendered video are unforgiving — GPU-seconds per conversation-minute at low latency mean COGS scale linearly with usage in a way text APIs do not — so growth consumes capital fast for a company an order of magnitude less funded than its enterprise-app rivals. The PALs launch reads two ways, and the unflattering read is strategy drift: a developer-infrastructure company simultaneously building consumer-facing AI humans with agency, calendars, and email access is splitting a 11–50-person team across two brutally competitive fronts, and "emotionally intelligent AI companions" imports all the regulatory and reputational hazard now engulfing the companionship category. Demand-side conviction is also unproven at scale: face-to-face AI remains a conversion hypothesis — many users demonstrably prefer voice or text to an uncanny video head — and if the interface shift stalls at novelty, Tavus's differentiated stack serves a market that never arrives. The Series B buys perhaps two years to prove the primitive is a platform, against competitors who can afford to be wrong for a decade.