They Raised; We Built

The scoreboard, updated this month. Spring Health: valued at $3.3 billion. Lyra Health: $5.58 billion at its 2022 peak. Pathos: a $365 million Series D in 2025. Woebot Health: roughly $114 million raised, then its consumer app closed in 2025. Wysa: around $30 million. Talkspace: acquired by Universal Health Services for about $835 million, the deal closing in August 2026. Billions of dollars, the best engineering talent in the world, and a half-decade head start.

And the category still has not produced what we are building: a foundational mental health large language model, designed by a licensed practitioner, trained on clinical data with reinforcement learning, available as a HIPAA-compliant API that any of the million-plus licensed providers worldwide can build on. Read that list again and notice what the money bought. Employer EAP channels. Consumer chatbots. Closed systems with proprietary data and no provider API. In Woebot’s case, a shutdown.

This is not an argument that those companies failed. Talkspace’s acquisition is a genuine validation that the market is real and consolidating; UHS called the combination the nation’s first end-to-end continuum of behavioral healthcare. It is an argument about what capital efficiency looks like when the founder is the domain expert. While the funded companies were optimizing consumer engagement metrics, a practicing clinician was building the thing the market actually lacked, one bootstrapped app at a time.

The Antiphon ledger, for comparison. Zero dollars of external capital. Live iOS and web applications with more than 3,100 real users. HIPAA-compliant infrastructure on Google Cloud. A reinforcement-tuned model in active development across five languages. Institutional review board approval. A SAM.gov registered federal contractor. Listed on CMS.gov as a Health Tech Ecosystem Conversational AI pledgee, alongside OpenAI, Google, Microsoft, and Anthropic. An appointed seat on CMS’s conversational AI workgroup. Every dollar of that enterprise value is pure: no dilution, no preferences, no liquidation stack.

Capital is not a moat. It is a head start with an expiration date, and in AI the expiration dates keep getting shorter. The durable moats here are clinical provenance, a live data flywheel with consented real-world data, and federal validation that cannot be purchased. They raised; we built. The raise we are doing now does not fund experiments. It scales something that already works.

Figures: company disclosures; Mordor Intelligence (2025) on the $1.82B AI mental health market growing toward $9.96B by 2031; Becker’s Hospital Review on the UHS-Talkspace close (August 2026).