The Five Year Desert to Product Market Fit and a $5.3BN Valuation | Shiv Rao, Founder @ Abridge

Key points

• **Abridge — the 'five-year desert' founded 2018, ARR inflection early 2023, now $5.3B valuation.** Series A1 (terrible name) was the hardest round — got 8 on 40 from Whittington Ventures + USV + Bessemer in the wilderness era. **Thesis never changed:** 'healthcare is about people having conversations — that's the most human signal that powers healthcare.' Pivoted on product (consumer app → enterprise SaaS), go-to-market (down-market → up-market), business model. **'You can pivot on the specifics but you cannot pivot on the thesis if you want to survive a 5-year wilderness.'**
• **~40% of Abridge model outputs are from in-house models.** 'Varies month-to-month. Could be 60% next month after distilling a new open-source variant and fine-tuning it.' Reasons to build own models: (a) **latency** — 'doctors have no patience; reduce, reduce, reduce; we want to be like good air conditioning, invisible.' (b) **bell-able tasks** — once a discrete task is solved, you can optimise on latency + cost forever. (c) **PnL control.** Reasons to ride the frontier: open-ended tasks where you'll never be perfect but want monotonic improvement. **Direct echo of Krishna Rao's '3-platform compute strategy' from [Issue 05 ILTB this week](/issues/2026-05-17).**
• **'Doctors now refuse to join health systems that don't have Abridge.'** Overnight, table stakes. Direct evidence of the AI-procurement-mandate dynamic Lemkin called out in [Issue 04](/issues/2026-05-10) and [Issue 05](/issues/2026-05-17). 'Every health system right now is going AI — every single one.'
• **Counter-position framework from Hamilton Helmer's 7 Powers:** 'Build in a way the incumbent can't because it would impact their current business.' Useful for any vertical-AI builder facing a Microsoft / Salesforce / Epic-style threat. **Doesn't compete with Epic — builds on top of Epic** ('we are the intelligence layer; the conversation is our wedge'). Doesn't see Microsoft (Nuance, $20B acquisition) or Nabla 'much anymore.' 'Bundling isn't the existential threat people assumed it would be — category creation matters more.'
• **'Closer to the flow of money' as the strategic next step.** Notes → orders → billing. Doctors aren't compensated for care delivered; they're compensated for care *documented.* Abridge knew from day 1 that 'notes are bills' and architected accordingly. **Three stakeholders to thread:** CMIO (Chief Medical Information Officer), CIO, CFO — all with different lenses. The CFO is the long-term enterprise-value unlock.
• **Doesn't sell data — 'trust is everything in healthcare; the industry moves at the speed of trust.'** Asked moral question: 'would you build a foundation model for healthcare for the greater good?' Answer: 'absolutely — but only with our health-system partners' eyes wide open. We have an 'earn the right' refrain inside the company: every new feature leveraging data requires explicit partner blessing in the contract.'
• **'Talent over frontier-model access — without question.'** Asked: 'six months exclusive access to next-gen frontier model, OR six months exclusive access to the best research/engineering pool — which?' Answer: B, immediately. **'Ultimately it's always about the people. If you've got incredible people, you're going to build your own models, your own primitives — AND ride the frontier wave.'** Cross-references Krishna Rao's 'talent density beats talent mass' from this week's ILTB.
• **The 'taste good things to have good taste' company value.** Pre-2022 written. 'Reading the latest arXiv papers about new ML models, thinking about latest UI/UX patterns, living at the edge of culture if we want to create it.' **Useful pattern for founders building product taste — taste is exposure × discernment, not innate.**
• **Wartime/peacetime framing — 'no other way to be.'** 'You're always competing. You're in a market moving so fast. Who's a peacetime CEO?' Even the Costco CEO admired by Shiv may technically be peacetime. **'The minute the founder transitions out of the CEO role, I'm out.'** Direct alignment with Chesky [Issue 04](/issues/2026-05-10) and the Koch Brothers [Issue 05 this week](/issues/2026-05-17) on founder-led durability.
• **'Sucks the most: the chapter where you don't feel pressure.'** 'Pressure makes diamonds. If you're not figuring out how to 10x your ambitions, if you're not finding new power pockets, if you're coasting — that doesn't feel good at all.' **Anti-cruise-control operating principle.**
• **Goldman 24x token consumption by 2030 = 'directionally right.'** Shiv: 'You just can't get enough of this technology once you start to use it. Even as inference costs come down, you end up using more and more. For certain jobs you'll never be good enough on — those are bottomless pits.' Companion to All-In's multi-sensory model 1000x token claim this week.
• **Forward-deployed engineers from OpenAI + Anthropic 'are the strongest sign vertical AI has years of opportunity ahead.'** 'If that wasn't a sign there's an incredible opportunity for the foreseeable future for vertical AI, I don't know what is.' **The labs are pivoting to embedded delivery because the frontier-model-only approach can't get into messy enterprise workflows fast enough.**
• **Henry Kravis advice on going public.** First meeting: 'Do you want to go public?' Shiv started a clumsy answer. Kravis interrupted: 'No, you don't. Why are you even thinking about it? The right answer was no, I don't need to. I have a mission. If I can do it in private markets, I will.' **Direct alignment with the Koch Brothers' 'over my dead body' framing from this week's All-In.**
• **Jensen anecdote — midnight cold-call.** Sent Jensen a 2-line email earlier in the day; Jensen called back at midnight on the way home. Lesson Shiv took: 'fall in love with whatever the job is. You can convince yourself. You have to bend your DNA to CRISPR new DNA in.' Now lives on airplanes 5 days/week (San Francisco Mon-Wed, customers Thu-Fri).
• **Founder-mode framing differentiated from execs.** 'Founder mode is just about tours of duty. Depending on what's on fire, you go crush whatever that challenge is.' Doesn't mean micromanaging. Cross-references Chesky's [Issue 04](/issues/2026-05-10) 'AI founder mode' framing — Shiv operates it as task-rotation rather than blanket-detail-control.
• **'Always-on personal trade-off and Pittsburgh anchor.'** Wife + 3 kids in Pittsburgh because aging parents (dad has heart failure). 5 days/week on the road; Saturday morning walks with mum + Sunday with dad are non-negotiable. 'Folks who say you can have everything are lying.' **Powerful counter-perspective to the lean-ops 'work intensity' framing that dominates this issue and Issue 05.**
• **'Single biggest advice to a graduate: be full stack.'** If you're a marketer, be the one who creates content AND edits the video. 'Super-IC' as the future role. **Direct echo of Schoening from [Issue 04](/issues/2026-05-10) on agency and Koch Brothers' 'capability bounded' from this week's All-In.**

Notable quotes

Themes

• Five-year wilderness then $5.3B vertical-AI inflection (2018-2023 thesis durability)
• ~40% in-house models + distillation monthly = vertical-AI moat architecture
• Counter-positioning vs Microsoft/Nuance bundling (build in a way they can't)
• Talent over frontier-model access — 'always about the people'
• Founder-mode as tours of duty, not blanket micromanagement