Investors should target Vertical AI companies that specialize in high-stakes, regulated industries like healthcare, as these firms possess deep domain expertise and proprietary data that general models lack. Focus on companies like Abridge that are moving "closer to the flow of money" by automating high-value administrative tasks such as medical billing, revenue cycle management, and clinical documentation. Look for mature AI startups that are transitioning from expensive "Frontier Models" (like OpenAI or Anthropic) to fine-tuned in-house models to improve profit margins and reduce latency. While legacy platforms like Epic Systems dominate medical records, specialized "intelligence layers" that integrate directly into existing workflows offer superior growth potential through high user adoption. Prioritize investments in companies that emphasize data privacy and "trust-based" enterprise contracts, as these are better positioned to capture the massive $5.3 trillion US healthcare market.

Abridge (Private)

• Abridge is a vertical AI company in the healthcare sector, recently valued at $5.3 billion following a $300 million Series E funding round.
• The company spent five years (2018–2023) in a "wilderness period" before the current AI boom, focusing on the thesis that the "spoken signal" (conversations between doctors and patients) is the core data set of healthcare.
• Product Strategy: They use a "wedge" strategy, starting with automated clinical note-taking to solve doctor burnout, then expanding into medical ordering, billing, and revenue cycle management.
• Technical Approach:
  • Approximately 40% to 60% of their model outputs are generated by in-house models (fine-tuned open-source models).
  • They use "Frontier Models" (like OpenAI/Anthropic) for complex, ever-evolving tasks but switch to in-house models for high-frequency, high-stakes workflows to reduce latency and cost.
• Competitive Moat: Deep integration into regulated healthcare workflows and proprietary data sets. They position themselves as an "intelligence layer" on top of Electronic Medical Records (EMRs) like Epic.

Takeaways

• Vertical AI Resilience: Abridge proves that "being early is being wrong" can be overcome by persistence. Investors should look for vertical AI companies that survived the "pre-LLM" era, as they often have deeper domain expertise.
• The "Verb" Status: When a product name becomes a verb (e.g., "to Abridge a note"), it indicates significant market penetration and brand equity.
• In-House vs. Frontier: Abridge’s shift toward in-house models suggests that for mature AI companies, owning the model stack is essential for controlling the P&L and user experience (latency).

Vertical AI & Healthcare Sector

• Healthcare represents roughly 18-19% of US GDP ($5.3 trillion market), making it one of the largest opportunities for AI.
• The "Tsunami" of Need: The US healthcare system faces a massive shortage of clinicians and high burnout rates (40-50% of doctors; 30% of nurses considering leaving).
• Market Dynamics: Healthcare is not one monolithic market but a collection of segments (Integrated Delivery Networks, Academic Medical Centers, etc.). Success requires moving "up-market" to large care delivery systems.
• Data Challenges: Large healthcare organizations have messy, unstructured data. Vertical AI companies succeed by handling the "unsexy" work of data cleaning, compliance (HIPAA/HITRUST), and workflow integration.

Takeaways

• Deflationary Economics: AI is being positioned as a tool to make healthcare "cheaper, better, and faster" by automating administrative tasks (billing, documentation) that currently require doctors to work 30-hour days.
• Investment Theme: Look for companies that move "closer to the flow of money" (e.g., billing and revenue cycle) rather than just providing clinical tools.
• The "Trust" Barrier: In healthcare, the industry moves at the "speed of trust." Companies that prioritize data privacy and refuse to sell patient data (like Abridge) are more likely to win long-term enterprise contracts.

Frontier Model Providers (OpenAI & Anthropic)

• Mentioned in the context of their "forward-deployed engineers" and partnerships with private equity firms to help portfolio companies implement AI.
• Relationship with Startups: There is a tension between foundation models and vertical applications. Shiv Rao argues that if you are fighting foundation models, you have already lost; you must leverage their tailwinds.

Takeaways

• Platform Risk: While OpenAI/Anthropic can build "features" (like legal or medical note-taking), they struggle to build the deep, regulated "last mile" workflows that vertical AI companies provide.
• Token Consumption: Estimates suggest AI agents will increase token consumption by 24x over the next five years, benefiting the underlying infrastructure providers.

Legacy Systems (Epic Systems)

• Epic is the dominant Electronic Medical Record (EMR) provider. While they have their own AI features, they are viewed as a platform to build on top of rather than a direct competitor to specialized intelligence layers.

Takeaways

• Bundling vs. Best-of-Breed: Despite the threat of "bundling" (where a large incumbent like Microsoft or Epic adds a feature for free), specialized companies like Abridge are winning by focusing on a superior user experience and specific "spoken signal" data.

Investment Themes & Risks

• Founder Mode: The discussion highlights the importance of founders staying close to the "bare metal" of the business and performing "tours of duty" in different departments (HR, Sales, R&D).
• Talent Arbitrage: Access to top-tier researchers and engineers is considered more valuable than early access to frontier models.
• Risk Factor: The "A1" or "bridge" round (raising a small amount to stay alive) is a sign of a company in a "weak spot," though Abridge successfully navigated this to reach a multi-billion dollar valuation.
• Pricing Models: There is a shift/tension between "per-seat" pricing (preferred by CFOs for predictability) and "consumption-based" pricing (preferred by AI tech companies).