Will the 10× Growth Promise of Vertical AI Crumble as Generalist LLMs Like Manus Dominate the Market?

Background and Question

Recent breakthroughs from foundation models—Claude Sonet, Qwen, DeepSeek V4, and the viral product Manus—have raised expectations for generative AI applications. Bessemer Venture Partners (BVP) predicts that Vertical AI’s market capitalization will be at least ten times that of legacy Vertical SaaS because it can tap the services economy with new business models. The author asks whether this 10× growth belief will collapse now that general‑purpose models dominate the charts.

What Is Vertical AI?

Vertical AI refers to LLM‑native applications built for specific industries, scenarios, or functions (e.g., legal, healthcare, finance). Unlike generic AI solutions, Vertical AI tailors workflows, data, and domain knowledge to solve high‑value, repetitive tasks that traditional SaaS cannot address cost‑effectively.

Why Vertical AI Is Needed

Traditional SaaS struggles with high‑cost customization for tasks such as contract review, medical record entry, and financial analysis. Vertical AI can leverage LLMs to automate these high‑ROI, high‑value tasks, delivering greater efficiency for professionals.

Where the Opportunity Lies

BVP cites U.S. Bureau of Economic Analysis data: software spending was ~1 % of GDP in 2023, while Business and Professional Services accounted for ~13 %, indicating a massive TAM for AI‑enabled services. Vertical AI can enter markets that legacy SaaS cannot, especially where ROI is currently too low for adoption.

Current Growth Drivers

Rapid advances in generative AI (e.g., ChatGPT launch in 2022) provide technical feasibility.

Early Vertical AI companies have achieved ~80 % of traditional SaaS ACV within a few years, with ~400 % YoY growth and ~65 % gross margin (BVP).

Key Challenges

Complex, chaotic industry workflows – Legal, medical, and financial processes require deep domain expertise and extensive data cleaning, making direct use of foundation models difficult.

Foundation models do not yet “understand” industries – Effective AI solutions need intent extraction, context gathering, verification, and expert evaluation, which exceed raw model capabilities.

Implementation Guidance

The author recommends a “first‑easy‑then‑hard” rollout: start with AI‑assisted high‑frequency, high‑value tasks (e.g., contract analysis) before moving to fully autonomous workflows.

Harvey’s methodology—“Expand firstly and then Collapse”—creates 30‑50 reusable AI patterns (e.g., case law research, regulatory review) that are later composed into end‑to‑end workflows such as international M&A due diligence.

Moats for Vertical AI

Four dimensions of moat are outlined:

Data moat – Proprietary industry data and high‑quality domain datasets bridge the “Massive Delta” between foundation model capabilities and real‑world performance.

Technical moat – Model fine‑tuning, multimodal processing, modular infrastructure, and engineering speed give a competitive edge.

Product moat – Deep integration into core workflows, trust, reliability, and accuracy (especially in regulated sectors) differentiate true Vertical AI from simple LLM wrappers.

Business‑model moat – Innovative pricing (token‑based, outcome‑based, subscription‑plus‑value), copilot, agent, and AI‑enabled service models, and strong customer relationships (e.g., Palantir) sustain long‑term advantage.

Emerging Business Models

Copilots – AI assistants that work side‑by‑side with users (e.g., GitHub Copilot, Harvey).

Agents – Fully automated workflows with minimal human intervention (e.g., Relevance AI, LinkedIn Recruiter Assistant).

AI‑Enabled Services – Companies that sell AI‑automated services at traditional service‑provider pricing (e.g., EvenUp for legal services, SmarterDx for medical billing).

Pricing can combine usage‑based, outcome‑based, subscription, or hybrid models to align value creation with revenue.

Conclusion

While general‑purpose LLMs raise the bar for AI expectations, the author argues that Vertical AI remains essential because it addresses domain‑specific knowledge gaps, complex workflows, and high‑value use cases that generic models alone cannot solve. Success will depend on building data, technical, product, and business‑model moats, and on integrating AI deeply into industry processes.