A 3-person+AI Agent team claims their AI hit $30M ARR
Investing in growth-stage AI startups is getting riskier
Another 3 big ARR milestones worth noting I tracked last week, they are Vercel, Apollo and WorkOS.
Vercel ARR crossed $200M
Vercel is a cloud platform as a service designed specifically for frontend developers to build, deploy, and scale web applications and websites quickly and efficiently.
It provides a "frontend-as-a-service" product that simplifies the deployment and management of the user-facing parts of applications, separating frontend deployment from backend infrastructure.
Vercel has crossed $200 million in ARR as customers like OpenAI, Runway, Granola & more flock to its web development/hosting service.
Apollo crossed $150M in ARR
Apollo.io is an end-to-end AI-powered sales platform designed to help businesses grow by streamlining their sales prospecting, engagement, and pipeline management processes.
It provides a comprehensive B2B database with over 210 million contacts and 35 million companies, enabling users to find accurate and verified contact information such as emails and phone numbers.
Now 1M+ GTM professionals use Apollo to book meetings, fill pipeline, and grow their companies. And its ARR just crossed $150M.
WorkOS ARR crossed $20M
WorkOS is a developer-focused identity platform designed to make Business-to-Business (B2B) SaaS applications "enterprise-ready" quickly and efficiently by providing essential enterprise-grade features through unified APIs.
It enables SaaS companies to integrate complex enterprise identity and access management (IAM) capabilities without building them from scratch, thus accelerating their ability to sell to large organizations.
It is now powering 1,000+ customers and has crossed $20m ARR.
Trend: Investing in growth-stage AI startups is getting riskier
A couple of recent issues may prompt AI startups to reassess their reliance on foundation model companies and their business boundaries—because these foundation model companies have no defined boundaries.
If you’re operating in a promising space, once you reach a certain level of success, there’s a high chance the foundation model companies will either build their own version or take actions that actively harm you.
The AI coding sector is a prime example. After OpenAI reportedly offered to acquire Windsurf for $3 billion, Anthropic drastically cut Windsurf’s first-party access to the Claude 3.7 Sonnet and Claude 3.5 Sonnet models.
What’s more, Anthropic’s newly released Claude 4 models have entirely ignored Windsurf, forcing the company to rely on a more expensive and complex workaround for access—increasing costs for developers. Windsurf CEO Varun Mohan stated that Anthropic’s measures were taken almost without warning, resembling a “supply cut.”
Anthropic cofounder and chief scientist Jared Kaplan responded bluntly, saying that due to rumors of OpenAI acquiring Windsurf, they decided to drastically restrict Windsurf’s API access to Claude 3.x models in order to ensure limited compute resources go to long-term partners.
“We just want to support customers who will continue to work with us in the future. I think selling Claude to OpenAI is weird for us.”
OpenAI did the same thing, likely. Its recent rollout of AI-powered meeting notes and connector features directly targets startups like Granola, an AI note-taking tool, and Notion’s AI initiatives.
Granola, which recently raised a 43 million Series B at a 250 million valuation, has been pivoting from a “second brain” for individuals to a “collective brain” for enterprises, emphasizing data storage and team collaboration to avoid being just an OpenAI reseller.
Meanwhile, Notion has integrated Granola-like features directly, turning AI meeting notes into just another feature. Now, OpenAI has gone even further—its ChatGPT product not only replicates these AI note-taking functions but also uses connectors to integrate with third-party tools like Notion, challenging their position.
For Granola, its value proposition—data storage and collaboration—now faces intense pressure from ChatGPT, which may offer even greater value in both respects. Many AI products may soon face similar challenges.
This trend has led some investors to warn that AI growth-stage investing has entered a high-risk era. Cathy Gao, Partner at Sapphire Ventures, says:
“Investors now must go beyond near-term revenue and ARR. They need foresight to assess whether a company can withstand competition from tech giants and whether it can build a ‘moat’ in a narrowly defined market segment. The current sweet spot is big-enough-but-not-big-enough-to-attract-the-giants niche markets.”
She believes future winners will be teams that can demonstrate real growth trajectories, sustainable revenue models, and avoid head-on clashes with big players.
As a result, investors are increasingly exploring AI-driven roll-up strategies. Not only VCs but also solo investors like Elad Gil are adopting this strategy. He has invested in two companies using this approach, including Enam.Co, which reached a $300 million valuation within just a year. The company focuses on boosting worker productivity.
Gil’s idea: acquire mature, labor-intensive firms—like law firms or other professional service providers—and scale them using AI. Then, use the improved profit margins to acquire more similar companies and repeat the cycle.
He says he’s been doing this for three years. While roll-ups existed a decade ago, they were mostly used to increase valuations without substantial business gains. In contrast, today’s AI-driven approach fundamentally changes the cost structure of these businesses.
So, for AI applications with low defensibility but strong cash flow, traditional fundraising may not be the optimal path—and for VCs, the risks are significant.