This is a weekly newsletter about the business of the technology industry. To receive Tanay’s Newsletter in your inbox, subscribe here for free:
Hi friends,
Cursor scaled to $100M ARR in just 12 months with a team of fewer than 20 people. Bolt reached $20M ARR in only 8 weeks. Midjourney surged to over $300M in revenue completely bootstrapped.
AI startups aren't just growing—they're shattering traditional SaaS growth records. But what's fueling this unprecedented acceleration?
My colleague
and I analyzed dozens of explosive-growth AI startups to uncover four distinct market segments where multiple startups are seeing this remarkable trajectory:Software-Creation Focused Products
Prosumer Creative Tools
First-Mover Enterprise AI Apps in Clear ROI Areas
Model Development and Deployment Infrastructure
Here’s how these categories have emerged, and what other founders and investors can learn from their success.
Software-Creation Focused Products
Coding has been widely recognized as the killer use case for AI models, and we’ve seen software creation focused AI startups scale at breakneck speed by deeply integrating into existing workflows and letting product value drive adoption in the product development process. Cursor, Windsurf, Bolt and Lovable are all hitting tens and hundreds of millions in ARR with teams under 50 by solving acute pain points (e.g. writing, editing, or generating code/apps) and offering great UX with minimal friction targeting either developers or those who want to create but hadn’t historically been developers.
Cursor went from zero to $100M ARR in just 12 months with under 20 people and no marketing, likely the fastest SaaS growth curve on record. It is now reportedly at $200M in run-rate. Unlike incumbents who bolted on AI features, Cursor was purpose-built for an AI-first coding experience. This AI-native orientation allowed it to move quickly and offer a cleaner UX than legacy IDEs, creating viral pull across the developer ecosystem.

The common theme is developers (and people that want to create but weren’t able to prior to AI) discovering, loving, and sharing these tools organically. Community (GitHub stars, Twitter demos, Discords), rapid iteration, and dead-simple onboarding drove viral growth without traditional sales.
Bolt hit $20M ARR in just 8 weeks by letting users build full-stack apps with a single prompt, expanding reach to non-technical creators. Even Cognition Labs’ Devin, positioned as the first AI software engineer, leveraged buzz, benchmarks, and exclusivity to drive thousands of inbound pilots pre-GA.
PLG Prosumer Creative Tools
Creative AI apps like Midjourney, Captions, Photoroom, and ElevenLabs have unlocked a viral loop: users create content with them, then share it across social platforms, turning every output into distribution.
Midjourney scaled to $300M in revenue in 2024 (up from $50M in 2022), entirely bootstrapped and without a sales team, by building a thriving Discord-native community and delivering stunning outputs that fueled organic growth.
ElevenLabs surged from $25M to $90M ARR in under a year, driven by its Creative Studio, which lets creators generate multilingual, lifelike voiceovers that power content creation across platforms.
Photoroom, now at $50M ARR with 200M+ users and 4M monthly downloads, has become essential for millions of SMBs and creators producing high-quality product visuals and social media content.
Heygen went from $1M to over $35M and flirted with profitability, all in the span of 18 months, helping users create avatar-based videos across a variety of use cases.
We see these companies as proof that prosumer AI can scale like SaaS, with stronger virality and lower CAC. Their growth is powered by community, shareability, and fast product iteration. They convert casual creators into power users with clear value, simple pricing, and constant delight. Their models scale well: Midjourney and Photoroom are profitable, Runway is layering in B2B and API channels, and Pika is turning its Discord-native community into a consumer video platform. The creative AI stack is one of the clearest opportunities for mass-market PLG in the AI era, and these companies are defining the playbook.
Similar to how AI helps democratize coding and expand the market of those who can build software, AI helps democratize video/image creation and editing and enables anyone to be creative, and these products have benefited from it.
If you don’t yet receive Tanay's newsletter in your email inbox, please join the 10,000+ subscribers who do:
First-Mover Enterprise AI Applications
First-mover applications like Harvey, Sierra, and Decagon have rapidly gained traction by targeting high-value enterprise workflows with innovative AI solutions in areas where copilots or agents have clear initial ROI and value.
Harvey has established itself as a leader in legal AI, securing firm-wide deployments at Allen & Overy and PwC, achieving over $50M in Annual Recurring Revenue (ARR) across more than 200 law firms, and raising $300M at a $3B valuation (with some of that contracted but not live yet).
Sierra, led by Bret Taylor, focuses on automating customer support using LLMs and has initiated numerous Fortune 500 pilots by offering seamless integrations and a compelling "replace Zendesk" value proposition. Similarly, Decagon competes in the same market and has also experienced significant growth.
A common thread among these companies is their swift establishment of enterprise trust. These markets are ones where AI is obviously going to be critical, and many enterprises are actively looking to “buy AI solutions”. These companies were first and early movers and established themselves as the leading and trustworthy product to buy in those categories. They recognize that in regulated and reputation-sensitive industries, delivering not only advanced AI models but also ensuring security, transparency, and enterprise-grade workflows is crucial.
Their go-to-market strategies emphasize credibility and relationship-building: Harvey leverages backing from OpenAI and references from prestigious law firms; Sierra utilizes founder-led go-to-market approaches and rapid integrations; and Decagon invests in domain expertise and compliance-ready solutions in an area where agents are showing clear ROI.
AI Model Development and Deployment Infra
Models have been the key unlock for AI and many startups have seen a ton of market pull in and around the model layer by either helping enterprises and AI applications make use of these models (deployment and inference) or sell to those that are training the models directly (labelling, GPUs, etc).
Together.ai, Modal, Fal.ai, Fireworks, Mercor, Baseten are some examples of companies scaling rapidly by building core infrastructure for training, deploying, and operating AI models.
Together.ai reached $100M+ in annualized revenue within two years by offering lower-cost, open-weight model training and inference—making it the go-to partner for startups looking for more control and lower infra costs.
Modal is building a serverless GPU platform with over 10,000 developers, focusing on developer experience and lightning-fast deploys—positioning itself as a Vercel for AI workloads.
Fal.ai went from $1M to $40M run rate in one year (And is now past $50M in run rate) by specializing in multimedia model inference, serving customers like Quora and Adobe.
Fireworks.ai brands itself as the fastest inference engine and has grew explosively to double-digit millions in run rate revenue, with customers including DoorDash, Notion, and Uber. These companies have grown fast by staying programmable, cost-efficient, and tuned for modern AI-native workloads.
Mercor, while adjacent, plays a critical role in powering human-in-the-loop model training. It built a global labor marketplace for prompt evaluators, data labelers, and QA workers—now serving OpenAI, Anthropic, and more. The company surpassed $100M in ARR by automating sourcing, vetting, and task routing, becoming the ops layer for high-quality data pipelines.
On the now public side, Coreweave also played into this trend, leasing GPUs on relatively flexible terms in a GPU-constrained environment growing from $20M to $2B in revenue in 2 years, with customers such as Microsoft, Mistral, OpenAI and NVIDIA.
What unifies this group is their focus on removing the core bottlenecks in the modern LLM stack—whether compute, inference latency, customization, or training data ops.
Hello Tanay,
I hope this communique finds you in a moment of stillness. Have huge respect for your work.
We’ve just opened the first door of something we’ve been quietly crafting for years—
A work not meant for markets, but for reflection and memory.
Not designed to perform, but to endure.
It’s called The Silent Treasury.
A place where judgment is kept like firewood: dry, sacred, and meant for long winters.
Where trust, patience, and self-stewardship are treated as capital—more rare, perhaps, than liquidity itself.
This first piece speaks to a quiet truth we’ve long sat with:
Why many modern PE, VC, Hedge, Alt funds, SPAC, and rollups fracture before they truly root.
And what it means to build something meant to be left, not merely exited.
It’s not short. Or viral. But it’s built to last.
And if it speaks to something you’ve always known but rarely seen expressed,
then perhaps this work belongs in your world.
The publication link is enclosed, should you wish to open it.
https://helloin.substack.com/p/built-to-be-left?r=5i8pez
Warmly,
The Silent Treasury
A vault where wisdom echoes in stillness, and eternity breathes.