Y Combinator Founder Paul Graham: Growth Supports Valuation
Y Combinator founder Paul Graham stated that AI startups are growing rapidly, which is the reason for their inflated valuations. He emphasized that it is not just because AI is a new trend, but because many companies are indeed delivering impressive numbers.
The public growth figures align with this statement. Anthropic's annualized revenue increased from about $5 billion in August 2025 to about $65 billion in July 2026. Cursor's revenue grew from about $500 million in June 2025 to about $4 billion a year later, and it was acquired by SpaceX for $60 billion in August 2026. Cognition's revenue rose from about $155 million to about $1 billion. OpenAI's annualized revenue increased from about $12 billion to about $40 billion.
The surge in valuations is tied to the surge in revenues. OpenAI completed a $122 billion financing on March 31, 2026, with a valuation of $852 billion. About eight weeks later, Anthropic raised $65 billion at a valuation of $965 billion, surpassing OpenAI, and submitted a confidential IPO application in June, rumored to target a valuation of over $2 trillion. Cursor was valued at $29.3 billion in November 2025, with an acquisition price of $60 billion.
Not every company has revenue. Mira Murati's Thinking Machines Lab completed a $2 billion seed round at a $12 billion valuation in July 2025, without disclosing products or revenue at that time. Ilya Sutskever's Safe Superintelligence was reported to have raised $2 billion at a $32 billion valuation in April 2025, with the company stating no product or revenue plans in the near future. Cognition raised $400 million at a valuation of $10.2 billion with an annualized revenue of about $73 million, a multiple of about 140 times.
In the first half of 2026, approximately $355.9 billion of the $412.7 billion in U.S. venture capital flowed into AI, accounting for 87.5%. Large transactions over $100 million made up the majority of the financing amount. The median post-money valuation for seed rounds rose from $16 million in 2024 to $24 million in the first quarter of 2026. AI Series D and later valuations surged about 6.6 times, while non-AI valuations increased about 1.6 times.
This is a primary market bid driven by growth expectations, not secondary market retail trading. Buyers include a16z, Accel, Thrive, Founders Fund, and sovereign and strategic funds, while sellers are founders, employees, and old shares from previous funds. Money is concentrated from large funds into a few companies with annualized revenues that can be rewritten monthly. Beneficiaries are programming, models, and vertical applications with existing paid workflows; those under pressure are labs without revenue priced at similar multiples, and early revenues eaten away by reasoning costs and manual support.
Source: Public information
ABAB AI Insight
Paul Graham founded Y Combinator in 2005, initially selecting companies like Airbnb, Stripe, and Dropbox based on weekly growth rates rather than stories. Stripe demonstrated growth through payment volume, while Airbnb proved growth through booking amounts, both raising valuations only after revenues became visible. He applies the same standard to AI this time: numbers precede trends. In Y Combinator's own batches, programming and agency companies are directly aligned with this post by investors.
The flow of money splits into two paths. One goes to companies with existing annualized revenues, where Anthropic, OpenAI, and Cursor use financing amounts to cover prepayments for computing power and sales expansion. The other goes to labs without products, such as Thinking Machines' $2 billion seed round and Safe Superintelligence's $2 billion, buying the founders' previous positions at OpenAI and Anthropic. Cognition managed to turn about $73 million in annualized revenue into a $10.2 billion valuation; the growth is real, but the multiple is not a number that would pass the 2019 software memo.
This differs from the internet's eyeball valuation in 1999 and the software's valuation of 20 to 40 times annual recurring revenue in 2021. We are currently in an expansion phase: revenues can grow from $500 million to $4 billion within a year, and valuation jumps are seen as prepayments for growth rather than pricing for steady profits.
Structurally, this represents a transfer of pricing power. Previously, valuations were determined by comparable companies and revenue multiples; now they depend on whether the next round is willing to accept higher growth rates. Model companies control interfaces and usage, while application companies carve budgets from salaries and outsourcing through workflows. Reasoning costs remain below gross margins, so astonishing revenues do not equate to astonishing profits, and pricing power temporarily resides with those few companies that can consistently deliver growth.
ABAB News · Law of Cognition
High valuations are often because growth first rewrites the denominator.
Revenue can prove growth but not gross profit.
High prices without products buy previous positions.