Where venture capital flows, innovation follows. And for more than a decade, few faucets have been watched more closely than Y Combinator. An analysis of their investment patterns since 2020 doesn’t just reveal the accelerator’s strategy—it provides a map to the entire startup ecosystem’s next chapter.
With Demo Day approaching this week & inspired by Jamesin Seidel’s YC Series A analysis, I wondered how YC investment patterns have changed since 2020.
Who could have predicted that crypto and data center real estate would be the categories swinging the IPO market doors open?
In late 2024, I predicted a thaw in the IPO market. We’re now seeing that forecast come to life with CoreWeave and Circle’s IPOs. Neither company is pure-play software, but their strong performances signal renewed investor appetite for the ragged edge of technology.
Whenever I hear about a new startup, I pull out my research playbook. First, I understand the pitch, then find backgrounds of the team, & tally the total raised.1
Over the weekend, I decided to migrate this workflow to use AI tools, & the process taught me something important about how we’re actually integrating AI into our work.
Tools are small programs that expand AI capabilities. ChatGPT might call a web search tool to read a blog post I’d like to summarized. Claude might call the terminal tool to change file permissions in my current directory. Gemini might call a tool to find the latest stock price of the most recent IPO I’ve been following.
Which level do I want to use AI?
I find myself asking this question more & more frequently & I think the answer means at work I’ll be using many AIs - not just one or two.
| AI Level Use Case | Description |
|---|---|
| Chat-Based AI | Find the best Italian restaurant in the North Beach neighborhood of San Francisco. |
| In-App AI | Find a document or generate an overview paragraph within Notion. |
| Browser-Based AI | Deep research queries, such as estimating the market size of data center construction. |
| Computer-Based AI | Transcribe a video call and upload the notes to an investment memo. |
| Multiple AI Agents | Newer coding agents (e.g., Codex & Jules) work in parallel on the same codebase. |
Why are there so many levels? It depends on the context I want the AI to have.
Pricing changes are hard.
Fundamental shifts in go-to-market strategy tied to pricing? Monumentally difficult.
We recently dove deep into one such transformation with Barr Moses, CEO of Monte Carlo, during a Theory Ventures Office Hours. Monte Carlo, a data & AI observability pioneer, moved from traditional annual contracts to a daily revenue model.
These were the three most important takeaways for me from the conversation:
Monte Carlo customers were used to buying usage-based rather than contract-based & the alignment was an important & critical evolution. Ali Ghodsi said the annual contract is “selling like Oracle in the 1980s.”
NVIDIA announced earnings yesterday. In addition to continued exceptional growth, the most interesting observations revolve around a shift from simple one-shot AI to reasoning.
Reasoning improves accuracy for robots - like telling a person to stop and think about an answer before they reply. Here’s an example where I asked Gemini to create a financial projection for NVIDIA for the next five years.
With Salesforce announcing its intent to acquire Informatica & Google’s announced acquisition of Wiz, 2025 is the best year in the last six in terms of M&A value.1
These massive acquisitions, totalling more than $32b propel 2025’s year-to-date number to decade highs - exceeding the fast-money pre-Covid days of 2020.
Only five months in, 2025 has been the year of stablecoins. A Fireblocks survey of banks conducted in May underscores how quickly the market is moving.
Why are so many banks and payment processors aggressively moving to adopt stables?
We’re looking for an investor to join our team.
We are seeking people who see alpha in ambiguity, who are passionate about crafting theories about the future & making them a reality.
The ideal person :
If you’re interested, apply here.
“Now with LLMs, a bunch of the perceived quality depends on your prompt. So you have users that are prompting with different skills or different level of skills. And the outcome of that prompt may be perceived as low quality, but that’s something that is really hard to control.”
Loïc Houssier, VP Product at Superhuman, shared this perspective on a recent podcast. AI products differ from classic software in that the experience is in large part determined by the user.