Mirendil at $5B with no product: read AI valuations right
Mirendil is in talks at a $5B valuation three months after a $200M seed, with no shipped product. How to read AI vendor valuations when you pick a stack.
A five-times valuation jump in ninety days tells you something about capital markets. It tells you almost nothing about whether the software works. Bloomberg reported on September 22 that Mirendil — founded in December 2025 by former Anthropic researchers — is in talks to raise up to $1 billion at a $5 billion valuation. The company has no public product. Reading AI vendor valuations correctly is now part of picking a stack, so here is how we do it.
What actually happened
Mirendil closed a $200 million seed in June at a $1 billion valuation, co-led by Andreessen Horowitz and Kleiner Perkins with Nvidia participating. Kleiner is reportedly leading the new round, with a16z in discussions again.
The company is building what CEO Behnam Neyshabur frames as AI for AI research: models that design experiments, evaluate results, and run the next training round with progressively less human input. Neyshabur spent more than five years at Alphabet co-leading reasoning research for Gemini before joining Anthropic in late 2024. A frontier model aimed at engineering and research work is planned for early next year.
Three months. Same two lead investors. Five times the price. No shipped product in between.
Why AI vendor valuations matter for your business
We are not going to tell you this is a bubble, and we are not going to tell you it isn't. Neither claim helps you ship anything. Here is the part that does.
A valuation prices the option, not the product. Investors at this stage are buying a probability-weighted shot at a very large outcome. That is a rational thing for them to do with a fund. It is a terrible input for your build-versus-buy decision, because you are not buying an option — you are buying something that has to work on Tuesday.
Round size predicts survival better than valuation does. The $200 million in the bank is the number that says a vendor will still answer support tickets in eighteen months. The $5 billion headline says investors expect enormous growth, which eventually arrives at your invoice. When we assess a vendor for a client, we look at capital raised and burn, not the multiple.
Track record moves to the founders when the product does not exist yet. With no shipped software, the only real signal is who is building it and what they built before. Neyshabur's Gemini reasoning work is a genuine credential. It is still a credential, not a roadmap, and your migration plan cannot be someone's résumé.
The practical filter we use with clients is three questions. Is there a product we can run against our own data today? Is there a documented export path? Can we swap this layer by config if the pricing changes? A vendor that fails all three is a research bet, and research bets do not belong in production workflows — they belong in a pilot with a fixed budget and an end date.
Key takeaways
- Mirendil is in talks for up to $1B at a $5B valuation, per Bloomberg — three months after a $200M seed at $1B
- Kleiner Perkins is reportedly leading; a16z co-led the seed, with Nvidia participating
- The company builds AI that automates AI research; a frontier model is planned for early next year
- No public product exists at the time of the reported talks
- Valuation prices an option; capital raised and burn rate are the better survival signal
- Filter vendors on three things: run it against your data today, documented export path, swappable by config
- Pre-product vendors belong in a fixed-budget pilot, not in a production workflow
Vendor diligence is a half-day, not a quarter. Rush Commerce builds automation with the model layer swappable by config, so a vendor's funding round never becomes your migration. Tell us what's in your stack and we'll tell you what we'd replace first.
- #ai-funding
- #vendor-risk
- #valuations
- #procurement
- #small-business
Tommy Rush — Founder, Rush Commerce
Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More
Get The Rush Report weekly — one email, zero fluff.
Keep reading
Huang: don't ship agents you can't contain, liability follows
Nvidia's Jensen Huang told Ezra Klein that labs shipping uncontrollable AI face civil and criminal liability. The same containment test applies to your deployment.
Read itVerizon's $70M AI training is free for small businesses
Verizon put $70M behind free AI training for small businesses and job seekers, with coursework from Anthropic, OpenAI, Google, Microsoft and IBM. Here's how to actually use it.
Read it