Inside this edition:
How Thinking Machines $2b seed makes IPOs look like Series A rounds
What the lab started by former OpenAI CTO Mira Murati just revealed it's working on
Why Renaissance Capitalist readers won’t be surprised by it (hint: The AI that is Seen, and the Deepfake Paradox posed by Stalin’s Photoshop that’s Unseen)
Why making AI deterministic is worth more than money to society
TL;DR: They're making AI reproducible, which sounds boring until you realize what is unlocks, and the Orwellian nightwares it may help us avoid.
Most up-to-date version with 900% less typos at renaissancecapitalist.beehiiv.com.
Most people wouldn’t turn down a billion dollar job offer if it were priced in Yen.
Former OpenAI CTO Mira Murati allegedly rejected one priced in USD from Zuckerberg during Meta’s Poachapalooza.
But her new AI lab, Thinking Machines, helps give that $1bn a sense of proportion.
Her lab raised a seed round of $2bn, valuing it at $12bn in July with a promise of revealing what it was working on soon.
They've now shared what they're working on. And if you've read prior posts, you'll be very familiar with what they're trying to solve, while others scratch their heads before returning to their infinite scrolls.
(which is why this publication isn't for everybody, but for intellectually curious builders, funders, and thinkers who care about the world we're building - and hopes to serve as a serendipity machine for those accelerating Renaissance ideals into our desired future).
Before getting to the great reveal… let’s give that IPO-sized seed, and IPO-sized valuation, a sense of proportion.
Thinking Machines (TM) record breaking seed was, according to Jay Ritter’s IPO research:
Nearly 7x the average raise of US tech IPOs in the first half of 2025
The average US tech company going public in H1 2025 raised $300M. TM raised $2bn at seed.
USD billions, 1H2025 US tech IPOs
Equal to one-third of an entire IPO market
TM's $2b represents 28% of all money raised by US tech IPOs in H1 2025, which totalled just over $7bn. (Or if you prefer: 36% of all tech IPO proceeds for the entire year of 2024)
Achieved in 1/33rd the time
The median tech company waits 14 years to go public.
TM reached 7x the scale of their IPO raise five months after founding.
Find this post with 900% less typos here and https://renaissancecapitalist.beehiiv.com/
If you've read this publication, you already have a sense of the problem that Thinking Machines' research last week makes headway on solving.
In my piece Stalin’s Photoshop, I explained the deep fake paradox that we (& tech) must solve.
And in my earlier piece "The AI that is Seen, the Crypto that is Unseen", I used Bastiat’s framing to describe AI as the "seen" (probabilistic, infinite versions of truth) and how crypto may be the "unseen" (deterministic, singular proof).
Thinking Machines' new research tackles (part) of this exact headache from a novel perspective.
The problem in plain English:
Ask ChatGPT the same question five times, get five different answers. Not because it's creative, but because of how GPUs (the chips running AI) randomly batch requests together. Imagine a chef making your order differently depending on who else is in the restaurant at that exact second. That's current AI, or LLMs more specifically here I ‘spose.
Their fix:
Make the GPU kernels (tiny programs inside Nvidia chips) batch-invariant - gigabrain for making sure your answer doesn't change based on who else hit the server at the same millisecond.
In the simplest terms: Making sure the same prompt gets the same answer, every time.
Why this matters:
For enterprises: Reproducible outputs mean auditable AI decisions (finally)
For scientists: Experiments you can actually replicate (incredible)
For AI training itself: Cleaner data for reinforcement learning (the process of rewarding correct answers gets messy when "correct" changes more frequently than the release date of Epstein files)
When you can trust that the same input produces the same output, you can finally build deterministic systems on top of probabilistic models.
Sound familiar?
It's the technical implementation of what I've been arguing: the marriage of infinite possibility (AI) with verifiable truth (determinism).
PS leaving NYC and already met with an incredible group of founders, public intellectuals, and public (investing) intellectuals.
If this or the challenges of democracy that tech might solve hit a nerve, hit me up before we head to Balaji’s Island off Singapore for the network state conference.
If you enjoyed this post or know someone who may find it useful, please share it and encourage them to subscribe here or at https://renaissancecapitalist.beehiiv.com/
WHO IS THE RENAISSANCE CAPITALIST?
Part adventure capitalist, part librarian — Daniel Attia is a (venture) investor & builder who writes in the third person and backs founders reinventing reality through preemptive.
(portfolio below)
His lens comes from a random sprint through high finance, startups, tech & media, venture, hedge funds, and the arts.
He mastered capitalism's grammar at Deutsche Bank and Goldman Sachs before being force fed its real-life principles as the first US hire at Payapps (acquired by Autodesk for ~$500M).
Daniel would later shape tech and market perspectives as founding Head of Research at Prof G Media, contributing to works like the NYT Bestseller "Adrift: America in 100 Charts."
His favorite capitalist pastime may be steering companies away from entropy toward rationalism (often mislabeled as "shareholder activism"), partnering with hedge funds, families, and shareholders who've grown weary of watching their capital fund executive delusions.
Today, he serves on the Foundation Council at the State Library of Victoria—the world's third busiest library—while moonlighting as consigliere to founders and CEOs at pivotal crossroads.
Daniel also serves as Special Advisor to VP Capital, a HK based hedge fund.
Daniel co-founded Pew Pew NYC, a non-profit art collective for the creatively curious
(which just unveiled Call me Lola in Mexico City, the first live-in art gallery hotel experience where 70% of art sales flow directly to artists!)
Select venture investments:
beehiiv (this very platform) – Because Tyler Denk always had Big Desk Energy
SymphonyOS – beehiiv for artists (Business Insider Top 13 Creator Startups to Watch, just like beehiiv. Led seed alongside Tyler too)
Harmonic Discovery – Precision pharmacology (JP Morgan Life Sciences Award winner)
Carry – Putting tax optimization on autopilot, built by Ankur Nagpal who turned his $250M exit lessons at 32 into your tax solution
Measured – Medicine minus middlemen led by dreamer and DREAMer immigrant Monji Dolon (seed with Initialized Capital)
True3d — Building livestreaming infrastructure for 3D by Meta livestreaming veteran Daniel Habib (still can’t believe we got an investor mention alongside YC founder Paul Graham)
EatBlueprint by Jeff Tang (now merged with Bryan Johnson’s Blueprint)
Atelier – Making manufacturing magnificent again (Co-led Series A with Macquarie Capital)