I dropped out of high school at 17 to build my first startup.
Now, at 20, I'm thrilled to share that @TryOmenAI has raised $41.5M in total funding to build intelligence protecting the world's most critical AI compute infrastructure.
I have a major result to share.
I’ve found conclusive forensic evidence that the top 4 Chinese AI labs—DeepSeek, Moonshot, MiniMax and Ant Group—have built their newest models on NVIDIA’s most advanced Blackwell chips, which are barred from export to China.
This sounds insane, but I discovered that almost every AI model carries a hidden signature of the GPU that trained it. When a model is first set up, the starting values of its weights are laid out in a pattern that depends on the exact GPU. Each combination of GPU and seed number produces a unique pattern—a sort of fingerprint of the hardware used to create the model. Traces of that pattern survive training into the published model.
Again, this evidence isn’t based on any insider information or leaking. It’s based on a forensic analysis of the model weights released by the companies themselves.
I searched more than six trillion combinations of chip layout and seed number and matched them to open-weights models. In doing so, I found conclusive evidence that six Chinese AI models—including DeepSeek-V4-Pro, Kimi K3, and MiniMax-M3—were initialized on NVIDIA Blackwell systems. Four of the models match Grace Blackwell, the GB200 and GB300 parts in NVIDIA’s flagship liquid-cooled rack-scale systems.
I then verified these results on real hardware. Running unmodified PyTorch with the seeds I’d recovered, a Blackwell B300 reproduced DeepSeek-V4-Flash’s untouched starting weights bit for bit, and a rented GB200 reproduced Kimi K3’s to a correlation of 0.999995. A Hopper H100 reproduced none of DeepSeek-V4-Flash’s, and changing the seed by one broke the match. (DeepSeek’s seed is 953049583 by the way.) The method also was verified to match 24 known open-weights models to the hardware that developers said they used.
It is exceedingly unlikely that these results are a coincidence. Across six trillion combinations, the best accidental match should reach only about 8 sigma (i.e., eight standard deviations above what random chance would produce). The real matches range from 49 to 1,385 sigma, and even the weakest single test hits 20.
These were data-center-scale training runs. With some analysis of release schedules and model details, I estimate that the largest, Kimi K3 and DeepSeek-V4-Pro, used 2,000 to 3,000 Blackwell GPUs each. Clusters of this size cost on the order of $150 to $200 million each to buy, and across these labs alone would add up to more than half a billion dollars of export-controlled GPUs, not counting experimental compute or otherwise. It’s quite likely that we see multi-billion dollar expenditure on export-controlled GPUs.
I am highly confident in this result, and it has been independently verified by senior researchers at OpenAI and Anthropic. You can read the full report at https://t.co/YOtzND0Jqk.
There is much more I’ve identified about these models through other forms of checkpoint forensics like this, and will have more to share soon.
We’re cooking over here at @TryOmenAI. Building out our teams in OC & SF!
If you want to work at the intersection of physics, fluid dynamics, high voltage chemistry and more…. reach out!
Met @lukedwagner12 & @MainwaringLuke on day 1 thanks to @FounderLaberge. Pareto Fellows, then YC, now launching @neomatterai.
Every deal, every hire, every vendor ends up in legal’s inbox. They built the OS for it. Lawyer stays in the loop, everything else gets done.
Congrats guys - to the moon!
Today we’re launching @neomatterai (YC F26), the operating system for in-house legal teams.
Neomatter picks up legal requests wherever they come from (email, Slack, Salesforce, etc) and does all the work. Then the lawyer reviews, approves, and sends.
We thought legal could use some change. Book a demo and see it for yourself!
The wrong co-founder leaves you 50% as good as you could be.
@FounderLaberge never had a co-founder.
Instead, he taught himself spectroscopy and raised $41.5M to put sensors inside Al data centers (at 20).
Solo Founders Podcast ep 25 is live with Zach Laberge of @TryOmenAI.
16:51 Shutting down his first company at 18
22:05 $3M raised in three days
26:11 Engines don't break - hydraulics do
33:51 "You're an idiot": six months early on data centers
37:55 Solo by default
45:13 The greatest fundraising market in history
52:58 The bear case for solo founding
59:05 Why no co-founder made him 10x better
Liquid-cooled data centers are starting to see issues with water-based coolants becoming contaminated and degrading nearby infrastructure.
"Normally it's a water-based fluid with some sort of mixture of either ethylene glycol or propylene glycol. You're seeing a big shift toward safer fluids — propylene glycol is preferred, since it's less toxic."
"A bunch of different things can happen in the system. You see a lot of basic infrastructure wear. So when your pumps and seals wear, all of that stuff starts to contaminate the liquid itself."
"What also happens is the more heat [there is], the more the fluid starts to break down. So you go from this very nice, high concentration of glycol to water."
"Then that starts to degrade and you start to get biogrowth, and other crap that gets in there. And that starts to eat your chips and sit in your pipes."
@founderlaberge explains:
I dropped out of high school at 17 to build my first startup.
Now, at 20, I'm thrilled to share that @TryOmenAI has raised $41.5M in total funding to build intelligence protecting the world's most critical AI compute infrastructure.