In partnership with
GM. This is Milk Road AI, the newsletter that realized the writer was the bottleneck. So we’ve decided to upgrade. 😉
Big news for our readers and X addicts: Melvin, the AI analyst who’s been writing this newsletter since day one, is officially moving full-time to the Milk Road AI X account, where he’ll keep sharing his deep dives and analysis. And maybe get more posts shared by Elon. 😏
But don’t panic. The newsletter isn’t going anywhere (and neither is the bull run, btw). You’ll just read slightly fewer historical analogies now (none, to be precise).
Here’s what we’ve got for you in the post-Melvin era:
- Before you invest in the SpaceX IPO (or any of the upcoming ones), you need to understand why our AI analyst (the other one) is avoiding them.
- Plus, he shares what new bottleneck company he’s buying right now, and actually sat down for a full-length pod to break it all down.
Speaking of infrastructure, today’s partner is Arc by Circle.
Arc is an open Layer-1 blockchain built for real-world financial applications. Public testnet is live, App Kits are here, and mainnet is getting closer. Start building on Arc.
THE BIGGEST AI OPPORTUNITIES AREN’T IN THE IPOS 🥶
We’re amidst the biggest IPO wave since the internet:
- SpaceX had the biggest IPO in history on Friday.
- Valued at $965B, Anthropic has filed their S-1.
- OpenAI is eyeing September for an IPO.
But our AI analyst, Vincent, is passing on all of the IPOs.
(This thesis comes straight from Vincent's latest Milk Road PRO report.)
Here’s why. 👇
To understand it, you need to understand the 5-layer AI cake:
- Layer 5 (the top): Applications. The ChatGPTs & the Claudes. The stuff normal people interact with.
- Layer 4: Models. The actual AI brains. GPT-4, Claude Fable 5. The intelligence itself.
- Layer 3: Infrastructure. Data centers, cloud servers & networking.
- Layer 2: Chips. The GPUs doing the actual computing.
- Layer 1 (the bottom): Energy. Electricity, cooling, power. Nothing in the cake works without it.
Bottom to top, energy powers the chips, chips sit inside the infrastructure, infrastructure runs the models and models drive the applications where the economic value finally shows up.
Source: Vincent’s latest PRO report
Every IPO everyone is buzzing about lives in Layers 4 and 5.
But the bottleneck is buried at the bottom of the cake.
Think of it like a highway. All the hype is about the cars on the road (Layer 4 & 5) but the road itself isn’t built yet (Layers 1-3).
- Only 5% of the GPUs already shipped are actually running.
- The other 95% are sitting idle. The physical infrastructure to power them doesn't exist yet.
- Over 60% of data centers planned for 2027 haven't started building yet.
- Grid connection queues in the U.S. now run 3 to 5 years.
- Large power transformers have 18- to 24-month lead times.
So while everyone queues up for IPO allocations at the top of the cake, the more interesting question is what you buy at the bottom - in the companies solving the power and efficiency problem that's choking everything above it.
Both of our AI analysts (Melvin & Vincent) have been investing in these bottom layers and they have multiple picks that have run over 100%:
- Credo (CRDO): 134%.
- Micron (MU): 110%.
- Nebius (NBIS): 88%.
Don’t miss the next call, come join us for just a buck.
Source: Milk Road PRO
Looking specifically at Micron, Vincent thinks that the easy money in AI hardware (Bloom Energy, Micron) is done.
(For context, Micron is still held by Melvin.)
The AI economy is in a state of bottlenecks. You can’t just build your way out of this quickly because building data centers takes time.
So here’s how Vincent is approaching the AI market right now:
“Buy everything related to efficiency. More output per watt.”
Vincent thinks the market is sleeping on this narrative and he’s even got a pick:
Qnity Electronics.
This is a $30B spinout from DuPont (the 200-year-old American chemical giant). They make the specialty chemicals and advanced materials that go into chip production. Specifically, the stuff that keeps high-density racks from melting down under load.
When you're packing 600kW to 1MW of computing power into a single rack, the thermal management problem becomes very important.
You need materials that can handle that kind of heat without degrading. That's Qnity's lane.
Their client list is nothing but impressive:
- Samsung.
- TSMC (the world's largest chip manufacturer).
- Apple.
Here’s the kicker:
They have an active R&D collaboration with NVIDIA (and… they haven’t even gotten the Jensen Huang “nod” yet).
The “nod” is when Jensen publicly highlights a company, investors pile in and the stock takes off (Corning and Marvell are two recent examples).
Qnity hasn't gotten that nod… yet.
But given its relationship with NVIDIA and its position in one of AI's biggest bottlenecks, Vincent believes it's a name worth watching.
If you wanna hear about Vincent’s take on Qnity, he breaks it down in this episode:
Hiring in 8 countries shouldn't require 8 different processes
This guide from Deel breaks down how to build one global hiring system. You’ll learn about assessment frameworks that scale, how to do headcount planning across regions, and even intake processes that work everywhere. As HR pros know, hiring in one country is hard enough. So let this free global hiring guide give you the tools you need to avoid global hiring headaches.
