GM. This is Milk Road AI, built for the people who want to understand what's happening in tech before their group chat does.

​Here’s what we’ve got for you today:

Nexo is back in the U.S. - and new clients get 30 days of Wealth Club Premier perks! Higher yields, lower borrowing rates, and crypto cashback - start here.

THE KID WHO GOT FIRED AND CALLED THE FUTURE

In 1956, a young mathematician named John Tukey walked into Bell Labs and told his colleagues that the most important invention of the next century wouldn't be the hydrogen bomb, the transistor, or the space rocket.

It would be the computer.

They laughed at him, and these were serious men at one of the most serious research institutions on earth. 

They had built the transistor and pioneered information theory, they knew the cutting edge, and this wasn’t it.

Tukey wasn't wrong, but he was just ahead of schedule.

The pattern shows up everywhere, once you start looking.

In 1977, Ken Olsen, founder of Digital Equipment Corporation, one of the most respected technologists alive said there was "no reason anyone would want a computer in their home." 

Three years later, Apple went public at a $1.8B valuation.

In 2007, Steve Ballmer looked at the original iPhone and laughed on camera. "Five hundred dollars? Fully subsidized? With a plan? That is the most expensive phone in the world, and it doesn't appeal to business customers because it doesn't have a keyboard."

The iPhone sold 6M units in its first year while Ballmer's phone, the Zune, sold 2M total and was discontinued four years later.

The people who are wrong about transformative technology don't usually sound wrong in the moment, and they sound completely reasonable but the people who are right tend to sound a little unhinged.

Which brings us to Leopold Aschenbrenner.

He's 24 years old, who had a 4.18 GPA from Columbia, where he enrolled at 15 and graduated at 19 as class valedictorian with a double major in economics and math-statistics.

He got fired from OpenAI for saying the building was on fire.

Then he wrote 165 pages explaining exactly why, raised $225M, and built the most concentrated, most aggressive, most interesting bet in public markets right now.

The guy who read the internet and went to war

In April 2024, OpenAI fired Leopold Aschenbrenner.

The official reason was a leak, OpenAI said he shared confidential material, while he said it was just a harmless brainstorming doc.

But here's the messy detail everyone glosses over: two months before they fired him, he had sent the board a memo.

The memo said OpenAI's security was "egregiously insufficient" and that foreign adversaries could steal the model weights, the actual AI brain and nobody would notice until it was too late.

OpenAI HR's response was to tell him his concerns about Chinese espionage were "racist and unconstructive,” and then they fired him.

Two weeks later, Ilya Sutskever left, then Jan Leike, the co-lead of the entire Superalignment team, left and publicly said safety was no longer the priority. 

The whole Superalignment division was eventually dissolved.

From the outside, it looked less like a routine HR action and a lot more like a house on fire.

So Aschenbrenner went home, wrote 165 pages, and blew the roof off.

The essay that accidentally became an investment thesis

In June 2024, he published a free essay series called Situational Awareness: The Decade Ahead.

It went everywhere, AI researchers debated it, and Silicon Valley printed and dog-eared copies.

The core argument was uncomfortably simple.

GPT-2 was a preschooler, GPT-3 an elementary schooler, and GPT-4 a smart high schooler.

One more jump of the same size gets you to AGI.

He called this counting the OOMs, a framework that maps out how much compute is improving, how much algorithmic efficiency is improving, and how much just deploying AI as autonomous agents ("unhobbling") unlocks on top of all that.

Put it all together, and this isn’t a better chatbot, it’s AI doing the work of a smart engineer, a thousand times over, for pennies.

And then the superintelligence follows fast, because the first AGI starts designing its own successor.

The final punchline was his warning: the U.S. is entering a race for the most powerful technology ever built, and it’s securing the lab behind it with the same urgency as a startup Slack workspace.

That wasn’t just a prediction, but rather he positioned for it.

In September 2024, he launched Situational Awareness LP, and it started with $225M, and by December 2025, it had $5.3B.

That’s roughly ~24x growth in 15 months from a 4-person team with no prior investing track record, run out of a San Francisco suite by a 23-year-old who was vibe-coding in his bedroom two years earlier.

Here’s where it gets interesting, his bet isn’t Nvidia, Microsoft, or any of the names you’d expect.

THE KID WHO GOT FIRED AND CALLED THE FUTURE (P2)

Everyone plays AI through the obvious names.

Buy Nvidia, Microsoft, and Google, clip your coupons, and watch the AI trade drift higher.

Aschenbrenner looked at that playbook and said, wrong question.

The real question isn't who wins the AI race but rather what does every single winner unconditionally need to keep going?

His answer was three things: electricity, physical compute capacity, and the networking to connect it all.

His #1 position going into Q4 2025 wasn't a chip company or a cloud platform, it was Bloom Energy, which is a fuel cell company.

The thesis is almost embarrassingly physical because connecting a new AI data center to the power grid takes 3-7 years of regulatory approvals, utility negotiations, and infrastructure construction. 

Bloom Energy's solid oxide fuel cells can be deployed directly at the data center in 90 days.

If AI is about to go vertical and the grid can't keep up, someone who can drop power anywhere, fast, has pricing power that no model capability benchmark can compete with.

Here's the part that should make you put down your coffee.

Our analyst Vincent spotted this exact trade three months ago, the same bottleneck thesis, same infrastructure-over-applications logic. 

He flagged Bloom Energy to Milk Road PRO members when it was trading at half its current price, and that position is up over 100% in 90 days.

And guess what, he wasn't alone. 

Across the PRO portfolio, our analysts have been stacking the same category of trades, the picks-and-shovels plays that don't make headlines but print money when the AI buildout accelerates. 

Credo Technology, Applied Optoelectronics, AMD, Galaxy Digital, they are all up double digits in the last 30 days alone.

This is exactly what Milk Road PRO is built around, focusing on the thesis early, not recapping what already happened or explaining moves after the fact, but getting ahead of where the market is going, the same way Aschenbrenner did.

Now, back to the rest of the portfolio.

The rest of the portfolio reads the same way. 

He owns CoreWeave (GPU cloud) and Lumentum (optical networking inside data centers).

He also holds Intel call options, betting the company’s new chip process works, along with EQT, America’s largest natural gas producer.

Then there’s the broader layer, Bloom Energy, Core Scientific, Iren, Applied Digital, SanDisk, Coherent, Cipher, Tower, and Solaris, all tied to the same AI infrastructure buildout.

And his only short going into Q4 2025? Infosys, an Indian IT outsourcing company, because he thinks Claude Code and Cursor will automate most of what they do.

Genius, lucky, or both?

The honest answer is probably all three.

The case for genius, he was buying Bloom Energy before the Oracle data center deal validated the thesis. 

He was buying CoreWeave before the IPO and was long Bitcoin miners-as-AI-infrastructure before that trade existed as a concept in any analyst report. 

The thesis-to-position alignment isn't scatter-shot speculation but rather about someone who worked inside the room where AGI is being built and invested like he believed his own conclusions.

The case for luck, the fund launched in September 2024, timed perfectly to the single largest AI infrastructure spending cycle in history. Microsoft, Google, Amazon, and Meta announced a combined $300B+ in 2025 capex in Q4 2024. 

Anyone buying AI infrastructure broadly in late 2024 did well. 

The question nobody can fully answer yet is whether his specific picks generated alpha above just owning the category.

But here's the thing that separates him from the usual hedge fund story.

He didn't raise $225M by being cautious but rather raised it by finding a very small group of people who had read his 165-page essay, believed it, and decided that if there was even a 30% chance he was right, the position sizing implied by that probability was larger than anything they owned.

Most people who get fired write a LinkedIn post and move on. 

Aschenbrenner wrote 165 pages, called his shot, and convinced some of the sharpest people in tech to back it with nine figures. 

Love the thesis or hate it, that's a different category of human being. 

We'll know in a couple of years if he was Tukey or just another guy who was very confident at exactly the wrong time.

Alright, that's it for this edition of Milk Road AI. We want to hear from you.

BITE-SIZED COOKIES FOR THE ROAD 🍪

Anthropic may raise $50B at a $900B valuation, nearly a trillion. AI valuations keep soaring; Anthropic would rival the biggest tech giants.

OpenAI launched advanced security for ChatGPT with Yubico hardware keys. As AI holds more sensitive data, account security is becoming critical.

Meta's business AI hit 10M conversations a week, up 10x since January. A 10x jump in 3 months signals Meta's AI is finding real enterprise traction.

MILKY MEMES 🤣

ROADIE REVIEW OF THE DAY 🥛