GM. This is Milk Road AI, where we find tomorrow's winners before Wall Street does.
Here’s what we’ve got for you today:
- ✍️ The most important AI infrastructure story of 2026.
- 🎙️ The Milk Road AI Show: Companies Are Hoarding Compute Like Gold.
- 🍪 Uber burned through its AI budget.
Kalshi is a regulated prediction market where you can trade on real-world events. Special offer for Milk Road Readers: Trade $10 on Kalshi → Claim $10 for free.
Prices as of 10:00 a.m. ET.
GOOGLE JUST FIRED THE BIGGEST GUN IN AI HISTORY
In 1943, the U.S. War Department needed to build the world's first atomic bomb before Germany did.
They didn't submit a budget request or establish a committee.
They called it the Manhattan Project, handed it $2B in secret appropriations, and told Robert Oppenheimer to figure out the physics later.
The point wasn't the bomb itself but rather the statement, we are moving faster than you, at a scale you cannot match, and we are not asking permission.
Google just made a very similar announcement.
They just raised $80B in equity, the largest equity offering in the history of profitable technology companies and told the market the same thing.
We are moving faster than you, at a scale you cannot match, and we are not asking permission.
Most people read this as a Google story but they are looking at the wrong company.
The real story is about a semiconductor designer in San Jose that almost nobody outside of Wall Street can name, and why this $80B just handed them what might be the most valuable contract backlog in the history of the chip industry.
But first, let's talk about what Google actually did.
The structure of the raise (and why it matters)
$80B sounds like one big number but it is actually three moves happening simultaneously.
$15B in mandatory convertible preferred stock, $15B in straight common equity and a $40B at the market program that lets Google bleed shares into the open market gradually, starting in Q3, without cratering the stock price with a single massive block sale.
But the real validator is the one buying $10B in a private placement before this thing even hits the open market.
His name is Greg Abel and he runs Berkshire Hathaway.
Warren Buffett's handpicked successor looked at Google's AI buildout, decided it was a reasonable long-term bet, and wrote the biggest check his firm has written in years.
When the most conservative large capital allocator on earth decides AI infrastructure is worth $10B of Berkshire's money, it suggests the underlying demand is real and the returns may still be underappreciated.
So why did Google actually need to do this?
Here is the uncomfortable sentence buried in Google's press release.
"The company is experiencing strong demand for its AI solutions and services from enterprises and consumers, at levels that are exceeding the company's available supply."
And the financing market appears to agree.
Google's convertible stock sale is reportedly oversubscribed as the company raises $80B to fund its AI infrastructure buildout.
Investor demand for the offering has exceeded the amount of stock available, as institutions bet on Google's ability to scale compute capacity for what the company itself describes as "unprecedented customer demand."
Google, the single largest owner of AI compute on earth, controlling roughly 25% of global cumulative AI computing capacity, running approximately 3.8M of its own custom chips is telling you they cannot keep up with demand.
The Cloud business grew 63% year-over-year in Q1 2026, the fastest rate since Google began reporting it as a separate segment.
The Cloud backlog nearly doubled quarter-over-quarter to $460B.
Revenue from generative AI products surged nearly 400% year-over-year in Q4 2025.
This is not a company with a demand problem but rather a company with a supply problem.
And so the obvious question becomes, if Google generates roughly $64B in free cash flow annually, has $100B in net cash, and can issue investment-grade bonds at basically risk-free spreads, why raise equity at all?
Because the math on waiting is terrifying.
Equity is the most expensive form of capital Google could have raised but they chose it anyway.
That tells you the speed requirement exceeded everything else.
Issuing $80B in bonds would have weakened the balance sheet, while waiting for free cash flow would have cost them 6–12 months of compute leadership.
A 3.3% dilution hurts on paper, but it puts $80B into infrastructure in Q3 2026 instead of Q3 2027.
The implicit statement here is extraordinary, the return on locking up compute right now is greater than the cost of the dilution.
For a company as disciplined as Google, that threshold is almost never crossed.
CRYPTO’S THIRD PMF MOMENT IS HERE
PMF = Product Market Fit
Until now, crypto has really had two major PMF moments: 24/7 trading and stablecoins.
Now a third one may be emerging with Kalshi: Prediction Markets.
Kalshi is a regulated prediction market exchange where people trade on the outcomes of real-world events (elections, inflation, sports, crypto prices and more).
And their growth has been massive:
- Recently raised $1B at a $22B valuation
- Institutional trading volume is up 800% in six months.
- Annualized volume is somewhere around $178B.
Kalshi is giving Milk Road readers $10 free when you trade $10.
Trade $10 → Claim $10 on Kalshi.
GOOGLE JUST FIRED THE BIGGEST GUN IN AI HISTORY (P2)
Now here is the part your financial advisor isn't telling you.
Google didn't raise $80B to let it sit on the balance sheet.
The vast majority of that capital is headed toward AI infrastructure: more compute, more data centers, more networking capacity, and most importantly, more TPUs, the custom chips at the heart of Google's AI strategy.
All of that money has to go somewhere, chips don't manufacture themselves.
You need someone to design the custom silicon, handle the physical implementation, coordinate the advanced packaging, and manage the manufacturing at TSMC.
That company is Broadcom.
Most people think of Broadcom as a boring semiconductor conglomerate that makes Wi-Fi chips for your router and that's like describing Amazon as a bookstore.
Broadcom actually runs two distinct high-margin businesses in parallel.
About 65% of revenue is semiconductor design, custom AI accelerators, high-speed networking chips, and the interconnects that tie massive AI clusters together.
The other 35% is infrastructure software, built around the $69B VMware acquisition, which generates roughly $27B annually at gross margins above 93% and is being positioned as the enterprise platform for private AI.
That software floor is the thing most analysts miss entirely.
It means Broadcom can invest aggressively in chip R&D through any part of the semiconductor cycle without depending on timing, because the cash register keeps ringing regardless of where chip demand sits.
But back to Google.
The relationship between Google and Broadcom on TPU chip design goes back nearly a decade.
Google controls the top-level architecture, designing the critical compute blocks at the heart of the AI workloads.
Broadcom handles the physical design, the implementation, the high-speed interconnects, and the manufacturing coordination at TSMC.
Broadcom manages the advanced packaging on the training-optimized chips, which are the most technically complex variants to produce.
Google's seventh-generation TPU, codenamed Ironwood is the chip this $80B is being built around.
Dual chiplet architecture on TSMC's 3nm process and 192 gigabytes of HBM3e memory per chip.
And it is 4x faster inference than the prior generation at 30% lower power.
Anthropic just signed a deal for up to 1M of them, the single largest TPU commitment in history.
For Broadcom, the TPU program translates directly into revenue.
The TPU v6 program alone was expected to generate over $15B in lifetime revenue for Broadcom.
The TPU v7 program is larger, more complex, and deploying into a period where Google just raised $80B to accelerate rollout.
One honest risk
There is a genuine caveat here worth stating directly.
Google has started working with MediaTek on certain TPU variants, specifically the inference-optimized, cost-sensitive chips where eliminating Broadcom's supply markup on HBM memory saves hundreds of millions at Google's scale.
This is real but the question is what it actually means.
Broadcom retains the training chip development, the highest-value, highest-complexity, highest-switching-cost portion of the relationship.
MediaTek is gaining share on inference variants.
Even accounting for this, Broadcom is projected to hold 60-75% of the custom ASIC design market through at least 2027.
Google is using MediaTek as leverage on lower-end inference chips, but that looks more like a negotiating tactic than a true replacement for Broadcom.
What this actually is
Google's $80B raise is not just a financing event but rather a declaration that winning the AI race matters more than short-term dilution.
The demand for AI compute is real, accelerating, and outrunning supply at the company that owns more of it than anyone else on earth.
The decision to sell equity rather than tap bonds or wait for cash flow tells you the speed requirement was non-negotiable.
The Berkshire participation tells you that the most patient money on earth sees the returns.
Most of it flows into TPU infrastructure and all of that infrastructure runs through Broadcom.
I already added a position in Broadcom on May 27 at $419.46 per share.
Milk Road PRO members saw the alert in real time, and the stock has already climbed to roughly $492 as of writing this.
If you want to see ideas like this before they play out, you can join us inside Milk Road PRO.
Broadcom reports earnings today, and I'm expecting a stellar beat along with strong guidance as hyperscaler AI spending continues to accelerate.
If I'm right, Google's $80B raise may end up being remembered as one of the clearest demand signals AI infrastructure investors ever received.
Alright, that's it for this edition of Milk Road AI. We want to hear from you.
Are you joining Milk Road PRO for just $1?
- Absolutely. Show me the research.
- I'm interested, but I need one more winning idea.
- Not yet. I'll wait until the next stock doubles.
COMPANIES ARE HOARDING COMPUTE LIKE GOLD 💰
In today's episode, we dig into why corporations are treating compute as the new corporate store of value, hoarding GPUs and TPUs over passive assets like Bitcoin.
Here's what you'll hear:
- Why Google's $80B equity raise (plus $10B from Berkshire) signals a massive AI CapEx wave and a race to lock in capacity.
- The real bottlenecks: chip fab cadence, power and cooling limits, plus fiber and optical interconnect supply.
- Why ASIC, optical, and fiber names like Broadcom, Corning, and Marvell are in focus as new rack standards expand the market.
- How "SaaS is back" with agentic apps like Robinhood's agent-managed accounts, and why cybersecurity matters more than ever.
Press play and dive in 👇
YouTube | Spotify | Apple Podcasts
Real Finance Blockchain is an EVM-compatible L1 that is built specifically for RWA tokenization. Read more about Real Finance Blockchain here.
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.
Your carrier is selling your data. Use code MILKROAD for 33% off your first 6 months.
BITE-SIZED COOKIES FOR THE ROAD 🍪
Special offer for Milk Road readers. Kalshi is giving Milk Road readers $10 free when you trade $10 on their platform.*
Uber blew its entire employee AI budget in just 4 months and had to cap it. $1,500/month per employee, the AI productivity boom comes with a real price tag.
Anthropic filed confidentially for an IPO, the first major AI lab to do so. Near a $1T valuation, this IPO would be bigger than any tech listing since Meta.
OpenAI expanded Codex with six new plug-ins for jobs like finance and sales. With 5M weekly users, Codex is now OpenAI's biggest enterprise bet.
*this is sponsored content.
MILKY MEMES 🤣
ROADIE REVIEW OF THE DAY 🥛
