GM. This is Milk Road AI, the only newsletter that watches markets like a scoreboard.
Here’s what we’ve got for you today:
- ✍️ Why Google still looks undervalued.
- 🎙️ The Milk Road AI Show: NVIDIA CEO’s Biggest AI Warning: AI Tokens Will Decide Which Companies Survive.
- 🍪 SpaceX’s $60B bet on a coding AI empire.
Consensus Miami is one of the largest digital asset conferences that’s going all in on crypto and agentic commerce. Grab your passes at 20% off.
Prices as of 10:00 a.m. ET.
THE MARKET IS SLEEPING ON GOOGLE
There is a famous scene in Moneyball where Billy Beane is trying to convince his scouts to buy players the entire league has written off as broken, old, or just too weird to touch.
The room pushes back hard and says, "We're not selling jeans here."
But Beane doesn't care, he only cares about one thing: what is this player actually worth versus what the market is pricing him at?
That gap is the opportunity.
And it doesn’t just show up in baseball, it shows up in markets especially when the asset looks obvious.
Because the market doesn’t miss bad companies but rather misprices great ones, especially when they’re already at all-time highs.
Too expensive, too late, already priced in.
Currently, Google is at $330, and Marvell is at $151, both of which are near all-time highs.
And this weekend, The Information reported that Google is in active talks with Marvell to co-design two brand-new AI chips, a memory processing unit, and a new inference-optimized TPU.
There is no contract signed yet, but just like in Moneyball, the market is focusing on the surface, the price, instead of the underlying value being created.
And that’s where the opportunity is.
The cheapest mega-cap AI winner on Earth
Here is a genuinely offensive sentence.
Alphabet trades at a lower P/E than the average S&P 500 stock, and that’s before you even strip out its $100B+ cash pile.
You’re getting Google Search, YouTube, Gemini, Google Cloud, and Waymo essentially at market-average prices.
Google trades at roughly 22x forward earnings, near its five-year average forward P/E, and well below where it traded in 2020–2021.
That is what you pay for a boring regional bank, not a company compounding revenue at 15% per year while growing its Cloud business at 35.8% and simultaneously building its own AI chips.
So why is the multiple this low? Because the market is still pricing Alphabet like a cyclical ad company, but that couldn’t be further from the truth.
Google is a full-stack AI company that just happens to also print money from ads while it builds.
Here is what you are actually getting for that 22.9x.
Google Services generated $139.4B in operating income last year, the cash machine that funds everything else.
Google Cloud just crossed $58.7B in revenue, and operating income tripled to $13.9B in a single year.
YouTube is quietly doing $40.4B in ad revenue without anyone noticing.
Waymo is already running a real driverless taxi business across Phoenix, San Francisco, LA, and a growing list of cities nationwide.
And then there is the thing that almost nobody is talking about.
The $100 billion secret
Google owns roughly 7.4% of SpaceX.
They originally invested around $900M back in 2015, when SpaceX was valued at $10B.
SpaceX is now being discussed for an IPO at a valuation of $1.5 to $2T.
At the $800B December secondary price, Google's stake is worth roughly $40B.
At a $2T IPO, that single position would be worth $100 to $122B, roughly 3% of Alphabet's entire market cap, sitting in an investment the market is not pricing in at all.
In Q1 2025, one SpaceX revaluation generated an $8B unrealized gain for Alphabet.
That was 25% of Google's entire net income for the quarter, from one line item most investors had forgotten existed.
When SpaceX goes public, and it will bring another $40 to $80B, could flow straight through Alphabet's earnings in a single quarter.
It would be like finding out your house has a second house inside it.
And here is the layer that really breaks your brain.
SpaceX now owns xAI after the February 2026 merger with Elon Musk's company, which means Google's SpaceX stake includes indirect exposure to Grok, the Colossus supercomputers, and Starlink's satellite AI data centers.
If Musk's entire AI stack outperforms, Google wins anyway.
THE ENTIRE INDUSTRY IS HEADING TO MIAMI
Where do you go to hear people talk about crypto, AI, and real capital?
If you surface-level takes, then it’s probably in my group chat.
But if you want institutions with deep pockets, you’ll have to head to Consensus Miami.
Consensus Miami is one of the largest digital asset conferences that’s going all in on crypto and agentic commerce.
Here are the key details:
- 20,000+ global attendees
- $4T AUM managed by finance giants attending Consensus Miami
- The ultimate intersection of crypto and AI
The best part?
You can get an exclusive 20% discount on passes with code MILKROAD.
THE MARKET IS SLEEPING ON GOOGLE (P2)
Now back to this morning's chip news.
The Marvell talks are not about desperation but rather about redundancy.
Google already has Broadcom locked in through 2031 on TPU development.
Adding Marvell as a third design partner alongside MediaTek gives Google something no competitor on Earth has right now: a fully diversified custom silicon supply chain that cannot be held hostage by any single vendor.
And every chip Google designs with Marvell instead of buying from Nvidia saves Alphabet approximately 75 points of gross margin.
That margin does not go to Jensen Huang, but rather it stays at Alphabet.
At the scale Google operates, that is billions of dollars per year quietly shifting from Nvidia's pocket to Google's.
The custom ASIC market is projected to grow roughly 45% in 2026 alone, and Google is building its own on-ramp to capture it.
But here is a riddle.
What company is simultaneously the second-largest custom AI chip designer on the planet, the dominant supplier of optical interconnects that every major AI data center runs on, received a $2B equity check from Nvidia, and as of this morning is reportedly in serious talks to become a core chip partner for Google?
That is Marvell Technology.
Revenue hit $8.2B in fiscal year 2026, up 42.1% year-over-year.
The Data Center segment surged 46.5% to $6.1B, which is now 74% of Marvell's total revenue.
This is no longer a diversified chip company but rather an AI infrastructure company that happens to have some legacy businesses on the side.
And before this morning's report, Marvell's custom AI chip clients were AWS, for the Trainium chip, and Microsoft, for the Maia chip.
That is a good list.
Adding Google would make Marvell the first company ever to design custom AI chips for three of the four major hyperscalers.
The only one missing is Meta, and they are reportedly evaluating Marvell for their next internal chip program right now.
The economics here are staggering.
Marvell earns roughly 5 to 10% of the total silicon value it designs.
If Google's combined chip programs reach $30 to $50B in annual production value by 2028, Marvell's cut is $1.5 to $5B in incremental high-margin revenue layered directly on top of what AWS and Microsoft are already paying.
That is enough to bend the revenue curve meaningfully above what any analyst currently has in their model.
Okay, so what could actually go wrong?
If hyperscaler AI CapEx plateaus in late 2026 or 2027, both names de-rate at the same time, which is the shared risk on this trade.
A rate shock would also hurt, since long-duration growth stocks lose multiple support when yields spike.
For Google specifically, the DOJ antitrust case is not fully resolved.
Chrome divestiture is still technically on the table, even if analysts expect behavioral remedies.
And Google spent $91.4B in CapEx in 2025. Any sign that ROI is disappointing will pressure the stock hard and fast.
Google is expected to ramp that even further, potentially reaching ~$180B in CapEx , as hyperscaler spending continues to explode.
For MRVL specifically, the Google talks have not produced a signed contract yet and that’s a real execution risk but regardless of whether this happens or not, this is still a high-quality company.
AWS represented over 40% of data center revenue at peak, and that single-customer concentration remains a risk.
Broadcom could aggressively cut prices to protect market share.
The verdict
I’ll be adding GOOGL and MRVL to my watchlist.
This chip news makes this setup too compelling to ignore.
Google is becoming a full-stack AI company, owning the demand side through Search, Cloud, Gemini, and now its own custom silicon.
Every chip it designs pulls margin away from Nvidia and keeps it in-house.
On the other side, Marvell is positioning itself as the picks-and-shovels layer of this entire buildout.
They’re designing custom chips for hyperscalers, running the optical backbone of AI data centers, and now potentially adding Google to a client list that already includes AWS and Microsoft.
Owning both means you’re not betting on one architecture, one company, or one outcome.
You’re owning the demand and the supply and letting the entire AI buildout work in your favor.
We just launched the Milk Road Watchlist where you can track every stock our analysts and I are watching in real time, I’ll be adding these names there, and you can join us for just $1.
I’ve already added Oracle there as well.
Alright, that's it for this edition of Milk Road AI. We want to hear from you.
Which side of this trade are you on?
- All in on GOOGL: The SpaceX stake alone makes this a no-brainer.
- All in on MRVL: 30% compounder, Nvidia-endorsed.
- Buying both: The pair trade makes too much sense to ignore.
TOKENS WILL DECIDE WHO SURVIVES 🪙
In today's episode, we sat down to break down Jensen Huang's framing that token consumption efficiency will determine which AI companies live or die in the capital and compute war.
Here's what you'll hear:
- Why the Anthropic-Amazon $25B deal and Anthropic's $100B AWS spend reshape the compute race against OpenAI and Grok.
- How rising inference costs invert SaaS logic: power users now increase unit costs instead of driving margin expansion.
- Why the panel favors AI CapEx names like Nvidia, AMD, and Oracle over broad software exposure going forward.
- How USD(.)AI, a 7.7% APY stablecoin financing GPU purchases, uses crypto rails to fund real-world compute demand.
Hit play and see for yourself 👇️
YouTube | Spotify | Apple Podcasts
Kalshi is one of the largest prediction markets that allows users to trade on the outcome of real world events. Sign up here for a free $10 and start trading.
Real Finance Blockchain is an EVM-compatible L1 that is built specifically for RWA tokenization. Read more about Real Finance Blockchain here.
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.
BITE-SIZED COOKIES FOR THE ROAD 🍪
SpaceX is partnering with Cursor to build a next-gen coding AI, with an option to buy it for $60B. The deal ties into its IPO story while aiming to reduce reliance on OpenAI and Anthropic.
Unauthorized users accessed Anthropic’s Mythos tool via a third-party vendor, though no systems were breached. The incident raises concerns about how easily powerful AI tools can leak.
Tim Cook is stepping down as Apple CEO after 15 years, passing the role to John Ternus. He led Apple’s rise from ~$350B to a $4T giant through global expansion and growth.
MILKY MEMES 🤣
ROADIE REVIEW OF THE DAY 🥛
