GM. This is Milk Road AI, the only newsletter that tells you who’s actually making money in AI.
Here’s what we’ve got for you today:
- ✍️ Amazon owns the infrastructure layer.
- 🎙️ The Milk Road AI Show: Global Chaos Is Rising… So Why Are We Buying?
- 🍪 Apple’s smart glasses are coming.
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.
Prices as of 10:00 a.m. ET.
AMAZON OWNS THE AI ECONOMY
In 1494, the Medici family of Florence was the most powerful institution in Europe.
They weren’t the best soldiers, politicians, or artists, but they were the bankers.
While Da Vinci painted, Michelangelo sculpted, and Botticelli churned out Venuses like a Renaissance content farm, the Medicis were doing something far more boring and far more lucrative: they were financing all of it.
They didn’t care who got the glory, they just needed everyone creating, because every brushstroke ran through their books.
In 2026, there is a new Renaissance, and it's called AI.
OpenAI is Da Vinci, Anthropic is Michelangelo, and Amazon? Amazon is the Medici family.
They finance almost all of it, owning the vaults, controlling the ledger, and taking a cut of every transaction.
Last week, Amazon CEO Andy Jassy dropped the annual shareholder letter.
We read the whole thing, and once you see what’s actually in it, you’re going to look at Amazon very differently.
First, let’s set the scene
Amazon's 2025 revenue just hit $717B.
By the time you finish reading this newsletter, Amazon will have generated roughly $13.6M, but the total isn't the interesting part, rather the mix.
For the first time in its history, Amazon's service businesses, AWS, advertising, Prime subscriptions, now generate more revenue than its product businesses.
Services crossed $403B, while physical goods brought in less.
Amazon started as a bookstore, became a general store, then turned its own internal tools into massive businesses, from AWS to logistics to a $68.6B ad empire.
The pattern is simple: build it internally, perfect it, then sell it to everyone else.
What's the next internal capability about to become a business? I'll let you read to the end to find out.
Here is the single most important data point in Jassy's letter, and it is not getting nearly enough attention.
AWS's AI revenue just hit a $15B annual run rate.
That is the first time Amazon has publicly disclosed this number, and Jassy immediately did something smart, he put it in historical context.
Three years after AWS launched commercially as a product, its total revenue run rate was $58M, but three years into the AI wave? $15B.
Jassy called it "the fastest commercial adoption of any technology Amazon has ever seen."
And remember, AWS itself was already the fastest-growing enterprise business in history when it launched, so we're comparing against a very high bar and somehow clearing it.
And this $15B isn't some accounting trick or creative bundling.
It includes real products that real companies are paying real money for:
- Trainium-powered compute for training and running AI models.
- Amazon Bedrock, giving enterprises access to models like Claude and Nova without managing infrastructure.
- SageMaker for building and training models.
- Kiro, a new agentic coding tool that rebuilt Bedrock’s inference engine in 76 days with just six engineers.
That last one should tell you everything, this thing is moving way faster than people realize.
The $200B "problem"
Here's where every financial headline got this story wrong.
Amazon’s free cash flow dropped from $38B to $11B in 2025, a 71% decline in one year.
And in 2026, Amazon plans to spend approximately $200B in capital expenditures, the largest single-year capex commitment by any company in the history of capitalism.
Investors panicked, the stock got choppy, but Jassy spent a significant portion of the letter explaining, very patiently, why the people panicking are thinking about this wrong.
His argument, translated from CEO into English, goes like this:
Data centers take 6 to 24 months to build before you can charge a single customer.
Chips and servers have 5–6 year lifespans. Buildings last 30+ years, so the cash goes out the door long before the revenue comes in.
In high-growth periods, the gap between building capacity and billing for it creates a temporary free cash flow crater that looks bad on a chart but says little about the underlying business.
Then he dropped the argument that should have ended the conversation: "We're not investing approximately $200B in capex in 2026 on a hunch."
The OpenAI contract alone is worth over $100B.
Multiple large contracts already exist, meaning the capacity is effectively spoken for before it’s even built.
Think of it this way.
You're a landlord, and a Fortune 500 company signs a 10-year lease on an office building you haven't built yet.
To build it, you take out a big loan, and your cash flow looks awful for 18 months during construction.
Then the tenant moves in, the rent checks arrive, and suddenly your terrible cash flow decision looks like a very good decade.
That's the entire situation. The bears saw the construction loan, while Jassy is showing you the signed lease stack.
He also made a historical comparison that investors should take seriously: AWS went through this exact same investment cycle during its first big growth wave.
The free cash flow pain was real, the operating leverage that followed was enormous.
He is explicitly telling you: same movie, different decade, stakes are 260 times higher.
Amazon bought the factory
Now for the story that could genuinely change how you think about Amazon as a company.
Most investors think of Amazon as a cloud company that builds chips for internal use.
Jassy is signaling something far more interesting: those chips are on their way to becoming one of Amazon's biggest standalone businesses.
Amazon has three custom chips. Here's the quick version:
- Graviton → Amazon’s in-house CPU, ~40% better price-performance, now used by 98% of AWS’s top customers.
- Trainium → Their NVIDIA alternative, ~30% cheaper, with next-gen chips nearly sold out before launch.
- Nitro → The invisible backbone, powering thousands of instances on a single server behind the scenes.
Together, these three chips are generating over $20B in annualized revenue, growing at triple-digit percentages year-over-year.
Now here's the sentence Jassy buried deep in the letter that should be in every headline but isn't:
"There's so much demand for our chips that it's quite possible we'll sell racks of them to third parties in the future."
At current production rates, if Amazon sold chips externally as a standalone business, Jassy estimates it would clock approximately $50B in annual revenue.
For reference, that would be larger than AMD's recent run rate. At semiconductor company margins, 40 to 50% gross, that's $20 to $25B in gross profit from a business that doesn't even officially exist as a segment yet.
When Amazon eventually breaks chips out as a separate reporting segment, and Wall Street will demand it once they see what's in there, analysts will apply semiconductor multiples to it.
The resulting re-rating could be violent in the best possible way but the chip story is only half of it.
That’s also why I added Amazon to my PRO portfolio about a month and a half ago and it’s already up +14.6%.
I’ve been screaming to buy this in the Discord for weeks now and it’s my second-largest position at this point, and my top position is already up +36.3%.
If you want the next ones before they move, come join us, it’s literally less than your last Amazon order.
CRYPTO SHOULD WORK HARDER FOR YOU
Most people hold crypto and hope.
The smart money? They're earning interest on it, borrowing against it without selling, and trading it.
Where can you do the same all in one place? Nexo.
And right now, new U.S. clients get 30 days of Wealth Club Premier (benefits normally reserved for loyalty program members):
- Enhanced interest rates on your digital assets
- Lower borrowing costs against your crypto
- Up to 0.5% cashback on trades
No need to sell to access liquidity. No juggling 5 different platforms.
*Disclaimer: Geographic restrictions and terms apply.
AMAZON OWNS THE AI ECONOMY (P2)
Let’s talk about what AI is doing to Amazon's other businesses, because this part gets almost zero coverage.
Start with Alexa. Amazon has 600M active Alexa endpoints.
They rebuilt Alexa from scratch on generative AI, and the results are not subtle.
Users are having twice the conversations, completing 3x more purchases, with streaming up 25% and smart home usage up 50%.
When you rebuild a product with a genuine AI upgrade instead of just bolting a chatbot onto it, the engagement numbers look like you accidentally put rocket fuel in the gas tank.
Now consider what this means for the advertising business. Amazon's ad revenue hit $68.6B in 2025, growing at 22%.
It is arguably the most defensible advertising business on Earth because, unlike Google (which knows what you search) or Meta (which knows what you click), Amazon knows what you actually buy.
And purchase-intent data at that scale is worth more per impression than almost anything else in digital media.
Jassy explicitly said AI-generated recommendations will be embedded directly into Alexa+ conversations.
Then there's grocery, which grew to $150B in gross sales in 2025, making Amazon the second-largest grocer in the United States.
Perishables same-day delivery saw 40x growth since early 2025, and Amazon Now, their 20-minute delivery product in India, is growing at 25% month-over-month.
This is what Jassy means by a business multiplier.
It’s not just a chatbot you add to a website, it’s improving how every part of the business runs.
Retail gets better routing and inventory, ads get sharper targeting, AWS brings in new spending, and Alexa starts driving real commerce.
Each piece reinforces the others, and the whole system is moving faster than it ever has before.
The constraint nobody is talking about
Here's the thing though, Amazon could be growing even faster, and they know it.
The conventional view is that chip supply is the bottleneck, and that’s true. More GPUs still means more growth.
But there’s a deeper constraint emerging alongside it: electricity.
AWS added 3.9 gigawatts of new power capacity in 2025.
They plan to double total power capacity by the end of 2027, and the Tennessee Valley Authority projects that data center electricity demand will double by 2030.
Nearly half of all data centers planned for 2026 in the United States are being delayed or outright canceled, not for lack of money, but for lack of grid capacity.
Jassy is explicit about this in the letter: there is unserved demand right now.
AWS could be signing more contracts today if it had more power.
The 20% to 24% growth numbers you're seeing from AWS are the constrained numbers, underlying demand is higher.
This has a counterintuitive implication: the companies that secured power purchase agreements and land early have a structural advantage that cannot be fast-followed.
You either got in before the land rush, or you're waiting years while your competitors sign customers you can't serve.
Amazon started its power procurement cycle long before the AI boom made it fashionable.
Two of AWS's largest customers asked to buy all of its available Graviton chip capacity for 2026.
Amazon had to say no because it had too many other customers to serve, and that is what genuine pricing power looks like in the wild.
The number hiding in plain sight
Here is the contrarian argument Jassy basically hands you in the letter, for free, and most people will still miss it.
Eighty-five percent of global IT spend still runs on-premises, which means most companies still run their tech on their own servers instead of the cloud.
AWS is at a $142B revenue run rate, massive by any normal measures, and it represents approximately 15% of total global IT spend.
As more of that remaining 85% gradually moves to the cloud, driven by the economics, AWS’s market grows far beyond what it looks like today.
This is the cloud migration tailwind entirely independent of AI.
Before you add a single AI workload, just the base cloud migration is one of the longest and most durable secular growth stories in enterprise technology.
Add AI on top of that?
Jassy has internally projected that AWS alone could hit $600B in annual revenue by 2036.
Amazon's entire company did $717B last year.
He is saying one division could approach that in a decade, at operating margins three times higher than the rest of the business combined.
AWS at $600B with 35% operating margins generates $210B in annual operating income from one division.
Go ahead and sit with that for a minute. I'll be here.
The verdict: the Medicis always win
Here is the honest assessment.
The bear case is real, $200B is a historic bet.
If AI adoption plateaus, if enterprise customers slow their migrations, if the hype collapses, Amazon will have deployed enormous capital into assets generating long-term returns but needing near-term customers to make the math work.
Free cash flow has already crumbled, and it is not irrational to be nervous.
But the bull case is not hype, it has signed contracts, disclosed run rates, and a historical pattern that has now repeated itself multiple times inside the same company.
The same people who called early AWS spending reckless are explaining why they've held for twenty years.
Most CEOs are desperately managing to the next quarterly earnings call, and they are terrified of missing estimates by a penny.
Their entire incentive structure is a 90-day horizon, but Jassy is building much further out.
The Medici family financed the Italian Renaissance for over a century.
They didn't pick winners, but rather owned the infrastructure that made winning possible.
Every painting sold, every sculpture commissioned, every cathedral built, the Medicis got a cut.
That is the business Andy Jassy is describing.
He’s not just in models, chatbots, or even cloud in the way most people think, he’s in the empire business.
Alright, that’s it for this edition of Milk Road AI. We want to hear from you.
Is Amazon building the most powerful position in AI?
- Yes, they’re quietly winning.
- Maybe, but it’s still too early.
- No, others will dominate the top layer.
Hackers are using prompt injections to break into crypto users accounts. Don’t be a victim! Use private, safe and encrypted chat Okara. Use the code MILKROAD to get a 20% discount.
Summ (formerly Crypto Tax Calculator) is a tax software built specifically for crypto. Get started for free with Summ.
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.
BITE-SIZED COOKIES FOR THE ROAD 🍪
Apple is testing four designs for smart glasses planned for a 2027 launch. The glasses will focus on cameras, audio, and Siri, not full AR displays.
U.S. officials are urging banks to test Anthropic’s Mythos model for security risks. The push comes even as Anthropic battles the government over AI use limits.
OpenAI launched a $100/month Pro plan with higher coding limits. It sits between Plus and the $200 tier, targeting heavy Codex users.
What if a bank app and crypto wallet had a baby? You’d get Brighty which is a money app that lets you use crypto and regular fiat money together without transferring funds.
MILKY MEMES 🤣
ROADIE REVIEW OF THE DAY 🥛
