GM. This is Milk Road AI, the newsletter that helps you separate signal from noise in the world of markets, tech, and artificial intelligence.
Here’s what we’ve got for you today:
- ✍️ Is the king of AI still untouchable?
- 🎙️ The Milk Road AI Show: The Scariest AI Prediction Yet… Is It True?
- 🍪 Silicon Valley vs. The Pentagon over AI weapons.
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 20% discount.
Prices as of 10:30 a.m. ET.
IS NVIDIA A BARGAIN OR A BUBBLE?
By April 1970, NASA had pulled off something arguably more dangerous than going to the moon.
They had made it boring.
Apollo 11 captivated 600M viewers, but Apollo 12 drew far less attention.
By Apollo 13, networks didn’t even bother airing the crew’s live broadcast from space.
The mission was running so flawlessly that the American public had essentially decided that landing on the moon was routine.
(Fun fact: CBS literally cut away from the Apollo 13 broadcast to air a rerun of The Doris Day Show. The moon had lost to a sitcom.)
Then, 55 hours into the mission, an oxygen tank exploded.
In an instant, the mission went from uneventful and forgettable to a life-or-death crisis unfolding on live television, and suddenly the whole world snapped back to attention.
Not because something new happened, but because they finally realized how fragile perfect always was.
NVIDIA just had its Apollo 13 moment.
Not the explosion part, not yet, anyway.
The part where perfection became so routine that nobody cared anymore.
They just posted the greatest quarter in semiconductor history, beat every estimate, guided above every forecast, and the market erased over $400B in value in the two days since earnings.
Because when you've beaten expectations fourteen quarters in a row, the fifteenth beat doesn't get a standing ovation.
Let's unpack why the market yawned at a company printing money faster than the Federal Reserve, and whether this is the dip you've been waiting for or the first crack in the oxygen tank.
The numbers (brace yourself)
NVIDIA reported fiscal Q4 2026 results last week and absolutely detonated expectations.
Revenue came in at $68.1B, up 73% year over year and 20% from last quarter.
Free cash flow alone was $34.9B in a single quarter, more cash than most Fortune 500 companies produce in an entire year.
Operating cash flow came in even higher at $36.2B, showing just how much raw cash the business is throwing off before any capital spending.
Here’s a stat that should melt your brain: Nvidia is currently pulling in about $500M in profit every single day.
At this pace, it could be making $1B per day by 2028.
At that point, they won’t report earnings calls; they’ll just report daily GDP.
And for the full fiscal year, Nvidia generated $215B in revenue and $96.7B in free cash flow.
And the gross margin? 75.2% up 170 basis points from a year ago, driven by the continued Blackwell ramp with an improved product mix and cost structure.
Most chip companies would sacrifice a kidney for a 50% gross margin.
NVIDIA is sitting at 75%, like it's normal.
So why did the stock crater?
Here's where things get weird.
NVIDIA guided Q1 revenue to $78B (plus or minus 2%), beating the consensus estimate of $72.6B by 7.4%.
And here's the wild part: that $78B guidance explicitly assumes zero Data Center compute revenue from China.
Meaning if they somehow get any China revenue at all, it's pure upside.
And this is their fourteenth consecutive quarter of beating estimates, and the stock dropped 5.6% the next day, erasing roughly $260B in market cap.
So what happened? Three things happened at once.
First, the buy-the-rumor, sell-the-news crowd showed up in full force.
Options pricing before earnings implied only a 5.6% move, the lowest for a Nvidia report in three years.
The surprise premium had been completely priced out.
Second, the narrative shifted, and investors are no longer asking if AI is real. They're asking whether we are at the top?
Third, and this is the big one: nobody knows what to do with the cash.
During the earnings call, UBS analyst Tim Arcuri basically said what every investor was thinking:
You're about to generate $100B in cash this year, the stock hasn't moved despite record results, and the price looks cheap, so why not plant a flag and announce a massive buyback?
NVIDIA CFO Colette Kress delivered a masterclass in corporate speak.
Lots of talk about supporting the ecosystem, securing supply chains, investing in partners, and pursuing “unique opportunities”.
Translation: Ecosystem investment comes first; buybacks come whenever we feel like it.
And the numbers back that up.
Capital returned to shareholders actually dropped from $8.1B in Q4 last year to just $4.1B this quarter, even as free cash flow more than doubled.
They're generating cash at a record pace and sending less of it back.
Meanwhile, they're guiding for $7.5B in operating expenses next quarter alone, a 47% jump year-over-year, because they're pouring everything back into R&D and infrastructure.
Translation: "We're keeping the money, and we're spending it. Thanks for understanding."
Apple returns $90B a year, Alphabet has a $70B buyback authorization, and Meta combines dividends with aggressive repurchases.
NVIDIA is sitting on a Scrooge McDuck pile of gold and telling investors to be patient.
That made Wall Street nervous, and nervous Wall Street sells first, asks questions later.
But here’s the part most people completely missed.
LET 2026 BE THE YEAR, YOU DON’T LET AN AGENT HACK YOU!
As OpenClaw was launched, the major use case came for hackers who trained it to steal SSH, crypto and wallet keys. Don’t become a victim to trendy AI platforms without having proper guardrails.
Okara is built with privacy at core. All chats are encrypted, protected and hashed with your client side keys. Nobody on earth can read your chats except for you.
The security protocol is open-sourced so anyone can verify.
- Okara comes with 30+ latest open source models
- Memory that grows as you chat more
- Unlimited Projects
- A store with agents like Reddit, SEO, GEO that help your business grow
Okara only costs $20/month for unlimited prompts. (Use Code MILKROAD for a 20% discount.)
If you care about privacy, Okara is built for you.
IS NVIDIA A BARGAIN OR A BUBBLE? (P2)
Alright, now let's get past the drama and look at what actually matters: what's under the hood.
Because honestly, the underlying business has never been stronger.
The growth engine behind all of this is Blackwell, Nvidia's latest GPU architecture, and specifically the GB200 NVL72 system.
I know those letters and numbers sound like a license plate, but stay with me because this thing is genuinely insane.
The GB200 NVL72 connects 36 Grace CPUs and 72 Blackwell GPUs into a single liquid-cooled rack that operates as one massive GPU.
It delivers 1.44 exaFLOPS of AI performance at FP4 precision.
(If that means nothing to you, it's 30x faster than the previous generation for real-time inference on trillion-parameter models. In English, it makes the biggest AI models run like butter instead of molasses.)
13.4 terabytes of memory, 130 terabytes per second of NVLink bandwidth across the 72-GPU domain.
And Nvidia isn’t stopping.
At CES 2026, Jensen Huang confirmed that all six chips in the next-generation Rubin platform have already returned from manufacturing and passed early testing.
Customer deployment is on track for the second half of 2026.
How much better is Rubin? Forget the alphabet soup of specs nobody can actually visualize.
Here’s what matters: Rubin is expected to be about 3.5x better at training AI and up to 5x more efficient at running it than Blackwell.
After Rubin comes Rubin Ultra in 2027, which is, as you probably guessed, even more ridiculous.
Then Feynman arrives in 2028, continuing Huang’s “one-year rhythm” of launching a new platform every twelve months that makes the previous one feel like it belongs in a middle school calculator drawer.
This is a decade-long upgrade conveyor belt that keeps getting better and keeps Nvidia getting paid every step of the way.
The data center is the whole story
Data Center revenue hit $62.3B in Q4, up 75% year-over-year, with compute at $51.3B and networking at $11B.
For the full fiscal year, Data Center pulled in $193.7B, up nearly 13x since the emergence of ChatGPT in late 2022.
Let that math sink in for a second: a business segment that was doing roughly $15B has scaled to nearly $200B in three years.
That's 91% of the entire company's revenue, and Nvidia is no longer a gaming company that happens to sell AI chips.
It's an AI infrastructure company that happens to still make graphics cards.
And the demand picture is borderline absurd, even Hopper chips and many of the six-year-old Ampere-based products are completely sold out in the cloud.
Six years old and still sold out, that's like Toyota announcing that the 2020 Camry has a waitlist.
And here's a number that most people skim past: networking revenue hit $11B, up more than 3.5x year-over-year.
For the full year, networking pulled in $31.4B, making Nvidia the world's largest networking business, period.
Why does networking matter? Because as AI clusters expand to tens of thousands of GPUs, the pipes connecting them become just as important as the chips themselves.
NVLink, Spectrum-X Ethernet, and InfiniBand are becoming the circulatory system of AI, and Nvidia owns all of it.
They're selling the entire nervous system, and they're claiming the title of the most performant, lowest cost-per-token inference provider with the largest installed base in the industry.
And that last part is critical.
If inference becomes the dominant workload (more on that in a minute), being the cheapest and fastest per token is the whole ballgame.
The $705B wave behind them
If you're wondering who's writing the checks for all this hardware, here's the answer.
The big five hyperscalers (Amazon, Google, Microsoft, Meta, and Oracle) have signaled a combined CapEx of roughly $705B for 2026.
That is nearly double what they spent in 2025.
Roughly 75% of that goes toward AI-specific infrastructure, and that's approximately $450B in AI spending in a single year.
Microsoft disclosed it has an $80B backlog of Azure orders that it literally cannot fulfill because it doesn't have enough power.
They have $80B worth of customers saying take my money and Microsoft's answer is we can't plug in enough servers.
And Nvidia is the one selling the shovels.
But wait, there's literally a wall.
Jensen Huang said it plainly on the earnings call: "Every data center is power-constrained."
Data centers in Nvidia's own hometown of Santa Clara are sitting empty because there isn't enough electricity to turn them on.
Multi-year waits for grid capacity are now common in major tech hubs.
Training workloads demand 100 to 200+ kilowatts per rack (up to 1 megawatt for frontier systems).
That's the equivalent of powering 100 homes with a single server rack.
NVIDIA knows this is the bottleneck, so they're pivoting from chip company to energy value chain orchestrator:
They launched an 800-volt DC power ecosystem with Schneider Electric and Eaton to support 1 MW+ racks by 2027.
They're exploring nuclear power for data centers in Japan and partnered with Bechtel to design modular 1-gigawatt data centers.
When a chip company starts worrying about nuclear reactors, you know the scale has gotten genuinely absurd.
But wait, there’s something even bigger on the horizon that could make today’s numbers look small.
What happens next could make or break the AI trade
Upgrade to PRO to access:
- The shift that could trigger the next explosion in AI spending.
- Why today’s numbers may look tiny in hindsight?
- What the market still hasn’t fully priced in.
- The risks that could blindside investors.
- My full buy-or-bubble verdict.
- Exactly how I’m positioning before the next move hits.
Summ (formerly Crypto Tax Calculator) is a tax software built specifically for crypto. Get started for free with Summ.
Mercuryo is the simple way to buy, sell, and spend crypto with low fees (yes, even with Apple Pay). Discover an easier way to use crypto.
Warbux is the easiest way to start trading crypto with zero deposits required. Get funded and start trading today!
BITE-SIZED COOKIES FOR THE ROAD 🍪
Anthropic won’t allow its AI for surveillance or autonomous weapons, while the Pentagon wants unrestricted use. The fight is over who controls powerful AI: companies or the government.
ChatGPT hit 900M weekly users and 50M paid subscribers, up from 800M in October. OpenAI also raised $110B in funding at a $730B valuation, one of the largest rounds ever.
Netflix backed out after investor backlash crushed its stock, signaling the deal was too risky. Rather than overpay in a bidding war with Paramount, it chose financial discipline.
MILKY MEMES 🤣
ROADIE REVIEW OF THE DAY 🥛
