Read more

One of you asked me about MRVL after Jensen said cast his spell on them. Now trading near $300, is it a good time to buy?

The floor was open. Ten names came in. Here is every one, and the rule I ran them through.

What this is

This week I opened the floor in the room. Drop any ticker, any sector, any topic, and I cover it. One person wants one name, it is a deep dive. A lot of you want a lot of names, I cover them all and run every one through the same rule.

You gave me ten. NOW. IBM. MRVL. BRUN. TXG. KRKNF. AUR. UBER. LEU. CEG. Then I added the eleventh myself, because half of you keep circling it without naming it, and because it is the map underneath everything else you asked for: everything NVIDIA is invested in, and everything NVIDIA quietly relies on.

Here is the rule. The same one every Anchor and every swing in my book has to clear. I quote it exact because it is load bearing.

I buy the exact raw materials, hardware machines, and power plants that force artificial intelligence to actually exist and operate in the real world.

Pass the line, a name can earn a seat. Fail it, it is a good company in the wrong book. Seven of these pass. Four fail, and the honest no is the whole reason this is worth reading. Let's go down the list.

MRVL. The name Jensen just anointed.

Here is the part the noise buried on Marvell, the name everyone is asking about after the Jensen fireworks.

The pop was not just a soundbite. On March 31, NVIDIA put roughly 2 billion dollars of equity into Marvell and wired it into NVLink Fusion, the rack platform that lets custom silicon plug straight into NVIDIA hardware. NVIDIA is now a shareholder, not just a neighbor (Does that explain his kind words in Taipei, you can decide). Computex was the victory lap for a deal already signed.

Underneath that, the business delivered. Q1 came in at 2.42 billion dollars, up 28 percent, with data center now 76 percent of revenue. They raised the full year to about 11.5 billion and the year after to 16.5 billion. The custom chips Marvell co-designs for Amazon and Microsoft are targeted past 10 billion by 2029. This is connectivity and custom compute, the literal plumbing between GPUs. It passes the line clean.

Now the part you actually asked. Is it inflated. Yes, right now. It trades over 100 times trailing earnings, and the forward multiple jumped from the mid 30s to about 50 in a single week while the numerator never moved. That is sentiment, not math. I owned this at 90 with my dad. I sold mine into a spring earnings pop, and I sold his on Friday, which was just poor execution on a small position I was rebuilding. Then Jensen opened his mouth on Monday and it is 300. That stings. It also does not change the read. Great company. Wrong price today. The analyst median target sits at 240, below where it trades. I want it back with a 2 handle before I rebuild, not a 3.

June 2, Computex. Jensen calls Marvell the next trillion dollar company. Plus 60 billion in market value in a day. The deal under it was signed March 31.

BRUN. A rocket with a red flag.

BRUN is Boost Run. It started trading on Nasdaq on May 11 after a deSPAC, and this one is dead center of the thesis. It is a GPU neocloud. It rents NVIDIA Blackwell compute to AI labs. Same layer as CoreWeave and Nebius, one rung above the chip.

The numbers are a rocket with a crack in the fuselage. Revenue is small and exploding, 11 million dollars last quarter, up 165 percent. The backlog is the story: 940 million dollars under contract against a target of 375 million in run rate revenue exiting this year. The marquee deal is 472 million dollars from Thinking Machines Lab, Mira Murati's startup, for 5,000 NVIDIA B300 chips on a take it or pay it term. NVIDIA handed it Exemplar Cloud status on Blackwell. Dell signed a 1.44 billion dollar supply agreement. It passes the line easily.

The crack. It is three weeks public and it already filed its first quarter late. A brand new company that misses its very first SEC deadline is waving a small yellow flag at you about its own controls, and cash on hand is thin for a business this capital hungry. The instinct to watch this one is right. Buy dips, do not chase a stock that already doubled. I want it on a flush toward the high 20s, not at 40 after the run, and I want to read that late filing for the internal controls language before I touch it.

NOW. Benchmarked against Microsoft and Salesforce.

Put ServiceNow up against Microsoft and Salesforce. Straight answer below.

The business is elite. Q1 revenue 3.77 billion, up 22 percent, current bookings up 22 percent, free cash flow margin 44 percent. They beat every line and the stock fell 17 percent anyway, which tells you the bar is now AI proof, not beats. The AI line is real. Now Assist crossed a 1 billion dollar run rate, they raised the AI target to 1.5 billion, and more than half of new business is no longer priced per seat. At their May conference they opened the platform to any AI agent through an MCP server, with Anthropic as the first design partner wiring Claude straight into ServiceNow.

The benchmark. Microsoft is the real threat, not Salesforce. Copilot already runs across 160,000 organizations, and Microsoft can bundle it into software you already pay for. That is reach ServiceNow cannot match. But in the actual job, IT workflow, most Microsoft pilots stall before production. Salesforce makes more noise with Agentforce and is the smaller danger, because it attacks from the customer service edge, not the core. When Salesforce pushed into IT service management head on, it got a couple hundred signups against ServiceNow's roughly 8,600 customers and 40 percent share. The moat holds. The contested ground is the new horizontal layer, the control plane over all the agents, and that fight is genuinely unresolved.

Here is the hard part. NOW fails my line. It is software sitting on top of the buildout, not the buildout. I actually owned it and rotated out for that exact reason, a great company that is not Physical Layer. I had my dad in at 90 and out at 125, and I played the leveraged version out around 105, so I have nothing but respect for the trade. The stock is 118 today, down by a third this year. But respect for a company and a seat in this book are different inputs to different rules. Best in class. Wrong book. Watchlist.

IBM. A dividend with a science project attached.

IBM is the value answer in a momentum room. It is cheaper than everything else on this list at about 24 times forward earnings, it pays a 2.2 percent dividend it has now raised for 31 straight years, and the AI book is bigger on paper than people think, over 12.5 billion dollars signed to date. The catch is what that book is made of. Most of it is consulting, bodies doing implementation work, not high margin software. Consulting grew 1 percent. Total company growth is mid single digits. The re-rating already happened, and then the stock gave a big chunk of it back.

It fails the line too. IBM does not sell the power, the silicon, or the cooling. The one piece that even brushes the thesis is quantum, where IBM is a genuine leader, building real hardware toward a fault tolerant machine around 2028. Treat that as a free lottery ticket stapled to a dividend, not a reason to own the stock. Both NOW and IBM are quality. Neither one is Physical Layer. That is allowed to be the answer.

AUR. Ahead of the competition? Mostly yes.

Aurora runs driverless Class 8 trucks. Still early, but ahead of the competition, and that is fair with one asterisk.

Ahead on the hardest version of the problem, open highway long haul, yes. Aurora has been fully driverless on the Dallas to Houston lane since April 2025, it is now across about 12 routes in the Sun Belt, and it has logged 250,000 incident free driverless miles running day, night, and rain. The edge is the sensor. Its FirstLight lidar sees about 450 meters out and reads speed straight off the return, which hands a driverless truck several extra seconds of reaction time at highway speed that a human never gets. The scale bridge is the moat: a hardware partnership with Continental and NVIDIA to mass produce the kit starting 2027, trucks built with both PACCAR and Volvo, and freight already booked with Hirschbach, Werner, Schneider, and Uber Freight.

The asterisk. Ahead does not mean alone. Gatik got to fully driverless at scale first, in short haul. Kodiak has more driverless trucks on the road today, in the oil fields. Aurora leads the hardest lane, not every lane. And the price tells you it is loved, a 15 billion dollar company on 1 million dollars of quarterly revenue and a 223 million dollar quarterly loss, sitting near its 52 week high, funded by selling stock. This passes the line as physical AI, the machines that move in the real world, but it is a swing, not an anchor. Size it like a lottery ticket you can survive going to zero, scale it only as it de-risks, and respect that one bad crash headline is the entire risk in a single line.

UBER. The demand layer, not the machine.

Uber is the perfect name to put right here, one slot under Aurora, because it teaches the whole filter in a single move.

Aurora builds the driverless truck. It passes. Uber books the load that rides on top of it. It fails. Same wave, different layer, and the line is the thing that tells them apart.

The honest part first. Uber owns no chips, no power, no data centers. It is a marketplace that clips a fee on rides and food. It does not force AI to exist. It monetizes the physical AI that other people build. By my rule, that is a fail, and I am not going to dress it up.

Now why it is still fascinating, and why it sits directly under Aurora in this post. Uber is the toll booth on autonomy. It is wiring robotaxis into its app instead of building the car. Waymo is already live on Uber in Austin and Atlanta, and Uber wants self driving rides in 15 cities by the end of this year. It signed a deal with NVIDIA to scale toward 100,000 robotaxis from 2027, running on NVIDIA's DRIVE chips. And here is the thread back to a name you already asked about: Uber Freight runs Aurora's driverless trucks, and Uber owns a stake in Aurora. The wave under Uber is the exact wave I buy.

The business is a cash machine. Profitable, about 9.8 billion dollars of free cash flow, bookings up 25 percent, and the stock is still down around 30 percent to the bottom of its range. The reason is one fear. If Waymo and Tesla decide they do not need a middleman and run their own apps, they route around Uber, and one shop on Wall Street thinks up to 40 percent of Uber's ride bookings are exposed to that. So the read is clean. Great company, cheap stock, wrong book. If you want this wave, own the layer Uber leans on, Aurora and NVIDIA. If you want Uber itself, hold it in a quality sleeve, small, and call it the exception it is.

KRKNF. The one that quietly works.

Kraken Robotics builds the eyes and the batteries for underwater war. Synthetic aperture sonar, towed minehunting systems, pressure tolerant subsea power. NATO navies are swapping crewed minehunters for autonomous ones, and Kraken sells the picks and shovels.

Here is what makes it rare. It is a small cap robotics company that actually makes money. Last year, revenue 102 million Canadian, up 12 percent, gross margin 62 percent, and positive EBITDA. Q1 grew 35 percent. It is sitting on a fat cash pile and a 615 million dollar acquisition that roughly doubles the business. The direct AI tie is fresh, a May deal to build automatic target recognition on top of the sonar, which is machine perception for the seabed. It passes the line as the physical layer of defense AI, perception plus power plus platform.

The catch is the chart. It is up over 200 percent in a year, it trades thin, and the acquisition still has to close and integrate. This is a buy on dips, not at the high. I want it on a pullback, call it the low 7s Canadian or better, sized small because the spread is wide.

TXG. My honest no.

10x Genomics. I am going to give the answer I would want, not the one that flatters the request.

10x is a genuinely good science company. It sells the instruments and the recurring kits that let researchers map single cells and tissue, 70 percent gross margins, over half a billion in cash, and a real new platform called Atera shipping later this year. The stock has tripled off the bottom.

It fails my line, and not at the margin. This is life science tooling. It does not force AI to exist. The only thread back to our world is that it generates the biological data that AI models train on, and that is a side story, not a reason to own it. Worse for the timing, the core business is basically flat once you strip out one time legal settlements, management itself guided revenue to step down for the next two quarters as buyers wait for the new machine, and the stock is at the top of its range. I respect the name. It is off thesis and late in the move. Watchlist, not book. That is the whole honest answer.

CEG. The nuclear name I owned, and sold.

This one I have a real history with, so you get the honest version.

Constellation runs the largest nuclear fleet in the country, and after closing the Calpine deal in January it is the largest power producer in America, about 55 gigawatts. It passes the line on sight. Power is the exact thing AI is short, and this is the power. The receipts are not slides. Constellation is restarting Three Mile Island for Microsoft on a 20 year contract, already more than 65 percent staffed, pulled forward to 2027, a year early. It signed a second 20 year deal with Meta off its Clinton plant. Two hyperscalers, two two-decade contracts, and a guide for more than 20 percent earnings growth a year through 2029.

So why did I sell it. Because out of all the names

Here is the honest other side, because you asked and you deserve it. There is a real argument they are not redundant at all. VST is Texas, gas heavy, levered to the spot power price. CEG is the PJM grid, nuclear pure, locked into those named 20 year hyperscaler contracts, lower drama. If you do not already own a nuclear name, CEG is the higher quality one to start with, full stop. I just could not own both and stay honest about my own cap. It also ran from under 50 dollars to 412 and has pulled back about 30 percent into the 270s, so even if you want it, I would scale into the 250s, not chase the bounce.

LEU. The fuel rod under all of it.

If CEG and VST are the power plants, LEU is the fuel.

Centrus Energy is the only American owned uranium enrichment company, and the first one licensed to make HALEU, the high assay fuel that every small modular reactor and advanced reactor design needs. Walk the chain and you see why it fits. AI needs electricity. Electricity increasingly means nuclear. Nuclear needs enriched fuel. And for 20 years America bought roughly a fifth of that enrichment from Russia. That door is closing, the Russian import ban phases out every waiver by 2028, and there is exactly one US owned company standing in the gap.

Washington knows it. Centrus just won a 900 million dollar Department of Energy award in January to build commercial scale enrichment, with options that could push it past a billion. It is sitting on a 3.9 billion dollar backlog out to 2040 and a 1.9 billion dollar cash pile. It passes the line as the deepest, most upstream node in the nuclear stack, the fuel under the fuel.

Now the honest no inside the yes. This is the riskiest name on the list. It swings up and down 50 percent on headlines, it trades at 64 times earnings even after halving from 464, it leans on government money that has to get appropriated every year, and both Urenco and Orano are building US enrichment to compete with it. It dilutes shareholders to fund the buildout. So this is not an anchor. It is the speculative tail of the nuclear trade, and you size it like a swing, not a foundation. I want it on the volatility, down in the 140s, not chased at 180 into an August earnings print. The thesis is real. The stock is a roller coaster, and you buy it that way.

The eleventh name. The machine under all of it.

Now the one I added. You keep asking me about NVIDIA from the side, through Marvell, through Boost Run, through Uber's robot fleet, through the anchors in my book. So let me draw the whole machine. Two questions answer it. What does NVIDIA feed, and what does NVIDIA need.

What NVIDIA feeds, and what NVIDIA needs. The whole machine on one card.

What it feeds. NVIDIA has quietly become the largest corporate investor on Earth. Over 40 billion dollars in equity bets in 2026 alone. Up to 100 billion committed to OpenAI as it builds. Up to 10 billion into Anthropic. 5 billion into Intel for a 4 percent stake. 2 billion into Nebius, which is NBIS, an anchor in my book. A billion into Nokia. A capacity backstop worth 6.3 billion under CoreWeave. And on Marvell, the 2 billion we just covered. The pattern is one loop: NVIDIA funds a customer, the customer buys NVIDIA chips, the revenue books, the stock rises, the cash funds the next one. Bulls call it an ecosystem. Michael Burry calls it circular and overbuilt. Both can be true at the same time, and you should hold both in your head.

One correction, because stale write ups get this wrong. NVIDIA actually sold out of several names late last year, including Recursion, Applied Digital, and its Arm stock. Selling the Arm shares did not cut the dependency, and that is the bridge to the second question.

What it needs. This is the part that makes me money, because the dependencies are public and ownable.

• TSMC, ticker TSM. Single source for the leading edge chip, and more important, the advanced packaging that stacks it. NVIDIA has booked roughly 60 percent of next year's packaging capacity. There is no second supplier at this volume.

• Memory, the high bandwidth kind. Three makers in the world. SK Hynix leads, Samsung is second, and Micron, ticker MU, the only US listed pure play and another anchor in my book, is qualified for NVIDIA's next platform.

• ASML. The only company that makes the machine that prints the chip. Ninety percent of all lithography. A monopoly two layers up.

• Optics. As the racks move from copper to light, NVIDIA leans on Coherent and Lumentum for the lasers. Lumentum, ticker LITE, is already in my book.

• Power and cooling. Every next gen rack runs past 300 kilowatts and forces liquid cooling, and Vertiv, ticker VRT, is the lead partner. Behind it sits the real bottleneck, electricity itself, which is the entire reason I own VST, why CEG is the name half of you keep asking about, and why LEU is the fuel under all of it.

• The CPU brain still rides Arm, ticker ARM, one of my five anchors. NVIDIA sold the stock and still pays Arm a royalty on every server CPU it ships.

So when you ask me about NVIDIA, the honest answer is I would rather own the road it drives on than the car. The cleanest ways in, ranked by the quality of the moat, are TSM, ASML, MU, VRT, ARM, and the optics names. MU and ARM are already anchors in my book, VST is the power under it, and LITE is in my pivot sleeve. The two I do not own yet, TSM and ASML, are the best shovels I am not holding, and that is the honest gap in my book. I built it around this map before I ever drew it for you.

My top 3

Eleven names is a lot, so let me be useful. If you only move on what I would move on, here are the three.

The three I would actually act on, ranked. The kicker is the one nobody asked for.

One. KRKNF. Kraken is my favorite of everything you sent. It is the rare small cap robotics name that actually makes money, it passes the line clean as subsea defense AI, and it is growing 35 percent with a deal that closes this quarter. The only thing wrong with it is the chart after a 200 percent year, so I want it on a dip, the low 7s Canadian, sized small because it trades thin. Of the eleven, it is the one I would buy first.

Two. MRVL. The best business on the list, full stop. NVIDIA owns a piece of it, the custom silicon and the optics are the literal plumbing of every AI rack, and the growth is real and still rising. The only thing wrong with it is the price right now. So it is not a buy at 300. It is a buy in the 200s, and I am patient.

Three. TSM. Here is the kicker. My single favorite name in this whole post is one not one of you asked for. Taiwan Semiconductor builds every leading edge chip and owns the packaging NVIDIA cannot get anywhere else, and NVIDIA has already booked most of next year's capacity. It is the closest thing there is to owning the entire buildout in one ticker, and it is the gap in my own book I am working to close. You called eleven names. The best one might be the twelfth.

The Request Line, from here on

This is volume one, and I am making it a monthly thing. I am calling it what it already is. The Request Line.

Here is how it runs. The floor stays open in the room. Free readers drop any name in the comments, and I cover it, run it through the line, and give you the structural read, exactly like the eight above. The Founding 50 get three things on top. Your request gets answered first. Every passing name comes back with the full receipt, the entry band, the trim trigger, and the kill vector. And I run it through the committee for the actual conviction score instead of a one line verdict. Free users see the conclusions. Pro users see the work.

That is the trade. You tell me what to look at. I do the work in public and keep the receipts.

Your call

Eleven names. One rule. Seven passed. Four good companies sat in the wrong book, and I told you so to your face, because that is the only version of this worth paying for.

So tell me where I am wrong. Of the seven that passed, which one would you actually put in your book right now. MRVL on a pullback. BRUN on a flush. KRKNF on a dip. LEU as the fuel-layer swing. CEG if you do not already own nuclear. Or one of the NVIDIA shovels I do not even own yet. And drop the name you want me to run next month. The loudest ticker in the comments goes first.

I buy the exact raw materials, hardware machines, and power plants that force artificial intelligence to actually exist and operate in the real world.

Real money. Real names. Real receipts.

You're in the room or you're not.

Subscribe to the Request Line

The floor is open. Every month I take the names you drop in the room and run each one through the same rule that governs my book. Real receipts on every read, the honest no included.

Drop your ticker in the comments. Free readers get the structural read on every name.

Subscribe free and call the next one in.

Founding 50

Pro is 99 dollars a year, locked for life, for the first 50 seats. It goes to 199 after they fill. The switch flips on July 22, my 18th birthday.

Pro readers get the Request Line in full: your name answered first, the entry band, the trim trigger, and the kill vector on every pass, plus the committee conviction score instead of a one line verdict. Not AI astrology. Not a 1 to 10 vibes score. The actual work.

Claim a Founding 50 seat.

Poll (Substack widget): Which name should I run through the line first next month?

• CEG, the nuclear name I sold

• MRVL on the pullback

• LEU, the fuel-layer lottery

• An NVIDIA shovel (TSM / MU / VRT / ARM)