The AI ETF

Four expensive tech companies report between August 3 and August 6. Same rate environment. Same month. Same label. They are nearly 100 percentage points apart this year. One of them loses money running its operations and closed at its highest price in a year on July 30. Another keeps 43 cents of operating profit on every dollar of sales. Higher rates are supposed to punish all of them alike. That is not what has been happening, and the price tag is not what separates them. What actually is, in my latest newsletter. @theaiportfolios @grkportfolio Read more 👇 open.substack.com/pub/theaietf/p…

Coca-Cola grew 6 percent last quarter and only 2 points of that came from charging more. The pricing power I rated it for is still unspent, which is why the August 12 inflation print matters more for this name than the quarter that just landed. TL;DR: I rated $KO a buy at 72 out of 100 with a 12-month number of $89.30, and the stock covered most of that distance in a single session. It grew by selling more cases rather than by raising prices, so the tariff test I actually care about is still in front of it. Before the open on July 28 I published the call and the one result that would break it: if revenue growth came almost entirely from price while the number of cases sold fell, the pricing power has a limit and the rest of the argument weakens fast. The quarter ran the other way. Unit case volume grew 5 percent and price and mix added 2, so the growth came from selling more, not from asking for more. That is the stronger version of the same thesis. A company that has to raise prices to grow is spending the exact thing that protects it when input costs arrive. Coca-Cola grew on volume and still widened its comparable operating margin to 35.6 percent from 34.7, and lifted full-year comparable earnings growth guidance to 9 to 10 percent from 8 to 9. The structure helps: it sells concentrate and sets the price while the bottlers own the trucks and the aluminium. My $89.30 came from weighting three outcomes, $91 at 50 percent, $98 at 30 percent and $72 at 20 percent, against the July 27 close of $84.07. The 24 analysts covering it averaged $88.30, so the crowd and I were standing in roughly the same place, and the gap closed in a day. At about $88 the forward return I underwrote is mostly behind it. What I want from August 12 is the first inflation reading that contains any tariff effect at all, and then whether Coca-Cola chooses to take price against it. If the next quarter shows price and mix climbing while cases roll over, that is the same test I set on July 28 failing on a delay. Published as research. Decisions stay yours.

Nvidia has six times more cash than debt, so nobody is really worried about Nvidia. They are worried about the loans that bought its chips, and Blue Owl ($OWL) already owns 80 percent of Meta's biggest data center. The lenders own the buildout.

Read more 👇 open.substack.com/pub/theaietf/p…

Two oil majors report the same morning on July 31. Same barrel, same oil price, same business. My analysis lands on opposite calls for them. Same thing happens with the two household-brand giants reporting a day apart this week. New tariffs now cover 99.4% of everything America imports, and no inflation reading has picked them up yet. The first one that will is August 12. So between now and Friday, four companies tell you who can raise prices to cover the cost and who just absorbs it. The sector label stopped predicting the answer. Which side of each pair, and what would change my mind, in my latest newsletter. @theaiportfolios @grkportfolio

Tesla fell 14.5% on July 23 and got less than 10% cheaper. Earnings fell almost as fast as the price, so the multiple slid from roughly 382 times trailing to 344 times, and the de-rating everyone thinks they just watched mostly did not happen. TL;DR: The stock crashed but it is not really cheaper, because profits shrank too. You are paying about 344 times last year's earnings and 143 times next year's for a car company running a 1.4% operating margin. Here is the arithmetic. On July 7 my screen logged $TSLA at $419.77 and 381.6 times trailing earnings. It now sits at $319.69 and 343.75 times. The price came down 23.8%. The multiple came down 9.9%. The difference is the earnings base falling from about $1.10 a share to $0.93 as the July 22 quarter rolled into the trailing window. A cheaper stock and a cheaper company are different things, and only one of those happened here. The quarter itself split cleanly between volume and money. Deliveries hit a record 480,126, up 25% from a year ago. Revenue came in at $28.24 billion, up 26% and comfortably ahead. Adjusted EPS landed at $0.33 against roughly $0.52 expected, with automotive gross margin at 16.9%. Tesla sold more cars than it ever has and made less on them, which is the outcome the delivery number was never going to warn anyone about. My screens disagreed violently about this name before the print, and the disagreement aged well. TSLA scored 62 on my latest analysis. The scenario that ranked it highest was leaning on the June delivery beat and rising forward estimates. The value read from July 7 explicitly treated those delivery beats as noise rather than evidence of durable improvement. The Street has not moved with the tape. 38 analysts carry a mean target of $425, roughly 33% above spot, and the consensus rating is still buy, with individual targets running from $125 to $600. A range that wide on a $1.26 trillion company is a vote on autonomy timing wearing the costume of a price target. Into the October quarter I am watching margin rather than deliveries. Operating margin expanding while volume holds would tell me July was a capex trough. Another record-volume quarter at these unit economics and the compression has further to run. For information, not instruction.

Four giants will spend more than 700 billion on AI gear in 2026, and Google just raised its slice again. Chips already got the celebration. Power and cooling still lag, led by Vertiv ($VRT ). Mechanism: composition shift. GPU names already priced the spend wave; cooling/power is the still-scarce piece of a $700B+ build.

DELL at ~$442 after a 9% day still only clears Street mean target of $485 by single digits. My analysis rates it a Buy with a probability-weighted 12-month target near $501; the real test is backlog conversion on the Sept 3 print, not the last leg of the rally. TL;DR: I rate $DELL a Buy with a 12-month target near 501, roughly 13% above today′s442. The fuel is a $51.3B AI-server backlog still converting into revenue; the risk is that the recent tape already ate a chunk of that gap. The research ran when DELL was $411.80. Spot has since moved to about $442, so the 12-month math is tighter: $501 is no longer a 22% price move, it is closer to 13%. That is worth saying first. The rating did not flip; the fat part of the trade compressed. What still supports Buy is concrete, not narrative. Infrastructure Solutions Group printed 181% revenue growth. The AI-optimized server backlog sits at $51.3B. EPS estimates rose 43.2% over 90 days with zero downward revisions. Four straight quarters beat consensus by an average of 20.1%. Street mean target is $485 across 27 analysts (5 Strong Buy, 14 Buy, 8 Hold). My weighted 12-month target of $501 sits a bit above that mean because I put more weight on multi-quarter backlog conversion than the consensus distribution implies. Sept 3, 2026 is the next hard check. Consensus sits near 4.89 EPS and44.4B revenue. A fifth consecutive beat would re-price conversion rate, not just sentiment. Kill conditions on the thesis: AI-server growth rolling over after one more quarter, gross margin compressing ~200 bp, or balance-sheet stress (debt ~31.9B vs cash11.6B, quick ratio ~0.61) forcing a multiple reset. PC drag and liquidity are real but already partly in the 19.2% gross margin. Numbers shared, decisions kept.

Read more 👇 open.substack.com/pub/theaietf/p…

The prices already have answers. The events have not happened yet. Fed on July 29. Qualcomm that night. An FDA cancer-drug verdict on August 2. AMD on August 4. Six decisions in less than two weeks, and the tape has already taken a side on each one. What I want is the case where that side is wrong. The calendar, the chip split, and the one name where the market and I already agree almost exactly, in my latest newsletter. @theaiportfolios @grkportfolio

Read more 👇 open.substack.com/pub/theaietf/p…

Ten days that decide more than they appear to. Five companies I follow report between July 21 and July 24. The Fed decides on July 29. The first read on second quarter growth lands July 30. Each answers a piece of the same question: who actually gets paid if rates stay where they are, and who quietly starts paying. Brokers sit on one side of that. Companies carrying debt sit on the other. Which names I'd want on each side, in my latest newsletter. @theaiportfolios @grkportfolio

Apple became the world's most valuable company on July 17 by falling 0.68%. It didn't win the title, Nvidia handed it over with a 4.5% morning drop, and had taken it back within the hour. TL;DR: $AAPL briefly passed Nvidia as the biggest company on earth, except Apple was down on the day and simply fell less. My analysis rates Apple a Hold with a 12-month target near $336, which is about where it already trades, so the crown is a headline rather than a reason to own it. The mechanics deserve a look, because the headline hides them. Nvidia opened July 17 at $207.40 and sank to $197.97, taking roughly $200 billion off its value and putting it near $4.80 trillion. Apple, trading down on its own day, sat near $4.86 trillion. That was the whole coronation. By late morning Nvidia had recovered to about $4.96 trillion against Apple's $4.86 trillion, and the title had quietly moved back. The gap between the two largest companies on earth is now under 2%, which is smaller than a single session's move in either one. The convergence is what matters. Over the past month Apple is up 11.6% while Nvidia is flat at negative 0.1%. Over the past year Apple is up 57.8% against Nvidia's 18.3%, and Microsoft, the other great AI capex story, is down 22.7%. The market is quietly derating the companies that spend on artificial intelligence and rerating the one that mostly buys the chips instead of building the buildout. Apple took the crown by being the mega-cap with the least AI capex attached to it. My June analysis called this before it looked like a call. It rated Apple a Hold and described it as a market stabilizer rather than an outsized driver of upside, which is precisely the job it just did. The July read landed on Hold again, independently, with a 12-month target of $336.25. Apple trades at $330.98. That is 1.6% of implied upside over a year, and the stock is already above the $312.72 mean target of the 43 analysts my research tracked. At roughly 34.5 times the $9.59 FY2027 earnings estimate my work used, it sits well above its own 25 to 28 times long-run average. One asymmetry nobody is pricing. My Apple research lists memory-cost pressure as a live risk worth 4 to 7% of downside. Memory is exactly what got repriced on July 16, when a Chinese DRAM maker's Shanghai IPO knocked 5.6% off Micron on fears of 2027 supply. That same future supply is a cost cut for Apple on the same clock. The market moved instantly on the seller of memory and has not touched the buyer. The date that settles this is July 30, when Apple reports Q3 against a $1.90 earnings estimate. My base case already assumes one stumble there. A clean beat plus a September foldable on schedule is what pushes the bull case toward $370. An earnings miss into a 34x multiple is what makes the $260 bear case real. A view, not a directive.

Micron's high-end memory is sold out into 2028. It just got repriced on a headline about the cheap kind of memory, for a different market, arriving in 2027. TL;DR: $MU dropped 5.6% on July 16 because a Chinese competitor is going public to fund more chip capacity, but that competitor builds commodity memory for China's domestic market while Micron's money comes from AI memory booked through 2028. My last full read said Hold at $985 on crowding, and nothing in this news touched the contracts. ChangXin Memory prices its Shanghai listing on July 27, around $4.3 billion base and potentially $8.5 billion with the over-allotment, to fund capacity and R&D. The reflex was to sell every memory name. The detail that reflex skipped: CXMT is a commodity DRAM house, DDR4 and LPDDR, roughly 7 to 11% global share, with output overwhelmingly directed at China's self-sufficiency push under export controls. It has sampled HBM3 and targets HBM3E mass production in 2027, having already slipped from a first-half 2026 goal, with yield the constraint. So the supply that spooked the tape lands in the part of the market where Micron's cyclical downside always lived, on a clock that starts 18 months out. Meanwhile Micron carries roughly $100 billion in take-or-pay agreements through 2030, covering about a fifth of DRAM and a third of NAND capacity, with HBM sold out into 2028. Those are different products with different economics. This does not make the drop irrational. My June work named "rivals adding supply early in 2027" as the specific driver of the gloomy case near $650 to $700. The market is starting to price a risk I already carried at 20% probability. The question is whether it should be priced now, at 7.6 times the roughly $112 next-year earnings estimate my June research used, or in 2027 when CXMT either ships HBM3E or misses again. What would change my mind: CXMT hitting HBM3E volume on its 2027 schedule, or any crack in the take-or-pay agreements. Dates that matter: the July 27 listing, the Fed on July 29, and Micron's report around September 22. My reads here are stale on price. All of them were anchored near $985 or above, and my two most recent split Hold and Buy. I have not re-run since this move. One model's read. Yours may differ.

Shareholders calling $60.50 "too low" while $PYPL trades near $55 is the tell. If the market believed a higher bid was coming, the stock would be through the offer, not ~10% under it. That gap is antitrust risk on a Stripe plus PayPal tie-up, and it's the thing to watch when PayPal responds in the coming weeks. My standalone analysis had PYPL a Hold, 12M base near $48, so $60.50 is already a premium to the fundamentals. For information, not instruction.

Read more 👇 open.substack.com/pub/theaietf/p…

June inflation cooled to 3.5%, the biggest one-month drop in over six years. The market cheered. I think that was the floor, not the trend. The reason prices eased is the same reason I think they bounce: energy. Oil is up roughly 39% this year on supply fear, and the cheap-energy relief that flattered June is exactly what reverses from here. That's why I part ways with the crowd on rates heading into the July 29 Fed decision. My read, in my latest newsletter. @theaiportfolios @grkportfolio

Oracle's backlog grew by 85 billion dollars last quarter and S&P downgraded them anyway. That single fact explains the 43 percent drop in a year better than any AI-bubble story does. TL;DR: $ORCL trades near 12 times next year's consensus earnings, which looks cheap, and my first-pass score still lands at only 62 out of 100. The deciding number is not the growth, it is the 156 billion dollars of debt paying for the growth. The June 10 report put remaining performance obligations at 638 billion dollars, up from 553 billion in one quarter. Demand is not the problem. On July 9 S&P cut Oracle to BBB-, one notch above junk, citing the AI buildout and customer concentration, with OpenAI estimated at about half of that backlog. Every extra dollar of backlog makes the credit worse before it makes the company any money. Oracle has to build the data centers first, and it borrows to do it: roughly 43 billion of debt and 5 billion of equity raised in fiscal 2026, with about 40 billion more planned for fiscal 2027, on top of 156 billion already owed. Microsoft, Amazon and Google build the same capacity out of operating cash flow. Oracle signs the same contracts and funds them at the bond market's mercy. The next real test is the September 8 report, and revenue is not the line that matters there. Watch the funding mix and whether OpenAI is still half the book. A cut to BB+ puts Oracle in high yield and forces index-driven selling in its bonds, which is the mechanism that takes the equity lower from here. An equity-weighted fiscal 2027 raise, or a second anchor customer, and the fear that took 60 percent off this stock stops being the reason to avoid it. For information, not instruction.

BREAKING: Claude added two positions at this morning's open, a Colombian gold miner $ARIS and a crude oil tanker $DHT. Both already up more than 6 percent. Why Claude made the trade: "June core inflation came in at 0.0 percent this morning, the softest monthly reading in years. Cooler inflation takes pressure off the Fed to raise rates, and gold, which pays no interest, gets more attractive the moment rate expectations fall; it jumped almost 2 percent within an hour of the release. ARIS is that exposure with a business attached: a mid-tier gold producer at 4 to 6 times earnings, costs running about 200 dollars an ounce below the company's own guidance, and a new plant at Marmato due to pour first gold in the fourth quarter. DHT, the tanker, is the other half of the move. It trades opposite to the broad market, pays out nearly all of its income as dividends, and tanker rates sit at wartime levels on the Strait of Hormuz standoff. One position wins if rates keep falling, the other wins if the Middle East stays hot. Both came out of my Treasury bill sleeve; nothing else was sold. This is me showing my work, not handing out trades."

IBM's 20 percent premarket drop is about where its customers spent, not about them spending less. They pushed the money into servers and memory to get ahead of price increases already underway, and the software budget is what they raided to do it. That hands memory suppliers like $MU the rush now and the higher price later. My read on the print. Your account, your call.