台账

Memory is a commodity cycle wearing a secular costume, and the cycle is turning where the wall says it should. The mechanism runs through the power bill: hyperscalers who cannot secure the megawatts defer the accelerators, the accelerator vendor holding thin inventory cuts its forward memory orders, and those cuts land just as the memory makers' new capacity arrives — margins built at peak pricing do not shrink in that world, they collapse. Micron sits at the receiving end of every link in that chain: it must spend heavily to build out production against orders it has not yet filled, its input is energy in an energy shock, and its product is priced by whichever competitor blinks first. The tells were on the tape by mid-July — the AI-capex rout ran straight through Asian semiconductor names, a Korean memory maker that listed in the United States at the top of the cycle fell more than eight percent in a week, and by the end of the month memory had sold hard with the sell-side turning. The expression is the stance itself, graded on the prints and deliberately not an options clock: a structural view on a theta clock is a donation, and the book buried that wrapper before it filed this thesis. The kill is pre-registered on the company's own numbers: two consecutive quarterly prints in which gross margin holds or widens quarter over quarter while next-quarter revenue guidance steps above the quarter just reported — a cycle that is not rolling over. The first scored test is the late-September print. Verdict due June 30, 2027.

Same cycle as Micron, without the secular story to hide behind. Storage is a commodity: the price is set by the marginal competitor's willingness to cut, the capacity comes on in lumps, and the customer is the same power-constrained buildout that is learning to defer deliveries. SanDisk had its run alongside memory and carries the same three burdens into the turn — heavy spending to build production against unfilled orders, an energy input repricing under it, and pricing it does not control — with none of the artificial-intelligence-demand narrative that lets a memory maker argue this time is different. That absence is the point of filing it separately: if the cycle read is right, storage shows it first and cleanest. It did on the tape — the week the AI trade cracked, this was the strongest grade on the demand side, down eleven percent on a single Friday. The expression is the stance, graded on the prints, and it shares one bet with the Micron filing under the book's concentration law: one thesis in two wrappers, sized jointly if ever sized at all. The kill is the same pair, on this company's own numbers: two consecutive quarterly prints in which gross margin holds or widens quarter over quarter while next-quarter revenue guidance steps above the quarter just reported. A cycle that stops rolling over ends the thesis. Verdict due June 30, 2027.

Held, not bought: legacy shares carried through the wall, filed as a claim so the hold can be graded. The mechanism is the endgame of the wall itself — the platforms that cannot get power from the grid go and build it, buying pipelines, funding small modular reactors, signing for geothermal, and become utilities behind an energy moat no startup can cross; that is what makes the large platform franchises the ultimate tollbooths of the thermodynamic age, and why the book's shopping list for a stress world starts with them. Alphabet is the specimen: an initiator of the AI computing revolution that treats power as its lifeline and invests heavily in self-developed energy, and the super-funder whose capital expenditure underwrites the backlog of the supply side this book owns. The tell arrived on schedule: the late-July print was good and the market did not care about the capex — communication services fell six percent on the week with the earnings cited as the drag — and the position was unchanged by design, which was the entire thesis; the next week the noise reversed hard. The expression is the founder's July rule for the mega-caps: keep the shares entered years ago, trim the rest into strength, and do not confuse preserving cheap old shares with a thesis that died — the thesis is still there, it is just not aligned with the short-term macro. The kill, in wall terms: Alphabet cuts its full-year capital expenditure guidance at a quarterly print. The day the super-funder stops funding the wall, the hold has no thesis. Verdict due December 31, 2027.

Held, not bought: legacy shares carried through the wall, filed as a claim so the hold can be graded. The mechanism is the endgame of the wall itself — the platforms that cannot get power from the grid go and build it, buying pipelines, funding small modular reactors, signing for geothermal, and become utilities behind an energy moat no startup can cross; that is what makes the large platform franchises the ultimate tollbooths of the thermodynamic age, and why the book's shopping list for a stress world starts with them. Alphabet is the specimen: an initiator of the AI computing revolution that treats power as its lifeline and invests heavily in self-developed energy, and the super-funder whose capital expenditure underwrites the backlog of the supply side this book owns. The tell arrived on schedule: the late-July print was good and the market did not care about the capex — communication services fell six percent on the week with the earnings cited as the drag — and the position was unchanged by design, which was the entire thesis; the next week the noise reversed hard. The expression is the founder's July rule for the mega-caps: keep the shares entered years ago, trim the rest into strength, and do not confuse preserving cheap old shares with a thesis that died — the thesis is still there, it is just not aligned with the short-term macro. The kill, in wall terms: Alphabet cuts its full-year capital expenditure guidance at a quarterly print. The day the super-funder stops funding the wall, the hold has no thesis. Verdict due December 31, 2027.

Held, not bought: the founder's long-held shares, kept through the wall and filed so the hold is graded rather than assumed. The mechanism cuts against the position, and the filing says so: in the last cycle the scarce input was compute; in this cycle it is electricity, and the chip vendor sits downstream of the watts — a hyperscaler that cannot secure power defers its accelerator deliveries, and the vendor holding minimal inventory is the first to feel it. That is the book's larger claim, that the economy cracks and the AI complex reprices into 2027. Nvidia is the vendor in that sentence, which is exactly why the hold is a separate thesis from the short book: the shares were entered years ago at a cost that makes them the cheapest claim on the endgame — the platforms that build their own power still need the compute — and preserving cheap old shares is not thesis death. The tells so far are the tape the rule was written for: an eight-hundred-billion-dollar session out of the AI complex in late July, then a fifteen-percent day in a mega-cap peer and a record-value session a week later — strength to trim into, by rule. The expression is the discipline: keep roughly a quarter of the old position and sell the rest of mega-cap into strength over the two months from July. The kill, in wall terms, is the deferral reaching the vendor's own guide: next-quarter revenue guidance below the quarter just reported, or data center revenue down quarter over quarter, at any quarterly print. Verdict due December 31, 2027.

Held, not bought: the founder's long-held shares, kept through the wall and filed so the hold is graded rather than assumed. The mechanism cuts against the position, and the filing says so: in the last cycle the scarce input was compute; in this cycle it is electricity, and the chip vendor sits downstream of the watts — a hyperscaler that cannot secure power defers its accelerator deliveries, and the vendor holding minimal inventory is the first to feel it. That is the book's larger claim, that the economy cracks and the AI complex reprices into 2027. Nvidia is the vendor in that sentence, which is exactly why the hold is a separate thesis from the short book: the shares were entered years ago at a cost that makes them the cheapest claim on the endgame — the platforms that build their own power still need the compute — and preserving cheap old shares is not thesis death. The tells so far are the tape the rule was written for: an eight-hundred-billion-dollar session out of the AI complex in late July, then a fifteen-percent day in a mega-cap peer and a record-value session a week later — strength to trim into, by rule. The expression is the discipline: keep roughly a quarter of the old position and sell the rest of mega-cap into strength over the two months from July. The kill, in wall terms, is the deferral reaching the vendor's own guide: next-quarter revenue guidance below the quarter just reported, or data center revenue down quarter over quarter, at any quarterly print. Verdict due December 31, 2027.

Owns the barrel and the refinery, in the week the barrel touched a hundred. The mechanism is physical: crude went through a hundred on a supply shock, not a demand boom, and the refinery is the tighter asset — diesel margins are at records. The second-quarter report already showed the flow-through: $17.2 billion of free cash on a capital plan the company has not raised. The thesis rests on cash, not on a gesture — nothing in the company's statements commits to a special payout. The risk is that the driver is a strait, and a strait reopens on a headline.

Heavy barrels on the safe side of the ocean. The mechanism is geography: a supply shock that runs through the Gulf — a strait, a pipeline, a drone — reprices every barrel that does not have to cross it, and the oil sands are the largest such pool on the continent, with pipe to the US Gulf Coast and refineries at the other end that were built to run heavy crude. Cenovus owns the sands, the upgraders and US refining, so it collects at both ends: scarcer crude, dearer diesel. The book already owns the barrel-and-refinery pair through Exxon; this is the same thesis with a different address, insulated from the Gulf by a border rather than a balance sheet. The honest tension is printed with the position: the strike narrative arrived in a packet the desk could not verify, the week's screen shows the market easing off a Saudi disruption, and Canadian heavy trades at a discount that can widen on its own pipeline politics. The expression is the common. Two kills, either one enough: crude settling below seventy before mid-2027 — the input goes, the thesis goes — or the heavy differential blowing out for ten straight sessions, which means the discount ate the shock. Verdict due June 30, 2027.

Rent, not blood. The buildout will burn gas as its marginal fuel — the turbines are the visible part of the argument, but the constraint is getting the molecule to the turbine, and the pipe that carries it charges a fee per unit moved regardless of what the unit is worth at either end. That is the point of owning midstream inside an energy thesis: the book's supply-side names are paid for supplying power; this one is paid for the volume that flows on the way, which makes it the least directional position on the long side and the one that collects whether Henry Hub is cheap or dear. Enterprise Products Partners runs fee-based midstream at scale — the tollbooth between the basin and the burner — and its economics are volume economics: the molecule pays the fee on the way through. The receipts so far are the mechanism, not a company print: energy led the tape through the July shock while this name sat with no print of its own, and the desk graded it honestly as untested. The crosswind was named at entry and has not gone away — a rising ten-year against a yield instrument — and it is a headwind to the price, not a kill to the thesis. The expression is the common, held for the fee stream. The kill is a volume fact: natural gas pipeline transportation volumes and total gross operating margin both down year over year at two consecutive quarterly prints — the flow that pays the toll falling in a demand bust. Verdict due June 30, 2027.

Charlotte's pick, adopted into the book: the moat is the tracker, not the panel. The mechanism is the wall's demand for watts arriving faster than firm generation can be built — utility-scale solar is capacity that gets built on a construction schedule, and the tracker layer is the engineered part of that build, while the panel underneath it is a commodity fought over by everyone. Nextracker is the global leader in utility-scale solar tracking systems, deliberately not in the hyper-competitive residential panel market: a technical moat, margins the desk called excellent, a record backlog, and demand heavily subsidized under the IRA — her strongest-fundamentals pick in the sector, and a candidate beneficiary of the rotation into grid infrastructure that the room called in July. The tells are the backlog and the margin, and the founder's standing directive was written before entry: track the coming prints and define what results justify scaling in. The expression is the common, unwritten on — the book's own lesson on this name is that a covered call in a bull market converts your best outcome into someone else's, and the right tail is what the thesis was filed to own. The kill is pre-registered on the two lines the moat claim rests on: reported backlog down year over year at any quarterly print, or gross margin down year over year at two consecutive quarterly prints. A tracker moat that stops filling its order book, or stops earning its margin, is a panel business. Verdict due June 30, 2027.

Charlotte's pick, adopted into the book: the moat is the tracker, not the panel. The mechanism is the wall's demand for watts arriving faster than firm generation can be built — utility-scale solar is capacity that gets built on a construction schedule, and the tracker layer is the engineered part of that build, while the panel underneath it is a commodity fought over by everyone. Nextracker is the global leader in utility-scale solar tracking systems, deliberately not in the hyper-competitive residential panel market: a technical moat, margins the desk called excellent, a record backlog, and demand heavily subsidized under the IRA — her strongest-fundamentals pick in the sector, and a candidate beneficiary of the rotation into grid infrastructure that the room called in July. The tells are the backlog and the margin, and the founder's standing directive was written before entry: track the coming prints and define what results justify scaling in. The expression is the common, unwritten on — the book's own lesson on this name is that a covered call in a bull market converts your best outcome into someone else's, and the right tail is what the thesis was filed to own. The kill is pre-registered on the two lines the moat claim rests on: reported backlog down year over year at any quarterly print, or gross margin down year over year at two consecutive quarterly prints. A tracker moat that stops filling its order book, or stops earning its margin, is a panel business. Verdict due June 30, 2027.

Listed apex of 24/7 geothermal baseload for off-grid compute; Thesis IX (exotic baseload) proxy, registered on the Sep 1 hyperscaler geothermal headline.