What We Told Hedge Funds This Week

Around the world of macro, before we get to the fun part

A lot happened this week, so let me take you around the world of macro first… not least because the setup we described going into this morning’s payrolls report has now been settled, and settled OUR way.

First, the Middle East. We are on the verge of MoU 2.0, and it feels remarkably reminiscent of the first one: a truce covering oil exports through the Strait of Hormuz, this time with a 60-DAY grace period during which transits are essentially toll-free, safeguarded by a combination of the Iranian regime and Oman, and with the US notably keeping its hands OFF the arrangement compared to the first MoU.

Recognizing the problems with the last deal makes me more comfortable that this deal survives longer. But should you celebrate by shorting crude with both hands? No. The backdrop is entirely different from March-April: positioning has already been washed (look at the chart below — the market came into this close to FLAT after the July article we wrote), so there is a limit to how far flat price can fall on a deal alone. The key question is not the grace period; it is what Iran intends to CHARGE after it. The current war-risk regime prices transit at roughly $12 per barrel on the most expensive routings (from the Strait), and therefore, there is clearly a LIMIT to how far the Iranians can take this, if they want to remain relevant.

So we FADE the second MoU in flat price terms. The real bear case is in refined products: gasoline and diesel cracks are where the war premium actually lives, and if the strait normalizes, it is the cracks that deflate first and hardest. Short refining margins, not Brent into a Chinese restocking bid.

Second, the July momentum rout. I will admit it on the record: I fell asleep at the wheel to some extent, because the price action felt SO at odds with the fundamentals coming into July that I struggled to take it seriously... and then it steamrolled everything anyway.

We now know why (one Leopold, 3-400k mini-Leopolds, and the biggest tech-book readjustment in the history of the Goldman prime data), and we also know what it was NOT: information. Positioning in the Mag7 got outright SHORT relative to normal times at the trough, and the recovery since has come for the right reasons, namely the reporting season. If you believe the order books of the three big clouds, Microsoft, Google and Amazon have gone from roughly $500bn of combined RPO/backlog a year ago to $1,630bn as of Q2. The backlog TRIPLED in two years.

You do not get a capex winter with a $1.6 trillion order book, and without Mag7 capex there is no infrastructure trade at all. So keep this chart within eyesight whenever someone shouts “bubble” at you.

Third, the Fed. And this is where the week turned into a home run for the nowcasting franchise.

Warsh has been hinting he could support a hike, Kashkari said EXPLICITLY that he would back raising rates, and the 1-year inflation swap is priced as if inflation falls apart from here... while our nowcast has US inflation probabilities in the SINGLE DIGITS and the pipeline pointing to another roughly 0.0% monthly CPI print next week.

I have never seen a discrepancy of this magnitude between what the Fed is signaling and what the inflation pipeline is doing. Into this morning’s payrolls, consensus stood at 80k, our withheld-tax indicator had us with the SOFTEST forecast on the street… and the actual print came in below even our call: MINUS 23k for the month.

The breakdown tells you exactly what kind of report this was, namely the World Cup HANGOVER report we warned about up-front: services even weaker than an already weak June, with Leisure & Hospitality down roughly 40k as the tournament hiring unwound on top of punishing seasonals.

Meanwhile, goods and manufacturing are quietly GAINING momentum (look at the green diamonds versus the blue dots in construction, manufacturing and durables below)... which is precisely the rotation our nowcasts have been flagging: weak(ish) services, firming goods.

And do note the government line before you panic about the headline: MINUS 50k from government alone makes the -23k look considerably more dramatic than the private economy actually warrants. Soft, yes. Collapse, no. But soft is all the front end needs.

So the scoreboard reads: softest call on the street on jobs, verified. A 0.0% CPI pipeline into next week, pending.

As I said this morning: we expect the Fed to be DOVISHLY surprised on both jobs and prices before September, and NO HIKE has to be the base case for the rest of the year accordingly. Insinuating a hike into this data was not hawkishness; it was a category error… and the front end is now doing the correcting for them.

Portfolio in focus: the Compute case… the market is repricing the megawatt

Now to my favorite subject of the week, because it ties every thread of our AI book together: the PRICE of compute. Bernstein came out this week with an analysis of pricing per megawatt across the AI clouds, i.e., the annualized revenue per contracted megawatt across CoreWeave, Oracle, Google Cloud and the emerging neoclouds, and the direction of travel is UP. Not stable. UP. In a market where the consensus narrative is “AI compute gets commoditized, and prices collapse”, the actual deal flow shows customers paying more per megawatt-year with every vintage of contract. That is what a supply-constrained market looks like, and it is the single most underappreciated data point in the entire debate.

So we pulled the ENTIRE deal flow: every announced contract where value, tenor, and megawatts are disclosed (or estimable), from the converted bitcoin miners to the neoclouds to the hyperscaler mega-deals, and computed the implied annualized dollars per megawatt.

The market for a megawatt-year turns out to have THREE floors, and knowing which floor a company lives on is the whole game. The bottom floor is the LANDLORD layer: WULF, Cipher, Hut 8, and Core Scientific, renting powered shells on 10-25 year leases at a remarkably tight $1-2.3m per MW per year (Cipher’s 39MW Fluidstack expansion at ~$2.1m, WULF’s 25-year JV at ~$2.3m, all of it Google-backstopped… the miners have become bond proxies with cooling towers).

The middle floor is the FULL-STACK layer: Nebius-Microsoft at ~$11.6m, IREN-Microsoft at ~$9.7m, Oracle-OpenAI at ~$13.3m per MW-year…five to six times the landlord rate, because the GPUs and the cloud layer live here.

And then there is the penthouse: Anthropic paying SpaceX $1.25 BILLION PER MONTH for the 300MW Colossus 1 cluster, GPUs included and available NOW... an implied ~$50m per MW-year, roughly 4x the term-contract rate. THAT single red diamond is Bernstein’s rising-price-per-megawatt finding compressed into one print: when immediate capacity is worth 4x contracted capacity, you are not looking at a commoditizing market; you are looking at a shortage with a price discovery mechanism.

(Do note: asterisked capacities are our estimates, scopes differ wildly, the AWS expansion prices low because it is usage-on-existing-infrastructure, and yes, the log scale is doing work in this chart... it has to, the market spans 50x.)

And the pricing power is not confined to the megawatt. NAND spot has broken out of its spring consolidation to fresh highs, and SanDisk’s print this week deserves a paragraph of its own: they beat by MILES again (revenue +372% Y/Y, an 84.6% gross margin IN NAND), and instead of squeezing every last dollar out of clients while spot goes bananas, management is introducing FLOORS AND CAPS in their pricing agreements… deliberately trading peak pricing for contract length.

The market sold the stock 8% because the guide was “only” in line, and I think that is exactly backwards: floors and caps take a great portion of the cyclicality out of an industry the market still insists on pricing as cyclical. Credit where credit is due… lengthening the cycle instead of squeezing it is the smartest thing a memory company has done all year.

On the leverage panic: my colleague Jamie has been out with good work on system leverage and real rates, and I want to chime in on the less-worried side of that debate. Nominal margin debt looks scary.

Margin debt relative to MARKET CAP, and especially its second derivative, looks outright conservative versus 2000, 2007, and 2021… the crowd is borrowing LESS aggressively against rising collateral than in any previous mania.

July was concentrated leverage blowing up (one fund and a Korean retail complex), not systemic leverage. And on the China-is-winning-AI scare that refuses to die: it is still built on OpenRouter’s 2% sample of price-chasing tinkerers, and even if Chinese open-weight models DO win the frugal end of the market, Jevons paradox says total token demand goes UP, and the compute bottleneck gets TIGHTER.

Cheaper tokens are an argument FOR the hardware complex, not against it.

Portfolio update: the drones woke up, and end-consumer scarcity is doing exactly what we bought it for

Two sleeves carried the book this week, and both deserve a word.

First, DRONES. DroneShield finally caught a proper bid, jumping 10% on Monday and snapping the brutal stretch that followed the July 28 guidance disappointment. The last week brought the kind of news-flow that separates a broken story from a merely oversold one: committed 2026 revenue of AUD 206m already on the books (95% of ALL of last year’s revenue, with a growing recurring share), two new European vehicle-mounted counter-drone contracts through their Benelux partner including a fresh AUD 23m order, the RfAI-3 broadband detection platform unveiled, and a Rear Admiral joining the board to soothe the governance nerves.

The ASIC investigation overhang is real, and it is why our conviction sits at 7 and not 10... but a 360% revenue quarter, a AUD 2.2bn pipeline, the US Safer Skies Act opening law enforcement demand and a NATO verified-supplier pool later this summer is not a company whose THESIS broke. The half-year report on August 26 is the next test. Overall, it feels like buying Drones here is a good idea...

Second, END-CONSUMER SCARCITY, our sleeve for the simplest idea in the book: when memory triples in price, the companies selling the DEVICES become inflation assets, provided they can pass it through. And they can. Dell is now up 81% on our May entry, HP up 30%, Apple up 7%... all three sitting at conviction 10, and all three rebounding hard over the last week for reasons that are fundamental...

The pass-through evidence keeps piling up: Dell hiked commercial prices 15-20%, HP is absorbing memory at 35% of bill-of-materials and still guiding through it, Apple raised Mac and iPad prices in June and, crucially, demand did NOT collapse… Scale is consolidating the market to exactly these names (IDC expects share shifts toward the biggest vendors with inventory and supplier leverage), and the sell side that downgraded the group at the lows in December is now chasing it back up. The Samsung mobile division just posted its first EVER quarterly loss because it has to buy memory at market prices... THAT is the input hurricane, and our ECS names are the ones holding umbrellas they charge admission for. We flagged this rotation when we built the sleeve in May: the memory trade has a second leg, and it lives in the device makers.

Third, and this deserves its own paragraph now that the data has come to us: the WEAK USD bet embedded in the book. Between friends, some 15-20% of the portfolio sits in the Decoupling sleeve, and that sleeve is, functionally, a soft-dollar basket: Energy Fuels and Barrick (up 15% since the July add) on the debasement and hard-asset leg, IONOS in EUR (up 16%), Linde and Steel Dynamics (up 50%) on the reshoring-priced-in-a-weakening-currency leg.

Wolfspeed x LITEON: the lottery ticket just got a barcode

And then there is Wolfspeed, our power-semi lottery ticket for the 2027 leg…

Wolfspeed and LITEON announced a strategic partnership with Wolfspeed’s silicon carbide formally QUALIFIED into LITEON’s 800VDC power sidecar and compute-rack PSU platforms for hyperscale AI data centers, with the explicit potential for broader adoption across multiple cloud platforms. If that sentence reads technical, here is the translation: LITEON is one of the biggest power-supply manufacturers feeding the hyperscalers, the entire industry is migrating to Nvidia’s 800VDC architecture (because 1MW racks cannot be fed with legacy silicon without setting the electricity bill on fire), and the 800V bus more or less REQUIRES 1,200V SiC MOSFETs, which cut conversion losses by 25-40%.

Qualification into the PSU platform is how a components company gets designed into EVERY rack a hyperscaler deploys, not one deal at a time but by architecture. Add the pattern of the last few months… the 10kV MOSFET first, the GE Aerospace MOU on high-voltage standards, the fifth-gen MOSFET launch, a dedicated data-center team planted in Silicon Valley… and Wolfspeed is methodically converting itself from an EV-cycle casualty into the power-delivery layer of the AI buildout.

So, structured by asset class, our best ideas from here..

In FX: short USD remains the anchor, against the EUR on the inflation-divergence story (our nowcast leads the front end by ~31 trading days, and the impulse is pointing one way) and against the JPY on positioning.

In Energy/Commodities: fade the MoU 2.0 panic in flat price crude, express the deal via SHORT CRACKS instead, stay long TTF into the thinnest storage winter in 15 years, and keep gold as the debasement hedge.

In Rates: received in the US front end, full stop. A Fed that discussed hikes into a -23k payrolls print and a 0.0% CPI pipeline will be dovishly surprised on BOTH jobs and prices before September... no hike is the base case for the rest of the year, and the front end still has repricing left to do.

And in Equities: stay long the memory and hardware bottleneck complex (SanDisk’s floors-and-caps just de-cyclicalized the story), keep the ECS sleeve at full conviction as the pass-through compounders, accumulate the drones on governance-discount days, and let the tripled hyperscaler backlog do the arguing with the bears.

The book sits at +10.5% YTD as of this morning’s open, having survived the worst momentum month in recorded history, holding some of the exact names that were being liquidated... which is not a boast, but rather, the entire point of sizing (and a -23k payrolls Friday didn’t hurt the FX and rates legs either).

横向滑动查看完整图表