What We Told Hedge Funds This Week
“Scott has started QE... just out of the Treasury building”
There is plenty to worry about, and still the book is up a handful of percentage points since the last letter. We are back in the QE-like regime; the war is past peak escalation, with the US and Iran talking about a phased reopening of Hormuz, and Korean chip exports just set a new record. Everything you need to know right now is that we are in a QE-like environment, and Scott Bessent has started it out of the Treasury building.
The regime, the dollar and the oil space
We moved from slipping between Lame Duck QE and QE-like for a couple of weeks in September to a decidedly QE-like regime. What moved the needle is that the inflation we track on a daily basis is moving slightly lower again. That is very helpful for the QE-like regime, which is, just to remind everyone, a very dollar-negative regime and, to some extent, a risk-friendly one in proxies like Bitcoin, gold, and equities. Lame Duck QE was the dominant regime for a short while: there is still liquidity in that regime, but it is not really working, which is why I call it a lame duck.
Admittedly, we have been slightly wrong-footed on the dollar view this month. But some of the dollar proxies, crypto in particular and gold, have fared okay since we turned dollar-negative back in July. Remember that back in July we had the first move to QE, like, exactly because inflation started to look more benign. The same thing has happened again.
The difference between the two regimes matters for the book, so let me spell it out. The three upper lines of the table, EURUSD, USDJPY, and the dollar versus the Swedish krona, represent our own neck of the woods, and they all point in the same direction: this is a weak-dollar environment. Fundamentally speaking, I know it is challenged right now by some of the energy moves, but this is an environment where you can still lean into the short-dollar trade, also in crypto proxies, gold, and silver.
And most importantly, if you look at the S&P 500 or world equities, we moved from a very neutral equity environment to one where the upside is five times as big as the downside when we scan through comparable regimes historically. That is a big shift and one you should take notice of in your portfolio, especially since the positioning is as bearish as it is.
Now to the oil space, which has been choppy over the past 24 to 48 hours. First, another Houthi attack on Saudi infrastructure. Then, a couple of hours before I wrote this, Reuters reported that the US and Iran are in negotiations around a phased plan for reopening the Strait of Hormuz. Our base case remains that the Iranian Revolutionary Guard is running out of money, and that changes the setup for a negotiation: both parties now have something in common they want solved, at least temporarily. It will probably be a limited or phased deal, and the US will not repeat the mistake of giving the IRGC five weeks to sell oil. But the notion that this war moves in sine waves is very much alive, and the past day or two mostly confirms that we are past peak escalation.
Then the diesel export ban. You cannot nowcast a decision like that; when it shows up, it shows up. Our read is that it was a test run to gauge the market reaction, and the reaction was more violent than Trump and his admin had hoped for. Two things surprised me when I went through the data. Europe is actually not on the receiving end of a lot of US diesel; Latin America is much more exposed geographically.
And the biggest losers in the equity space are not in Europe at all but on the US Gulf Coast: Trump is essentially telling the refiners in Texas and Louisiana that they cannot sell their diesel to the client willing to pay the most. That is bad for Valero and the likes, and we have been leaning short those refiners for a few weeks. Europe will replace the barrels, most likely from Nigeria and India, by paying up. Cracks go up in Europe, and so do the margins of the European refiners. The disinflation Trump buys at home becomes inflation in Europe, and from a cynical perspective, I understand why he is doing it.
So this is the setup. Inflation is not looking that dire outside of energy (especially the case in the US still); core is still pretty benign in our readings, and that goes for the other aggregators of inflation data as well. We have an acceleration of trade flows, especially within semiconductors. And we have a very bearish setup in positioning: the market is very long energy and very long dollars. The World War Three playbook is back on in many portfolios. If we are right that we are past the peak of escalation, and we end up with weekly hopeium in the form of headlines that a deal is approaching, then this reminds me a lot of early April, when we got the first benign wave of hopeium out of the negotiations.
The game changer in my thinking this month is a chart regular readers know I bang the drum on: South Korean exports, the best live gauge of earnings we have, especially within technology and semiconductors. The first ten days of September delivered a new record on daily exports, higher than June.
Remember how everyone called the peak in AI in June because this chart rolled over? We are now accelerating again. In year-on-year terms, it looks even more incredible, because September is typically not a strong export month: we are talking something like 250% growth in chip exports. Stunning numbers, and Micron reports next Wednesday.
Yes, there are early signs of what was developing ahead of 2022: inflation, central banks responding, an inverting curve. But remember the leads and lags. The gap between the first hikes and the response in the economy is typically 9 to 12 months, which leaves plenty of time to harvest returns.
Why Bessent is leading the QE-like scenario
It is increasingly clear to me that Scott Bessent is orchestrating something very similar to QE, just out of the Treasury building instead of the Fed building. It is probably also why he never wanted the Fed job when it was offered. At the Fed, he would need a committee to vote for it. In his own building, he has the mandate to tweak a great deal on his own, and that is what he has done.
Here is the point I made to clients this week. It does not matter whether he uses the TGA, the idle dollars already sitting in the Treasury building, or whether he issues T-bills to fund the buyback of T-bonds. Both scenarios end in de facto QE. If he uses the TGA, the cash lands as bank reserves directly. If he issues a lot of bills, the Fed has an almost automatic program in place, reserve management purchases, under which it bought bills earlier this year and late last year, and it will restart if bill issuance rises.
So he has the upper hand: he can issue bills and force the Fed to buy them indirectly, or he can use the cash he already has to buy back bonds. In both cases, more dollars arrive. That is why the regime is labelled QE-like: it is like QE; it is just not called QE. I mean that literally. We are in a QE environment.
Has it been a success in containing bond yields yet?
Not really. But it would have been worse if he had done nothing, and he has plenty of firepower left. I am turning more upbeat by the week, not least because Bessent reiterates that he will increase the pace every time he is asked.
Note what it does to funding markets: reserves are landing in the system, private money growth is already back above trend, and you see the first, very, very first signs of something slightly more constructive for bonds and credit markets. I am not turning upbeat on bonds yet. I wrote that last week, and I stand by it, but the direction has changed. And that is the most important thing for the names in our book that live and die by financing markets, which is the In Focus below.
Has the dollar trade worked? Versus the yen or the euro, we are slightly in the money since entering it. It has worked in gold, and it has particularly worked in crypto. Everything about the Bitcoin trade right now is a response to what Bessent has done.
The divergence between Bitcoin and gold is material, and I have to admit I picked the wrong horse to some extent: I am invested in both, and if you count the equity positions linked to gold, I probably have more gold and natural resources than Bitcoin... But the fundamentals are the same for both, and they are getting stronger by the week. Everything you need to know right now is that we are in a QE-like environment, and Scott has started QE out of the Treasury building.
In focus: IREN, the neoclouds and the compute case
IREN is the name most of you asked about ahead of this letter, by some distance. It was also, to my surprise, one of the most traded names on a large Scandinavian platform last week, even though we seem to be among the few in this part of the world talking about it. So let me explain why it is still in the portfolio when it has not really performed for the past year. Everything AI has performed; IREN not so much.
Look at IREN like this. It is a data-center operator: a Bitcoin miner earlier, now a neocloud. It has sold a lot of its capacity forward at a fixed price to Microsoft, and a little bit to Nvidia. So it has a fixed revenue side, or at least a very high degree of visibility on the revenue side. The issue, and this is where I have to criticize management, is that when you lock in your revenues, you also need to lock in the cost side related to that deal. And they have not done that. They have not issued all of the debt needed to build this out; there is still roughly 8 billion left to raise. They have bought what they need from the semiconductor companies for 2026, but the revenue is very locked in relative to the cost base in 2027, which is made up of funding costs and hardware costs. And we have seen interest rates go up and credit spreads blow out for some of these AI-related names.
I come from investment banking, and one of the first things we’d tell managements was never to run one leg fixed and the other variable. Investors dislike it because it hands them the fluctuations.
Take the extreme case: rates at 25% and data-center construction costs up 300-400%. IREN would be in serious trouble, because it would still have to deliver to Microsoft and keep building while the revenue side stayed locked. It could try to renegotiate, but from a very weak position. A light version of exactly that has unfolded over the past year: the revenue side was locked, and the cost side went up. And on a fixed-price contract, every cost shock comes straight out of the margin. Just take a look at Oracle, which has had to abandon a data center project in New Mexico due to the very aforementioned reasons here.
To some degree, this holds true for Cipher and a few of the other names in the space as well: a larger degree of fixed revenue versus floating costs. The comparison table below shows who carries the most of it. CoreWeave is the extreme case, funded almost entirely with debt and with an interest bill last quarter that exceeded its operating result.
Nebius is the opposite, with customers prepaying 50-60% of the build cost on its landmark deals. IREN sits in the middle on funding, but it has one thing neither of the other two has to the same degree: its own grid connections, around 4.5 gigawatts of secured power. Power, not GPUs, is the scarce input, and that is what a hyperscaler cannot buy quickly at any price.
So why do I keep it? Because the whole QE-like environment that Bessent is orchestrating is exactly what a name like this needs. IREN is so dependent on functioning financing markets, and Bessent is moving in that direction. I am not turning very upbeat on bond markets yet, but you see the first signs that are slightly more constructive for bonds and credit, and that should be very good news for IREN. If we get into this QE-like scenario and it solidifies, this is probably one of the names with the most asymmetric outcome space in the book, because it has not moved for a year, and that is simply because its costs have been rising against locked revenues. So it stays in. But I have generally lowered my exposure to the neoclouds as a consequence of this asymmetry between revenues and costs, and I will not add to IREN until the roughly 8 billion of funding is clear.
And the compute case itself is intact; if anything, it is stronger. Data centers are up about 48 billion since 2023, while every other category of US private construction is down about 124 billion. This is what is holding up the construction cycle. The frontier debate does not change that: if Anthropic slows the frontier, my best guess is that it spends more on inference instead, because anyone using Claude these days can feel the lack of inference capacity.
Inference runs 24/7 and needs power and installed capacity. That is IREN’s side of the remix. The exposed side is that its customers rent the newest GPUs, and a slower frontier could mean lower rental prices for flagship chips on renewal. Both sides are true, and the funding is what decides which one wins.
Portfolio Watch: A portfolio is a portfolio
We are up approximately 12% this year, so the pace is slightly more pedestrian than it has been for the past two or three years. I am decently satisfied with that given the turbulence, including two or three major sell-offs in AI that we have to some extent navigated. We are still trending towards roughly 40% annualized since this portfolio started, and there is live money in it.
The AI names first, because the remix I described above is where the book is built. Remember the ongoing discussion on whether Micron and SanDisk are still very cyclical. Look at their guidance: the cyclicality is much, much less pronounced than it was. SanDisk now has four-year deals with its clients, which is a major shift from selling NAND more or less spot. That gives them almost four years of transparency on their flows, and still we are pricing SanDisk at 10x and Micron at 8x on run-rate earnings, roughly half the P/E of the S&P 500.
Marvell, the custom-chip designer I bought ten days ago, is priced higher, as you would expect for a less cyclical profile, and it is off to a good start. And CrowdStrike, the software on top of AI in cybersecurity, sits at almost 200x. I know software has a smoother revenue profile, which is why the market prices it like that. But there are galaxies between how the market views the cyclicality of the two, and given the visibility we have around the flow of Micron and SanDisk, especially if I am right that we move towards inference, they are way, way too cheap. That view is also why I prefer them over the Koreans.
The remix in one sentence: underweight Nvidia, own the rest of the semiconductor stack. Training is one enormous synchronized job, many GPUs working on the same model and constantly swapping results, so raw computing speed and state-of-the-art chips matter. Inference is billions of small, separate tasks — every prompt anyone sends — so it does not need the synchronized workflow.
It needs memory, and it certainly also needs storage, which is the underestimated part of the remix: when you store a project in an AI assistant, it needs to sit close to the model, and that is NAND flash. SanDisk, to put a name on it. We have already seen a slowing in the pricing of state-of-the-art memory chips and an acceleration in storage chips, NAND versus HBM, in response to this debate. On top of that come the custom-chip designers, Broadcom and Marvell, whose hyperscaler-designed chips are substantially cheaper than Nvidia’s comparables. The biggest loser is Nvidia, and to some extent AMD, maybe even more so.
One more thing on the security names. If you feel CrowdStrike is a little expensive, and I sympathize, SentinelOne is the alternative: it grows almost as fast, 22% versus 25% on ARR, and it trades at one-sixth to one-seventh of the price on EV to recurring revenue.
It is not state-of-the-art the way CrowdStrike is, but it is a cybersecurity company growing almost as fast and decently equipped for the AI world. We hold both.
Alt-health. Panorama was the third Phase 3 on DT120, and it was actually the weakest of the three. Still very good versus placebo, but a little less good than the other two, so it was fair that we got a bit of a sell-off once the market opened. The real issue is what drives the stock from now until next year. There are no revenues worth mentioning, and nothing commercialized yet. We now know they have three rock-solid Phase 3 results; the next step is the approval process: a pre-NDA meeting with the FDA this year, an application early next year, and a commercial launch that depends on the FDA and could take a year or more. RFK has promised fast-tracking of this area of drugs. Everything we wanted from the trials is now here. What is missing is the next trigger.
It could be M&A: the product is there and ready for approval, and depression and anxiety are probably the two biggest diseases you can treat apart from obesity. For a company like Johnson & Johnson, this looks like a natural fit, especially with Eli Lilly having bought a smaller psychedelics player a month or two ago. But I am not in the know, obviously. So I am caught between a rock and a hard place: I am considering selling some of the holding simply because there is no evident trigger, while the most likely trigger between now and launch is exactly the one nobody can time.
I still hold a lot of sympathy for Beam, and if I sell some Definium, I may add to it. The results this month were super good, but there were safety concerns for some of the participants, and that is a red flag for the authorities that they now have to address. The gene editing works in the study, and that is a really good start for a newcomer to the biotech space.
So the results solidified the case, and the safety signal weakened the outcome space. We have had the same discussion around the GLP-1 drugs: as soon as Novo had a couple of patients losing eyesight, it started hurting the stock versus Eli Lilly. Safety is a big thing in this space. That is why Beam has not performed. But what they are doing works, and I am more comfortable with the thesis than I was before this study.
The power theme. I have quite a large exposure to solar, maybe 8% of the portfolio, and a decent exposure to fuel cells and what goes into them. Bloom has done a couple of hundred percent since New Year, which is very good. Nextpower’s problem has been the financing side, not its own but its clients’: solar projects are funded in the debt markets with a high degree of leverage, because the revenue side is very steady.
When credit costs go up and the revenue side stays constant, the outcome is not good. It is that simple. It is not the business cycle, but rather, financing costs. Nextpower, as well as IREN, is incredibly vulnerable to credit markets, and that is exactly why I take comfort in what Bessent is doing.
Energy Fuels is a story in itself: our reshoring-of-supply-chains bet. It is a nuclear company, but it is also a rare-earths company. On the nuclear side, we have reached the point where it is a pretty solid cash-flow business: the inventory cost of the uranium on the balance sheet is roughly half of what they can sell it for, which is a decent margin for a commodity, and it will turn into a cash cow as the buildout of nuclear continues.
The other part is very investment-heavy, because they use all of the proceeds from uranium to build the rare-earth chain, almost at the orders of Bessent and his team. China is a de facto monopoly in rare earths, and Energy Fuels is the only company that can go all the way from mining, in Madagascar, for example, to refining, processing, and making the permanent magnets that go into an EV or a fighter jet. MP Materials cannot do that. It is a full supply chain inside a controlled sphere.
The whole case has been hurt by the trade truce between the US and China, and we got a prolongation of that truce this week. As long as there is a truce, the market does not know whether it needs the supply chain. But when the can stops being kicked down the road, and at some point it will, this is the hedge. When the rare-earth licensing regime was floated last year, it was the only thing in the portfolio that worked for two weeks; everything else was sold. We downsized it twice into escalations, from 15-16% of the book at the peak to a little less than 5% now. It is still a big position, but it was a lot bigger, and it still needs to be there for the day the truce breaks.
Trades and themes across asset classes
In FX: short USD as a lean. Slightly wrong-footed this month by the energy channel, but back in a QE Like regime where EURUSD, USDJPY and USDSEK all point the same way. EURUSD sits on the downtrend line from the January high; we hold the lean small and do not add until it breaks clearly above again.
In Rates: not upbeat on bonds yet, but the first, very first signs of something slightly more constructive for bonds and credit are here. Bessent has plenty of firepower, and whether he funds it with the TGA or with bills, more dollars arrive. Constructive on funding markets before we are constructive on duration.
In Commodities: gold and Bitcoin both, with Bitcoin the better horse so far. Oil past peak escalation, a phased Hormuz reopening under negotiation, short the cracks and short the barrel from this zone; short the Gulf Coast refiners into the diesel ban, long European refining margins if you want the other side.
In Equities: the remix. Memory and storage on top, Marvell as the custom-silicon and networking expression, sell Nvidia, buy everything else in semis. Security, through CrowdStrike and SentinelOne. Definium held while we look for the trigger; Beam kept and possibly added to. The power sleeve and IREN are a bet on Bessent fixing financing markets. Energy Fuels stays as the hedge for when the trade truce stops.
Have a great weekend, and good luck out there!
