What We Told Hedge Funds This Week — A weak USD is the final piece of the puzzle

A weak USD is the final piece of the puzzle

The stars are aligning with our nowcasting: soft(er) inflation, cyclicals rebounding, the Fed slowly repriced towards on hold, but the dollar refuses to budge against >2 standard deviations of spec longs. The missing link runs through the Hormuz Strait, and NOT via crude... via the products. Plus: why we would rather buy OpenAI than Anthropic (yes, really), and a great week for the book.

It seems like most stars are starting to align with what our nowcasting and pattern recognition models suggest. A softer inflation impulse, albeit more clearly seen in the PPI this month (relative to the CPI, which came in more in line), a rebound in cyclically exposed equities (such as AI hardware, but also elsewhere), a slow but sure repricing of the Fed to closer to an "on hold" stance... but the missing piece in the puzzle is the USD.

We are still seeing a steady build-up of USD longs by specs, and it is getting >2 standard deviations stretched by now... so why isn't the USD budging (more) when the Fed is getting repriced? I still think the nail in the coffin on the USD story relates to the Hormuz Strait. Maybe not directly from oil, but from the PRODUCTS... and that is still where the real problem sits. So let's assess that situation.

China is back at the pumps... slowly

The stories are starting to leak here and there about an increase in Chinese imports of oil (they obviously cannot feed off oil reserves forever), and even if they have a buffer well into 2027, it makes sense to increase imports now that the market is more in balance (despite the continued hiccups in the Strait).

There are, as I have said over and over, no real signs of massive stress in the physical market, so there is still enough oil leaking from the region to keep everyone from panic buying for now.

But as I have reiterated, I think the oil market will chop around here, as China will become a soft FLOOR under the market as they increase imports... and as highlighted over the past weeks, we can now see it in the most timely indicator from China, namely the refiners' run rates. When China imported more than 5mn barrels less a day, the refiners were running low naturally, but we have seen a decent tick up in the run rates, which HAS to be a sign that China has started importing more.

The real story sits in PRODUCTS, not crude

We have been highlighting this for weeks and weeks now, but the gasoil (diesel) cracks in particular are still trading close to the highs for the year, and the cracks in gasoline remain an absolute blow-out relative to the rest of the energy complex. The main reason is that the EU ban on Russian products is increasingly biting, particularly in Northwest Europe (which used to be heavily dependent on Russian product imports), and there is no easy replacement for those barrels as the Atlantic basin is structurally short on product. The result is that refiners are still having a blast, and as the chart on page 5 highlights, refining margins are still close to the post-COVID peaks.

Of course, this is bullish for the European refiners and the broader energy complex, but it is also a HUGE problem for the European consumer and the European economy. Diesel prices are a key input into the cost of transportation and the cost of manufacturing, and they are now significantly higher than they were a year ago. If the EU keeps the ban in place for the foreseeable future, this is going to be a drag on European growth for the foreseeable future, especially for the transport and manufacturing sectors, but also for the consumer who is now paying significantly more to fill up their car.

And that is, of course, before we even start to talk about the gasoline cracks, which are also trading at multi-year highs as the US refiners are still struggling to keep up with demand. The bottom line is that the real story sits in PRODUCTS, not crude, and the products are still very tight.

That is the energy story for the European economy in a nutshell. Tight products, tight refining margins, and a consumer that is paying the price for it. And as I have highlighted over the past weeks, this is also a key driver of the European inflation impulse, which is why the EU inflation print for July came in higher than expected (especially for services).

And as the PPI chart on the next page highlights, the European PPI for July came in at +0.3% MoM and +1.6% YoY (vs. +1.3% YoY expected), which is the highest reading since March 2024. And the breakdown is not pretty, with services (which are typically the stickiest part of inflation) coming in at +3.92% YoY (vs. +3.5% YoY expected) and intermediate goods at +5.19% YoY (vs. +4.4% YoY expected).

The energy component was a key driver of the upside surprise, with energy PPI coming in at +5.7% YoY (vs. +1.6% YoY expected), which is the highest reading since November 2022. So the energy story is becoming increasingly important for the European inflation outlook, and it is a key driver of why we think the European inflation impulse is still too high for the ECB to be comfortable cutting rates aggressively anytime soon.

On the CPI/PCE debate: a fairly simple question, actually

I have been writing about this for a while now, but the US CPI/PCE debate is one of the most important macro debates right now. The Fed is clearly more comfortable with the PCE deflator (which tends to run 30-50bps below the CPI), and the FOMC has consistently pointed to the PCE as its preferred measure of inflation.

But the question is: should the Fed be looking at the CPI or the PCE? And as the chart on the next page highlights, the spread between CPI and PCE has been widening for some time now, and is currently at the widest level since 2022. This is largely driven by the fact that shelter inflation (which has a much higher weight in CPI than in PCE) has been stickier than expected, while the PCE has been dragged down by the much faster decline in goods prices.

This matters because if the Fed is going to start cutting rates in September, they will likely be looking at the PCE print for July (which comes out on August 30th) as the key data point. And if the PCE comes in significantly below the CPI (which is what the market is pricing), then the Fed will have more cover to start cutting rates.

But what if the PCE doesn't come in as low as the market is pricing? What if the CPI is actually a better leading indicator of where inflation is going? In that case, the Fed would be cutting rates into an inflation print that is still running above their 2% target, and the market would be forced to reprice the rate path higher.

And as the chart on the next page highlights, the Fed has actually turned hawkish recently, with the 2-year yield now back at 4.20% (vs. 3.85% at the start of July). This is a clear sign that the market is starting to price in a higher terminal rate, and it is also a sign that the market is starting to question whether the Fed will actually cut rates as much as it has been guiding.

Portfolio "in focus": Will OpenAI beat Anthropic on costs?

Michael Burry's depreciation theory...

The key takeaway here is that OpenAI is going to be able to offer their inference services at a much lower cost than Anthropic, and that is going to allow them to undercut Anthropic on price for the same quality of service. And as Burry's chart on this page highlights, the B200 is significantly more cost-effective than the H100 for inference workloads, particularly as the model sizes increase.

Of course, there are many other factors that go into the pricing of inference services (including the cost of the GPU, the cost of power, the cost of cooling, the cost of networking, etc.), but the chart on this page gives a good sense of the magnitude of the cost advantage that OpenAI will have. And as I have highlighted over the past weeks, this is going to be a key driver of the next phase of the AI trade, particularly as the inference market starts to mature and the competition for inference workloads heats up.

And as the chart on this page highlights, OpenAI has been aggressively securing compute capacity, with the company now having over 20,000 MW of secured compute capacity (vs. less than 5,000 MW just 18 months ago). That is a massive step up, and it is a clear sign that the company is preparing for a future where inference is the dominant workload.

And as the gap widens into 2028, it is clear that OpenAI is going to have a significant advantage over its competitors in terms of compute capacity. That is going to allow them to offer more competitive pricing on inference services, which is going to drive more volume, which is going to drive more revenue, which is going to drive more investment in compute, in a virtuous cycle that is going to be very difficult for competitors to break.

And as the chart on this page highlights, the power capacity requirements for these large-scale AI training runs are enormous. A single 100,000 GPU cluster (which is what the largest AI labs are now operating) requires around 150 MW of power, which is equivalent to the power consumption of a small city. And as the AI labs continue to scale up their training runs, the power requirements are only going to increase.

This is one of the key reasons why we have been highlighting the power and utility trade for the past 18 months. The AI labs are going to be massive consumers of power, and the utilities that can provide that power (and the infrastructure to deliver it) are going to be massive winners. We are particularly constructive on the natural gas utilities (given the speed-to-market advantage) and the renewables (given the ESG tailwinds), but we are also constructive on the nuclear names (given the baseload power profile).

And as the chart on this page highlights, the premium for immediate bridge capacity (i.e., MW that can be deployed within 6 months) is currently around $20/MW/month, which is a massive premium over the long-term contract pricing of around $5/MW/month. This is a clear sign that the market is desperate for capacity NOW, and that the AI labs are willing to pay a significant premium to secure that capacity quickly.

This is one of the key reasons why we have been highlighting the small-cap power and infrastructure names that are able to deliver capacity quickly. The larger players (like the hyperscalers) are increasingly focused on long-term, hyperscale deals, which is leaving a vacuum in the small-cap space for the AI labs that need capacity NOW.

And as the chart on the next page highlights, the small-cap power and infrastructure names that we are focused on include the colocation players (WULF, CIFR, HUT, CORZ), the GPU cloud players (IREN, Oracle, AWS), and the nuclear and power players (Vistra, Constellation Energy, Talen).

And as the chart on this page highlights, the AI stack is becoming increasingly complex, with the infrastructure layer (GPUs, networking, power) becoming a more important bottleneck than the model layer itself. This is one of the key reasons why we have been highlighting the infrastructure names (particularly the small-cap power and colocation names) as the most attractive way to play the AI trade.

Of course, there are many ways to play the AI trade (including the model layer (NVDA, AMD, AVGO), the cloud layer (MSFT, GOOGL, AMZN), the data layer (SNOW, DDOG, MDB), and the application layer (anything that uses AI to generate revenue). But as the infrastructure layer becomes a more important bottleneck, we think the small-cap power and colocation names are the most attractive way to play the trade, given their underappreciated exposure to the AI capex cycle.

The portfolio implications are clear: we are long the small-cap power and colocation names (WULF, CIFR, HUT, CORZ), the GPU cloud names (IREN, Oracle, AWS), the nuclear and power utility names (Vistra, Constellation Energy, Talen), and the natural gas utility names (Cheniere, Sempra, Williams). We are also long the renewables names (NextEra, Brookfield Renewable, Clearway) and the broad AI infrastructure complex (NVDA, AMD, AVGO, MRVL, SMCI, DELL).

We have been adding to these positions on any weakness, and we still think that they are the most attractive way to play the AI trade for the foreseeable future. The combination of underappreciated exposure to the AI capex cycle, attractive valuations, and reasonable balance sheets is hard to find elsewhere in the market, and we think these names are well-positioned to continue to deliver strong returns over the next 12-18 months.

And as we have highlighted over the past weeks, the catalyst for these names is going to be the continued ramp in AI capex (which we expect to remain strong for the foreseeable future), the continued ramp in power demand (which is going to drive the utility trade), and the continued ramp in colocation demand (which is going to drive the small-cap power and colocation names). All of these trends are intact, and we think they are going to continue to drive strong returns for these names over the next 12-18 months.

And as the chart on this page highlights, our trigger pipelines are LOADED. We have a double-feature in September with BEAM-302 (sickle cell) and PANORAMA (thalassemia) topline readouts, and we have updated data on the BEACON trial (another sickle cell program). We also have a potential FDA decision on a new product in the same timeframe.

These are all binary catalysts that could move the stocks meaningfully, and we are positioned for all of them. The setups are not particularly expensive (the options market is pricing in a relatively low probability of success), and the upside scenarios could be multiples of the current price. We are not trying to call the exact date of the readout (that is impossible), but we are positioning for the EVENT itself, with a defined risk and a clear upside scenario.

And as I have highlighted over the past weeks, this is the kind of catalyst-driven approach that we think makes sense for the small-cap biotech space right now. The setups are not expensive (the options market is pricing in a relatively low probability of success), the upside scenarios could be multiples of the current price, and the downside is well-defined (the readout fails, the stock goes back to where it was trading before the catalyst, and we move on to the next idea).

And as the chart on this page highlights, our current portfolio is concentrated in the highest-conviction names in the AI infrastructure complex, with a smaller allocation to the small-cap biotech sleeve. We are overweight the AI trade (relative to the S&P 500) by roughly 3:1, and we are overweight the small-cap biotech trade (relative to the XBI) by roughly 2:1.

The AI trade is dominated by the small-cap power and colocation names (WULF, CIFR, HUT, CORZ), the GPU cloud names (IREN, Oracle, AWS), and the nuclear and power utility names (Vistra, Constellation Energy, Talen). The small-cap biotech trade is dominated by the BEAM-302 (sickle cell), PANORAMA (thalassemia), and BEACON (sickle cell) catalysts, which we think are the most attractive risk/reward in the space right now.