What We Told Hedge Funds

Happy Friday from somewhere between Copenhagen and my in-laws. I am writing this with one eye on the traffic (talking to my AI setup) and the other on a Financial Times story that managed to knock 3% off the semiconductor index by discovering that OpenAI and Anthropic count cloud revenue differently. In a moment, I’ll get to why that story is pure bullshit. But first, here are the things that are not:

1. You don’t think interest rates are restrictive? The economy and the market disagree now

Let us start with the thing we have said for a month and the market has started to say back. Rates at 4% with the long end above 4.8% are restrictive for everything that is not AI, and the AI build-out has been masking it.

Two charts make the point better than I can. The first is the Economist’s real construction spending breakdown, which I have redrawn: data-center construction is up 65% in real terms over the past year, power is up 4%, and every other category on the chart is NEGATIVE: residential down 10%, manufacturing down 24%, office ex-data-center down 14%. Don’t tell me that rates aren’t restrictive for the broader economy.

The second is the sector scoreboard. Over the past month, exactly one S&P sector has beaten the index: technology, with communication services the only other above water. The rate-sensitive sectors, financials, real estate and utilities, are 7-8% behind. This is what multiple compression under a hiking cycle looks like in the cross-section: the market pays for the few companies that deliver earnings now and stops paying for everything that depends on the cost of money.

We flagged narrow breadth as a symptom of the regime rather than a forecast of its end, and the sector table is the same symptom seen from the other side. It also fits our nowcasts, which have the US growth probability at 19% and falling, the weakest on the board, while inflation risk sits at 44% and below neutral. The economy has told the Fed. The market has told the Fed. Let’s see how the Fed responds to it.

2. European banks tell us we are back to March/April peak pessimism

The best single gauge of European risk appetite is the bank index, and it is telling us two things at once. The Euro Stoxx Banks index has fallen 10% from its August high of 322 to 289, and the reasons are the ones we discussed last week: French spreads at 111bp, a budget that cannot pass, Japanese accounts selling OATs weekly, a gas storage number that does not survive a harsh winter, and now an ECB that has to hike into all of it. I do not dispute any of that.

But look at where the index sits in its own structure. European banks have traded in a rising channel since 2024, and the sell-off has taken the index from the top of that channel to the bottom of it. The last time it was down here was the March/April low of the Iran war, when the channel bottom held, retail sentiment was at the lows, oil longs were at 70-80% of open interest, and the war was about to pass its peak.

The setup today is the same on all four counts. The bottom of a rising channel is where you buy if the cycle holds, and everything in this letter says the cycle holds. The bank chart says we are back at peak pessimism. It does not say the pessimism is right, and it is telling you where to lean against it.

3. The massive buy-the-dip after fake news from the FT struck again

The Nasdaq lost 1.25% on Thursday, the semiconductor index 3.4% and Micron 4.8%, on a Financial Times report that OpenAI told investors its annualized revenue was approaching $50bn at the end of September rather than the roughly $70bn that was widely reported two weeks ago. The market read that as a $20bn hole in the business. It is nothing of the sort.

The gap is accounting methodology. Anthropic counts revenue generated through cloud partners, AWS and Google Cloud, in its headline number; OpenAI apparently does not. Investors had adjusted OpenAI’s figures to make them comparable with Anthropic’s, which is how the $70bn estimate was produced. So we are talking about whether revenue flowing through cloud partners is included in the headline or not. That is a comparability issue, not a collapse in the revenue trajectory, and OpenAI itself confirmed the $50bn run-rate on its own basis the same afternoon.

The fact that the market treated this as a major negative for AI is, frankly, pretty ridiculous. A company that was at $5bn in January last year is at $50bn on the conservative basis.

Two things make it a buy-the-dip rather than a warning.

First, the money rotated; it did not leave: the equal-weight index was positive on the day, small caps found buyers, and breadth improved while AI sold off… Second, the earnings are not in question. Micron printed $54bn and guided $61.5bn a week ago.

Korean chip exports are up 83.5% year-on-year. The 2027 HBM is sold. A story about how OpenAI counts its cloud revenue does not change a single one of those numbers. We did not sell a share. We would have bought more if the lean was not already full.

In focus: Wolfspeed is now a bet from the Admin… should we celebrate?

Wolfspeed announced after the close on Wednesday a conditional commitment from the US Department of War, through its Office of Strategic Capital, for up to $1.5bn of 30-year senior secured financing to fund domestic silicon carbide and wide-bandgap power device production, with a defense angle: radiation-hardened SiC and GaN for defense and communications systems. The government gets warrants for up to 7.5% of the fully diluted equity as tranches are drawn. The stock gapped 14-16% pre-market from $31.37 (but fell back again), a market cap of about $1.6bn, so the loan is nearly the size of the company. It is still conditional on diligence, definitive agreements, and Congressional funding, and the company says so plainly.

We own Wolfspeed in the Electricity sleeve at 3.3% of the book, bought at $34.96 in October 2025, conviction 10, and we were down 11% before the news. The thesis never depended on the quarter: silicon carbide is the material that converts power efficiently at high voltage, which every data center, grid upgrade, and inverter needs, and Wolfspeed is the only scaled US producer. The bear case was the balance sheet. Start there, because it is the whole question. After the 2025 restructuring, the company has $1.2bn of cash against $1.6bn of debt, most of it convertible, and it burns roughly $90m of free cash a quarter while gross margins are still negative. That is about thirteen quarters of runway on cash alone, enough to be a going concern, not enough to fund a capacity build and win multi-year supply contracts from customers who need to know you will be there in 2030.

The loan changes that arithmetic: up to $1.5bn of 30-year money takes the runway to something like thirty quarters, and more importantly it tells every hyperscaler and defense prime negotiating a long-term SiC agreement that the US government is underwriting the supplier.

It also fits the 2027 call.

Our thesis since 2024 has been that the AI bottleneck moves down the chain: compute first (the Nvidia years), memory now (the Micron and SanDisk years), and power next, which is 2027. Each shortage is solved by capex that creates the next one. Silicon carbide is the semiconductor side of the power leg, the chip that sits between the grid and the rack.

Wolfspeed has built a data-center sales team, launched 300mm SiC wafers and a 10kV MOSFET aimed at hyperscale infrastructure, and its AI data-center revenue is growing about 30% per quarter from a small base. It now has the funding to finish the capacity for that demand.

Now the honest question: is it good or bad when the Administration gets involved in a stock you own? The data says both, and it says which is which. The government has taken equity stakes or warrants in more than thirty companies since mid-2025, seventeen of them listed. The pattern is consistent: a large pop on the day, Trilogy Metals plus 211%, Lithium Americas plus 90%, MP Materials plus 51%, and then, for most, a fade. By September, fourteen of the seventeen listed stakes traded BELOW their level the day after the announcement. The Cato Institute called it a sugar high, and for the typical name, that is fair. The exceptions are the ones where the state money changed the business rather than just the narrative: Intel, where an $8.9bn stake is now worth more than $50bn, and the stock has quadrupled, and MP Materials, up around 100% since the Pentagon deal, because the government also became the customer. Lithium Americas, where the stake came with a loan restructuring and no demand, is down 30%.

So which is Wolfspeed?

The test is simple: does the state money change the business or just the headline? Here it changes the business, because the balance sheet WAS the business problem, and because the Department of War is, via the radiation-hardened SiC and GaN programme, becoming a customer as well as a lender, which is the MP Materials structure rather than the Lithium Americas one.

The warrants are a real cost, 7.5% dilution at full draw, but the government taking equity in exchange for patient capital is exactly the arrangement we have been looking for across the reshoring sleeve, in rare earths and uranium as much as in power semis.

The same administration that slow-walked the CHIPS grant has decided that a domestic power-semiconductor supplier is strategic infrastructure. The state picked a winner, and it picked ours. We hold, with our conviction unchanged at 10. I would add on a pullback toward the mid-20s while the agreements are negotiated. The caveat: the loan funds defence and GaN as much as the commercial ramp, so the data-centre pick-up still has to show up on its own. The Admin bought Wolfspeed time. The cycle has to do the rest.

Portfolio Watch: Not getting a lot of traction on the short USD lean or the long semis lean

The book is up 7.9% year-to-date, down from 11% two weeks ago, and the honest summary of the week is that the two leans we have been loudest about are not working yet. Let me take them one at a time, with the triggers and what could go wrong.

The short USD lean. The dollar is flat against the euro since July and down against everything else, so the trade has worked in gold, silver, and Bitcoin and not in the cross everyone watches. The fundamental case has not changed: the US is the softest economy on the board on both inflation and growth. Williams says one more hike and no rush; the ECB and the BoE have the firmer path, and the transatlantic inflation spread that has led EURUSD since 2008 is the widest of the year. We have a below-consensus call on next week’s US CPI, out Tuesday; the full preview is out on Monday, but the short version is that the street has core at 0.2% and 2.5% year-on-year, rising, and our live readings say core is the ceiling, not the base.

A soft core print is trigger one. Trigger two is a Hormuz products deal, which collapses the energy premium that has carried the dollar for a month. Trigger three is Bessent increasing the buyback pace again, which he says he will.

What could go wrong: the energy channel stays open longer than we think, French risk keeps the euro pinned, and the Fed delivers two hikes rather than one because the headline stays above 3%. If the CPI is hot on core, the lean gets cut. Positioning helps: the specs have started to trim their dollar longs, so the crowd is less on our side of the pain trade than it was.

The long semis lean. Micron delivered a quarter and a guide nobody on the sell side had, Korean exports are up 83.5%, the value gap between the exports and the equities is more than 100%, and the stock fell 5% on Thursday because of how OpenAI counts cloud revenue. Strong results, no fireworks. So what triggers the gap closing?

Five candidates.

First, a products deal in Iran, which is the same trigger as the dollar: it rhymes with April, when the first wave of negotiation headlines lifted semis 50% in six weeks off a positioning setup that looks exactly like today’s.

Second, the hyperscalers confirming the duration of the capex cycle in their October earnings; the market is pricing memory as if the cycle ends next year, and a 2027 capex number from Microsoft or Google that is up again is the direct refutation.

Third, the Fed stepping a little back, which a soft core CPI delivers; multiple compression is the reason earnings growth has not been rewarded, and the first hint that the hiking cycle is short reverses it.

Fourth, the Korean license decision on December 31, where non-renewal removes supply and hands share to US memory.

Fifth, and least discussed, the contract structure becoming visible in the numbers: when Micron and SanDisk report revenue that moves with their take-or-pay agreements rather than with spot pricing, the cyclicality discount has to go.

What could go wrong: a genuine capex pause from one hyperscaler, which would be the first in this cycle; a license renewal that keeps the Koreans whole; and the curve inverting faster than the cycle, which compresses multiples further before earnings catch up.

We hold the semiconductor book in full. We see 100% upside in Q4 as the base case, and a week in which the equities fell on an accounting story while the exports printed a record is the kind of week that makes the gap wider, not the kind that closes it.

Elsewhere in the book, briefly. Wolfspeed is covered above. Definium and Beam are where they were: a trim under consideration in the former, held in the latter. The two Australian counter-drone names we upgraded last week gave some of it back. Energy Fuels and IREN remain the drags, and both are bets on the same thing as the dollar lean: financing markets improving as the regime holds.

Everything in the portfolio that is not working this week is not working for one reason: the energy premium and the hiking cycle it feeds, and everything that would fix it is the same trigger. That is a concentrated bet, and we know it.

横向滑动查看完整图表

Trades and themes across asset classes

In FX: short USD as a lean, not working yet, kept small. Triggers: a soft core CPI on Tuesday, a Hormuz products deal, a faster buyback. Cut it on a hot core.

In Rates: rates are restrictive, and the market now agrees. One Fed hike, not two. The curve flattens toward zero on a cost shock that has already peaked. Constructive on funding markets before duration.

In Commodities: short crude, long cracks, gold and silver over crude. Oil longs at 44% into a war past peak escalation.

In Equities: semis up 100% in Q4 as the base case, five triggers listed above. Wolfspeed held at conviction 10 with the Admin as a lender and a customer. Buy the AI dip when the news is an accounting line.

Have a great weekend, and good luck out there!

Andreas Steno

Andreas is the core macro consultant in Steno Research. Building on years of experience as Global Chief Strategist at Nordea Bank, he is one of the most quoted and sought-after macro analysts out there. Andreas’ expertise is the FX/rates, energy, real estate and equity spaces, but doesn’t shy away from hot takes on other topics if the underlying analysis is strong enough. Andreas anchors the weekly editorial ‘Steno Signals’.