The AI Supercycle Podcast from QF Markets provides independent capital intelligence for the AI industrial economy. From semiconductors, AI factories and data centres to power grids, energy markets, critical materials, orbital compute and embodied intelligence, we follow where capital is being deployed, where the binding constraints are emerging, and what it costs to build through them.
Each weekly episode connects the latest developments to the multi-year forces behind the AI build-out, and sets out a clear base case along with the evidence that would change it. For traders, investors and anyone trying to understand where the capital in this cycle is really going. Published weekly by QF Markets.
The AI Build-Out’s Power Bottleneck
[0:00] On Monday evening, a public notice went up in Starbase, Texas. It was dry, the sort of thing that nobody reads. A request for property tax abatement, to be considered at a city commission meeting on the 9th of December. What it describes is anything but dry. SpaceX, the rocket company that now owns XAI, is proposing a gas-fired power plant and a supercomputing facility, with an estimated $24 billion of investment in the first phases. It would build, own and run it. And by making its own electricity, it could avoid a year's long wait to plug into the Texas grid. A week earlier, voters in 40 out of 41 local contests across America had favoured limits on data centres. SpaceX didn't draw up a $24 billion plan in a week. But put those two stories side by side and you see the same constraint from both ends. Power is the bottleneck and the most ambitious players have decided that if the grid can't give them it quickly they'll make their own. That is the story of this week. Not whether AI demand is real but who owns the megawatts who has them under contract and who still has to build them with borrowed money.
[1:26] Hello and welcome to the AI Supercycle podcast from QF Markets, where we follow the capital flows, physical constraints and commercial returns shaping the largest industrial build-out of our generation. From semiconductors and data centres, through power, energy and critical materials, all the way into orbit, and onto the enterprise workflows that ultimately have to pay for it. I'm Tim Hardwick, founder of QF Markets, and I'm recording this on Thursday, the 12th of November, October 2026. This one is a little shorter and a little different. Prism is sitting this one out and I'm putting my own trading front and centre, because last week was the kind of week that tests a process rather than a thesis.
[2:10] We'll start with the macro regime and what the market was really telling us underneath the headlines. Then I'll walk you through how I've been trading it. After that, a closer look at power and why gas has become one of the most interesting threads of the AI build-out. Then natural resources and the global macro picture. And we'll finish with what matters next, including a fortnight that runs from Nvidia's results to the APEX summit. We'll also cover the latest news that hits the headlines before we publish this episode.
Market Turbulence and the Fed’s Next Move
[2:41] So let's start with a closer look at the macro regime. You'll all know last week was pretty rough. The S&P 500 had its worst week since the tariff shock of April 2025. And this week has brought a recovery. I'm not going to replay the price action, you've probably seen it on your own charts, but what interests me is one day in the middle of it. On Wednesday, before the selling became indiscriminate, the market did something that it rarely does. Power developers and the cloud builders funded with debt were sold hard, while Amazon and Microsoft rallied sharply. My reading is that for one session, Traders and investors stopped treating AI as a single bet and started to ask much better questions. Who already has their capacity? And who still has to build it?
[3:30] Thursday then sold everything, as markets do when fear takes over. But I think Wednesday was the signal and Thursday was the noise. Behind all of this sits the price of money. The Fed held rates at 3.75-4% in October, after a hike in September, with two members voting to hike again. Then October payrolls came in at 165,000, with earlier months revised upwards. On Tuesday, we got October's inflation numbers, and they ran a little hotter than expected. Consumer prices rose 0.4% on the month, and core prices rose 0.3%, each a tenth above forecast. And that's the third month running at that core pace. One quirk worth knowing. There's no official annual inflation rate for October this year, because the comparison data from October 2025 was never collected during the government shutdown. So for once, the monthly number is the story. And the market hasn't needed much persuading. Fed funds futures have expected a December hike since late September and after Tuesday, CME's FedWatch tool shows that it's fully priced. Most of the markets expect a quarter point, but a small minority is now betting on a half point. An idea that briefly gained a lot of support in early October before fading away.
[4:56] Now, as we know, Kevin Walsh doesn't do forward guidance. He's been clear about that since he took the chair, and he hasn't changed his mind. But that hasn't stopped his colleagues from thinking out loud. And this week, several of them have been unusually vocal. Governor Christopher Waller said before the inflation release that a hot core number should put half a point hike on the table in December. And that only hiking a quarter could put the Fed's credibility at risk.
[5:25] Dallas Fed President Laurie Logan said it may be possible to move faster, and that she thinks that rates are still too low. And Chicago Fed President Austin Goolsbee, who is usually one of the more patient voices, described the latest data as borderline disturbing. When the chair stays quiet, the rest of the committee becomes the forward guidance. So the question for December isn't really whether the Fed hikes, it's how big the step is. Now oil has fallen a long way since October, which should help headline inflation in the coming months. But it's core inflation that has the committee worried, and lower oil doesn't fix that on its own. If you think of the AI build-out as a construction site, interest rates are the weather. Owners of the finished buildings have more shelter from the rain, although they could still feel it. The developer with the half-built tower, the scaffolding and the bank loan feels every drop. And they're going to get very wet.
Trading Through Uncertainty
[6:25] OK, so let's now turn to my own trading and how I've been navigating a week like this. I judge my trades on process, not outcome. A good trade is one that followed the system. Sound probabilistic reasoning, clear risk management and protecting capital. A trade can lose money and still be a good trade. A trade can make money and be a terrible one. That's just getting lucky. Last week was a very good week to remember that. Not that I had terrible trades last week, because I was flat going into Thursday. When the Nasdaq broke through the support that had held since August, the first thing that I did was nothing at all. And I mean that deliberately. A broken flaw tells you that the market has changed its mind about something. It doesn't tell you what. My job at the moment is not to have an opinion about the next move. It's to check whether my risk is where I planned it to be, and whether the reason I held each position is still true. The second thing was to take Wednesday seriously. When the hyperscalers are rallying on the same day that power developers are being sold hard, I read that as the market sorting the build out by layer. So I went back through my watch list and sorted it the same way. Companies with operating capacity on one side, companies with contracted capacity still to deliver in the middle, and companies that still need permits, turbines and loans on the other.
[7:51] Now this is exactly where I never talk about the AI trade as one thing. The layers diverge. And last Wednesday, they diverged in a single session. The third thing was the calendar. In Formula One, when the safety car comes out, the field bunches up. Nobody can overtake. And everyone knows the race will restart. They just don't know exactly when. This week and next feel a bit like that to me. Inflation, CoreWeave, Nebius, and then also we have NVIDIA's results on the 17th of November. When binary events are stacked up like that, I'd rather run smaller sizes and be patient about where the restart comes, rather than pretend the race is still on. Hence why I'm currently flat, and waiting for the system to give me a clear signal. There's a psychological point here too, and it's one that I think about a lot in my performance work. After a week like last week, there is a tendency to want to do something, anything, to feel back in control. Revenge trading after a loss or panic selling at the low, both come from the same place. The need to end an uncomfortable feeling. The traders who last are the ones who can sit with that discomfort long enough for the market to show its hand.
[9:11] A pilot in turbulence doesn't wrestle the controls. They hold the attitude, trust the instruments, and let the aircraft ride it out. They may also slow down. And that's what I've done this week, is to slow things down, look at what the market's doing, wait for the signals from my system, and be patient.
Who Owns the Megawatts?
[9:31] So now let's take a closer look at the physical build-out, and the question in this week's title, who owns the megawatts? Let's start in a small town rather than in Washington. In Perry, Ohio, about 87% of voters back to measure requiring hyperscale data centre developers who want certain tax incentives to give financial guarantees to the community. Three towns in Maine approved six-month pauses, and five communities voted for outright bans. Not every one of those votes is binding, but the direction is clear. UBS estimates that if pauses like these spread, they could affect around 30% of the capacity America expects to add in 2028. Now that's an analyst's estimate, not capacity already lost, but it tells you where the risk sits. I think of this like the green belt around an English city. You don't ban property. you make new buildings slow, conditional and expensive. And when you do that, the people who already own the houses are in a stronger position than the developer waiting on planning permission with a loan ticking away and costing them money.
[10:41] The physical queue is just as long as the political one. Reuters, citing Wood Mackenzie, reports that a new gas turbine in the US now takes around six years to deliver on average. Global orders are well ahead of what the manufacturers can build, and the developers are buying second-hand turbines and converted jet engines, with refurbished units selling at a big discount to new ones, but still at a hefty price.
[11:06] If you bought a car in 2021, you'll remember this. A chip shortage held up the production lines, and suddenly a year-old second-hand car cost almost as much as a brand new one. The AI chip boom is doing the same thing to power equipment. Now this is where I think the simple version of the story needs refining. It's tempting to split the world into owners and builders, but there are really three groups. There's capacity that's already operating, there's capacity that's under contract but still has to be delivered, And there's capacity that's still a proposal, waiting on funding, permits and equipment. Each carries a different kind of risk. Constellation Energy is a good example of the middle group. It raised its 2026 earnings guidance on Monday and pointed to two 20-year deals signed before the vote. One with Amazon and one with Google. Those contracts give real visibility, but part of the Amazon deal is an expansion expected early in the next decade, and the first upgrade behind the Google deal isn't due until 2028.
[12:14] A contract puts you on the right side of that green belt that we talked about, but you still have to build the house. Now the most ambitious players are trying to move themselves up that list by owning their power outright.
[12:27] The Financial Times reports that Anthropic has committed about $3.4 billion, for 1.7 gigawatts of gas-fired generators. And now SpaceX with its proposed plant at Starbase. A SpaceX affiliate has already filed plans for an 8-mile gas pipeline into Starbase, due in service by early 2027, originally to make fuel for Starship. The company that liquefies methane for rockets may end up burning gas for compute.
[12:56] SpaceX also reported a nine-fold rise in its AI revenue last quarter, much of it from renting computing capacity to Google, Anthropic and others.
[13:06] So it's already an owner of compute, and it wants to be an owner of power too. But buying generators or proposing a plant doesn't make a capacity operational. Until it's built and running, it carries construction risk like anyone else. And it isn't only in America.
[13:24] On Monday, Norway moved to rein in data centres. Its grid operator, Statnet, had paused new connection reservations, with the pause expected to run until August 2027, and the government plans to set out licensing principles next autumn. The striking number is the demand. Requests for new connections now add up to more than Norway's entire peak electricity demand, and data centres make up most of them. The government wants to favour critical uses such as defence, health and the financial system. And it wants the industry to contribute to new generation, including gas with carbon capture and offshore wind. When a country that exports power starts rationing it, you know the bottleneck is real. Here's the twist though. The biggest owners are also buyers. Microsoft said that some third-party capacity it expected arrived late. And Azure still grew 48%. Alphabet has started renting computing power from other providers and it expects that to squeeze its margins. Even the giants own some of what they need and rent the rest. So scarcity doesn't reward everyone who owns a data centre. It rewards the right capacity, in the right place, on the right contract, at the right time.
Gas, Oil, and Global Supply Risks
[14:41] Okay, so now let's turn to natural resources and the global macro landscape. Natural gas is the thread running through this whole episode. It is one of the main routes that developers are pursuing for data centre power, which is why SpaceX, Anthropic and even the Norwegian government are reaching for it. But the turbine queue means gas power isn't quick. It's only quicker than some of the alternatives at some sites. For me, that makes the equipment makers and the contracted supply at least as interesting as the commodity itself. Oil is telling a very different story. Last week, Brent had its steepest weekly fall since the pandemic, and then fell further on Monday after President Trump said over the weekend that further strikes on Iran were highly unlikely while negotiations continued. That's quietly helpful. Lower energy prices, ease headline inflation, and the cost of running the physical layer, although they don't settle the argument over underlying inflation on their own.
[15:42] On materials, China has also extended its pause on the rare earth, industrial demand and lithium battery technology controls that it announced in October last year. The pause now runs to the 10th of January 2027, the same date America's suspension of its own affiliates rule ends. And President Trump is expected to meet President Xi at APEC in Shenzhen on the 18th and 19th of November. So the deadline hasn't gone away, it's just been pushed back to January. Which to me seems like the can has been kicked down the road.
What to Watch: Earnings and NVIDIA
[16:16] And finally, what matters next? Core we've reported on Tuesday evening, and it's the clearest picture that we have of what the price of money does to the build-out. Revenue came in at about $3.7 billion for the quarter, but the net interest expense was $846 million, against $310 million a year earlier.
[16:37] Now that's the cost of building with borrowed money. The other number I'd pick out is the power. More than 1.7 gigawatts active against more than 4.6 gigawatts contracted. Coreweave isn't just a builder, it already runs a lot of capacity. But the gap between those two numbers is the part that still has to be delivered and financed at today's rates. Management also expects around $9 billion of customer prepayments and deposits over the next three months, which shows that customers are willing to help fund that gap. Today, on the 12th of November, Nebius reports before the US Open, as a second read on whether the NeoCloud model can still fund itself. Friday brings producer prices. Then NVIDIA. Its board has set up a committee of independent directors with outside counsel to review reports concerning its chief executive. And three house committee chairs have asked for a closed-door briefing on the 18th of November. The board has said it has no indication that company secrets or national security were compromised.
[17:43] I think we're going to see a lot more about NVIDIA in the news over the next few days, as there is much more scrutiny on the company. What I have noticed in the meantime was that other chip makers and equipment makers held up well whilst NVIDIA fell. The market seems to be treating this as a question about one company, rather than a fault at the chip layer. And as we've seen in the media, they're calling it key person risk. NVIDIA reports on the 17th of November, the day before that briefing, and we'll monitor that closely. The other big stories this week are giving traders and investors some reasons for optimism. Anthropic set an indicative price range for its initial public offering that would value it up to $1.4 trillion.
[18:26] If it completes at the top of the range, alongside a $26 billion private placement backed by Google, NVIDIA, AMD and Microsoft. And TSMC's board approved up to a billion dollars of investment across Anthropic and OpenAI. TSMC expects revenue growth of about 44% this year and says demand still outstrips capacity, even while sounding more cautious about 2027. When the world's most important chipmaker wants to own a slice of its customers, that tells you where it thinks the value will sit. The UBS chief executive said this week that until recently, the market has been pricing perfection, and that expectations for how quickly AI pays off were somewhat exaggerated. And I'd agree with that. But notice that's a statement about timing, not about whether the build-out happens. Turbines on long waits, towns voting on data halls, a rocket company proposing a gas plant, and a model developer buying generators all point the same way. The build-out is going ahead, maybe more slowly and at a higher cost, but it is going ahead. And in that world, the questions that matter are simple. Is the capacity running? Is it under contract? And who is paying for the part that still has to be built?
Breaking: NVIDIA’s CEO Steps Down
[19:47] And just before we wrap this up, some breaking news that just came in as I was finishing recording this episode. Jensen Huang has stepped down as chief executive of NVIDIA and left its board, following days of pressure over reports about his personal conduct. Chief Financial Officer Colette Kress becomes the interim CEO, while the company searches for a permanent successor.
[20:10] I think this is one of the biggest stories of the year for the AI industry, and it deserves more than a minute at the end of an episode. So I'll be back in the next few days with a short, special episode on what it means for the AI build-out, layer by layer.
[20:25] Okay, so that's a wrap for this week. If you found this episode useful, and I really hope you did, please follow the podcast and share it with someone trying to make sense of the economics behind the AI build-out. Just one last disclaimer from me, everything discussed here is for information and education only. It isn't investment advice or a recommendation to buy or sell any financial instrument. And with that, thank you for listening. See you next week.