Subscribe
Share
Share
Embed
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.
[0:00] The chipmakers are reporting strong demand, some AI infrastructure borrowers are facing nervous lenders, and after last week's sell-off, technology shares bounced on Monday.
Buy the Dip or Relief Rally?
[0:11] Is this a buy-the-dip opportunity, or a temporary relief before prices fall again? This week's earnings could help us judge, but this week brings another complication. Brent crude has reached a five-month high, putting energy costs back in the spotlight. Behind the spending announcements and order books sits a harder question. How much cash will these investments generate? And can the companies funding them afford to wait? I can't tell you whether Monday marked the bottom, but we can examine whether the reasons for the sell-off are easing, or whether prices have simply recovered ahead of the evidence.
[0:59] 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 Tuesday the 20th of October 2026.
[1:31] This market picture includes last week's close, Monday's trading and news available this morning. This week's main earnings releases are still ahead of us as I record. Some may be out by the time you listen, so I'll focus on the questions they need to answer. Last week we looked into the economics of a large AI data centre. The chips, the building and the power infrastructure have to be paid for, while the revenues arrive over time. Now that timing gap explains why healthy demand and a nervous market can exist together.
[2:04] In this episode, we'll work through the macro regime, the AI supercycle, natural resources and geopolitics, before bringing it all together in my read from the framework and what to watch next.
Rates, Credit Spreads, and AI Financing
[2:16] So let's start with the macro regime and the cost of money. September's US inflation rate rose to 3.7%. Prices across services remained a concern too. That is consistent with some energy costs working their way through the economy, although we can't attribute every rise in plane tickets or hotel prices to oil. As of Monday, markets still expected the Fed to hold rates at its October meeting but they strongly favoured another increase in December. On Thursday, the 15th of October, the 10-year Treasury yield eased to around 5.1%. It was still elevated but it had fallen that day even as some AI-related corporate debt remained under pressure. Government borrowing costs and the financing conditions facing individual companies were moving in different directions.
[3:08] For a company building a data centre, the question is straightforward. What will it cost to borrow the money? And can the project earn enough to pay it back? Treasury yields are part of that answer, but only part. Think about applying for a loan. The Bank of England might cut its base rate, But if your lender becomes less confident that you can repay, the offer that you receive could still be more expensive. Companies face a similar calculation. Investors look at the benchmark rate, then ask for an additional return to compensate for the risk of lending to that business. Now that extra premium is the credit spread. So the benchmark can fall while the risk premium rises. The final borrowing cost depends on which moves further.
[3:59] On Friday, one widely followed measure of the US high-yield spreads stood at 3.31 percentage points. In broad terms, investors wanted that much additional yield above comparable government debt to hold those riskier corporate bonds. That is a useful indication of the mood among lenders. But a developer cannot finance a building at the market average. It has to secure its own terms, based on its customers, its finances, and the project it promises to deliver. For some AI borrowers last week, those terms looked demanding, judging by the yields investors wanted on their bonds. Now, that does not mean an existing fixed-rate interest bill suddenly goes up. A company that has already locked in its funding has some protection. The pressure becomes more immediate when it needs another loan or has debt coming due. Floating rate borrowing can respond sooner. And in this build-out, the next funding requirement may arrive well before the first customer payment.
[5:05] The same timing problem applies when companies raise money by selling shares. The postponed listings of SB Energy and Firmus show that a planned fundraising cannot be treated as cash already in the bank. At the other end of the AI industry, the model developers are competing for capital too. Reuters reports that OpenAI is seeking about $30 billion in fresh funding, with an accelerated timetable ahead of Anthropik's expected stock market listing. Neither the fundraising nor the expected listing should be treated as completed. The two businesses are approaching some of the same investors. Even investors enthusiastic about AI have choices to make. How much do they commit? Which businesses do they back? And at what valuation?
[5:54] For companies selling shares, a lower valuation can mean giving up more ownership to raise the same amount of money. For borrowers, tougher terms can mean a larger interest bill or tighter conditions. Either way, the money has to arrive in time. A business can have strong demand and a promising future, but its contractors, equipment suppliers and lenders still need paying while it gets there.
AI Demand Versus Returns
[6:21] This brings us to the apparent contradiction at the heart of the AI supercycle. ASML reported record third-quarter sales of €11.7 billion.
[6:32] It said the capacity it currently plans for 2027 and 2028 would not fully meet customer demand. Some customers are already expressing interest in placing orders for 2030. That is a striking indication of their ambitions. although it is not the same as having firm orders booked that far ahead. TSMC's results have also showed strong demand for advanced chip manufacturing. I think these are really encouraging signals from the supply chain. However, they do not answer the question of what customers will eventually earn from all of that equipment. CoreWeave sits further along that chain. It buys access to the infrastructure needed to provide AI computing services, and sells those services to customers. At the end of June, it carried about $35.6 billion of financial debt, alongside substantial lease obligations. It remains loss-making on a net income basis, while management points to a revenue backlog of about $100 billion.
[7:35] This backlog provides visibility. It is not cash in the bank today. Some core weave bonds traded below 80 cents on the dollar last week, with the yields reaching about 16%. These prices are showing investors demanding substantial compensation for the risk. There was some relief on Monday. Reuters reported that the yield on its 2031 bond eased to about 15.13% from above 16% last week. Now that doesn't sound like much, but it does matter, and we should acknowledge an improvement when it happens, even while recognising that the yield remains elevated. Now there is a credible bullish argument here. Goldman Sachs upgraded Corweave to buy on Monday, pointing to tight computing capacity and the company's performance advantages. That is a broker's assessment, but it helps explain why some investors see an opportunity in the sell-off. Valuable computing capacity and difficult financing conditions can exist at the same time. Now strong demand matters, and so does the cost of getting that capacity into service.
[8:45] We should also distinguish pressure on one company from pressure across the banking system. JP Morgan estimates that most global banks have limited exposure to CoreWeave, although it acknowledges incomplete disclosure. That puts the banking risk in perspective. It does not establish that shareholders, bondholders and other lenders face the same degree of protection. There is also an important support arrangement. Under an agreement disclosed in September last year, Initially valued at $6.3 billion, NVIDIA agreed to purchase residual unsold core weave capacity through April 2032, subject to contractual conditions. So the chip supplier is also helping support demand for computing services provided by its customer. Now that can be commercially useful to both sides, but it is not a blanket guarantee of CoreWeave's debt, and it does not remove the need to deliver working capacity on time.
Power Delays and Project Economics
[9:43] We now have a real example of what happens when the physical and the financial clocks fall out of step. Microsoft has withdrawn from an agreement for one N-Scale data centre in Loughton, Essex. The site was intended to begin operating early next year, but its expected grid connection had slipped into the 2030s.
[10:03] Microsoft says this concerns a single site, and both companies say their wider partnership continues. So we should be careful about reading it as a retreat from AI. A customer may want the computing power, yet still be unable to wait for that particular project. Now put that alongside another report, N-Scale is in talks with Anthropic over a further computing agreement in West Virginia that could be worth around $45 billion. That agreement is still being negotiated. Its headline value is neither cash received today nor annual revenue. Nevertheless, the contrast is revealing. The same developer can be discussing enormous additional demand in one location, while losing a customer agreement in another. The question becomes much more specific than whether the AI demand is strong. Can this project deliver the computing power in the right place at the right time on terms that leave an adequate return.
[11:03] While a project waits, financing costs can accumulate, equipment ages and the cash cushion shrinks. A delay can begin as a timing problem and become a problem with the economics of the whole project. Gartner added a longer term warning last week. It predicts that more than 80% of specialist AI cloud providers, often called neoclouds, will cease operating independently by 2030. Now that is a forecast, not a roll call of future bankruptcies. Some businesses could be acquired or merged, but it does illustrate the pressure on providers that need substantial capital while competing to turn computing capacity into profitable revenue.
[11:46] Think of a builder's merchant when credit tightens. The customer paying cash on delivery is in a different position from the customer asking for another 90 days. In AI, the equivalent question is how much a company can fund from its existing business and how much depends on investors financing the next stage. Large companies such as Microsoft have substantial operating cash generation. That gives them options. It does not make every investment worthwhile or every commitment easy to fund. And weakness in memory and equipment shares does not prove that demand has collapsed. Equally, strong orders do not prove that demand expectations are unchanged. Investors may be reassessing valuations, future margins, customer finances, or several things at once.
[12:35] Mundy brought a rebound in technology and semiconductor shares. The improvement in Corweave's bond yield means it was not entirely an equity-only recovery. But one session cannot tell us how durable that repair will be. The more useful question is whether confidence in the businesses is improving alongside the terms on which they can raise money.
Oil, Inflation, and Energy Risk
[12:58] This takes us from demand for computing to the energy and infrastructure needed to deliver it. And this morning, the oil market has complicated that picture again. Brent oil briefly reached just over $111 per barrel, its highest level in five months. So the easing we saw during Monday's trading session has given way to renewed supply fears.
[13:21] The immediate backdrop is reports of attacks on vessels in the Gulf and Houthi claims of attacks on Saudi energy and other infrastructure. Saudi authorities had not confirmed those claims in the report, and Reuters could not independently verify them. I think we need to be careful here because we shouldn't turn that into a claim that a particular quantity of oil production has been knocked out. Prices can rise because traders fear a disruption before any additional loss of supply is established. That is the risk premium. Buyers paying more because dependable supply feels less certain. For the wider economy, sustained higher oil prices can feed into transport, distribution and other business costs. Some companies can pass those costs on, as we discussed last week. Others have to absorb them, and households can find that they have less to spend elsewhere, as energy prices increase.
[14:18] For AI investors and traders, there is a second connection. Persistent energy inflation could make it harder for central banks to ease policy, leaving financing conditions restrictive for longer. That is a possible transmission mechanism, not a prediction of the next rate decision. Nor does a rise in Brent oil translate directly into the same rise in a data centre's electricity bill. Local generation, tariffs and power contracts do matter here. As N-Scale's experience shows, getting a connection can matter more than the quoted energy price. There are differences within natural resources too. Higher crude prices can support producers' revenues, while fuel buyers face a bigger bill.
[15:01] Refiners sit between the two. Their returns depend on the gap between crude oil costs and the prices of the products they sell, as well as operating and transport costs. That is why this week's Valero and airline results are useful. Can businesses pass costs on? Are margins holding up? And are customers changing their behaviour? A higher oil price alone cannot answer those questions. Behind those prices is the question of whether energy can move reliably from producer to customer. The latest report from Kepler estimates that at least 18 vessels were reportedly attacked over two weeks. Now that helps explain the concern, but it is not a measure of how many barrels have failed to reach the market. We need to separate threats, reported attacks, confirmed damage and actual disruption to the deliveries, because they are all connected, but they are not interchangeable. Shipping operators also have to consider crew safety, insurance and whether a route remains commercially workable. Oil can still be moving while the journey becomes more expensive and less predictable.
[16:10] One busy day through the Strait of Hormuz would not establish normal passage. Equally, the one very weak day could not establish a permanent closure. Potential relief from changes to Russian energy restrictions also needs to be judged by the fuel that actually reaches buyers. A policy announcement cannot tell us how much additional supply will arrive, or how quickly. The same test applies to diplomacy. An agreement does matter, but operators and insurers would need confidence that vessels can move safely. If deliveries remain reliable and tensions ease, some of the risk premium could unwind. If disruption worsens, energy costs could remain a complication for inflation, margins and investment. That gives us another reason to avoid drawing a sweeping conclusion from Monday's rally. The backdrop can change while the share price is recovering. And cheaper oil, when it comes, will not by itself connect a data centre to the grid or refinance a developer's debt. It can improve the conditions around a project while its own problems remain.
PRISM Framework and Market Outlook
[17:18] Okay, so let's move on to the read from my framework, and we'll discuss my base case as well. As many of you know from previous episodes, PRISM is the framework that I use to analyse the AI supercycle, and I've designed it over many years to turn complex, fragmented, and sometimes conflicting evidence into structured, probability-weighted decision intelligence.
[17:41] It integrates multiple layers of analysis, including everything that we discuss in this podcast, such as macroeconomic and structural conditions, enterprise and industry fundamentals, capital formation and infrastructure development, positioning, sentiment and market behaviour, quantitative and statistical evidence, price, risk and also execution signals. PRISM produces structured evidence assessments, base cases, alternative scenarios, confidence levels, decision triggers, key risks and thesis invalidating evidence. However, it does not claim to eliminate uncertainty or replace human judgment. Its purpose is to make that judgment more systematic, transparent and responsive to changing evidence. So here's my read based on the framework. There are early signs of relief, but I don't yet see enough evidence to call this a durable recovery. The encouraging side is real. The latest chip-making results support strong underlying demand. Mundy brought a rebound in technology shares and some improvement in Corweave's bond yield. We should acknowledge both of those. But we need to watch the markets closely this week because the unresolved question is whether that improvement lasts and whether more businesses can secure funding on workable terms.
[19:03] This morning's renewed rise in oil adds another complication. It does not tell us where technology shares go next, but it weakens the argument that energy is already providing dependable relief. So I would resist confidently labelling this either a temporary relief rally or the moment to buy the dip. A rebound does not establish a new uptrend, and I think it's too early to tell. Equally, a difficult week does not invalidate the long-term demand for computing. Nothing has really changed in the physical layer. There is a time horizon problem in the question too. A rally that fades over the next few sessions could still occur during a period that eventually offers attractive long-term investments. Someone trading this week and someone investing over several years are asking different questions. And from my own perspective, I trade on multiple time horizons. And that changes how I view the markets. it changes the lens through which I view the markets. For this episode, I'm asking whether the rebound is showing signs of becoming a sustained recovery. A more convincing recovery would have more companies participating, rather than relying on a handful of names. So we're talking breadth, because what we've seen over the last few months is a degree of concentration. And we've talked about concentration risk in a number of previous episodes. It would also have support from company results and improving financing conditions.
[20:31] No single earnings beat or a day's trading would resolve this. And the growth of an industry is not the same thing as the return earned by every participant.
Demand can expand while individual projects struggle to pay back their investors. So I'm not trying to call the bottom of the current price action in the markets. I'm asking whether the evidence behind the recovery is improving. That is a judgment we can update as new information arrives, without dressing uncertainty up as a precise probability.
Earnings to Watch This Week
[21:02] So what am I watching this week? First and foremost, I'll be watching the earnings. Now I'm recording this before the releases, so by the time you listen, some may already be available, and you'll have the results to analyse. The questions remain useful either way. So let's start with the company's spending on AI. Meta reports on Wednesday, the 21st. Its latest guidance puts this year's capital expenditure, including finance lease principal payments at $130 to $145 billion.
[21:34] I'll be listening for how its cash generation compares with those commitments, and what management can show about the returns. Maintaining a large spending number is only part of the story. External financing would not automatically be a warning sign. A partnership can share costs and risks on sensible terms. So what matters is the cost of the money and the obligations that remain with the company. SpaceX is scheduled to report on Tuesday the 20th, with Tesla on Wednesday. So I'll be looking at how much cash their operating businesses generate, how much the next phase of investment consumes, and how they expect to fund the difference.
[22:16] The useful comparison is between the businesses paying today's bills and the ambitions drawing on tomorrow's capital. Then we'll look at the customers putting AI to work. IBM also reports on Wednesday, so its commentary can help us understand what enterprises are actually buying and how that demand is turning into business. An order, recognised revenue and cash received are different stages of that journey. The infrastructure suppliers provide another perspective. LAM Research's Wednesday results can shed light on the equipment demand, while Blackstone, which owns data centre operator QTS, reports on Thursday. Their commentary can help distinguish a delayed individual project from wider pressure across the build-out. And beyond technology, United Airlines is scheduled to release results on Tuesday, with American Airlines on Thursday. Alongside Bilero, they provide a window into fuel costs, pricing and demand, and that helps connect the energy story to the wider economy.
[23:17] One final point. Separate the results from the market's reaction. A company can report growing profits and still disappoint expectations. Its shares can rally because the news is less bad than feared, without its longer-term economics having improved much at all. So listen to what management says about the future, look at what the numbers actually show, and then consider the price reaction. Stronger evidence across all three would be more persuasive than a headline beat alone. The Federal Reserve meets on the 27th and the 28th. These company reports will arrive before that decision, adding evidence on demand, costs and investment.
[24:00] And, you know, there's going to be plenty of noise. The useful question is whether we are learning something that changes the economics behind the rebound. So, a buy-the-dip opportunity or a temporary relief rally? At this stage, I would treat the rebound as an improvement to examine, rather than a verdict. The earnings ahead may strengthen the recovery argument or expose weaknesses that a rising share price has temporarily obscured. The latest chip-making results still show strong demand, which we've covered previously in this episode. Customers are planning years ahead, yet power delays are already affecting individual projects, and renewed oil market tension is complicating the wider backdrop. At the same time, investors are asking harder questions about the cost of financing those plans, and the returns they will eventually produce. The physical clock runs at a pace of power connections, construction and equipment delivery. The financial clock runs at the pace of interest payments, refinancing dates and investor confidence. And it never stops ticking. A business with sufficient cash and dependable revenues may be able to absorb a delay. A business relying on its next funding round may have much less room.
The Question: Who Can Afford to Wait?
[25:18] That is the question running through this week's story. Who can afford to wait?
Trading the AI Super Cycle
[25:24] Okay, so before I finish, last week I promised to tell you a little bit more about the book I'm working on, Trading the AI Super Cycle. Well, I started it as one book, but at around a quarter of a million words, it became clear that I needed a bit more space. It's now a three-volume series, with one thread running through it, Following the Money. Volume 1 looks at where the capital comes from, what it costs, and where it goes first. That means interest rates, financing, geopolitics and the spending decisions of big tech before moving into the models, chips, networks and data centres at the heart of AI. Volume 2 follows that investment into the physical economy. Computing needs electricity, grids, materials and the transport to get things built. As we've seen this week, a customer can want the capacity and a developer can have a plan Yet a power connection can still hold everything up. This volume explores those constraints and what they mean across energy, commodities and the companies that supply them. Volume 3 asks, how do you turn that understanding into a disciplined trading and investing process? How do you weigh different scenarios, choose a way to express your view, time and entry and protect your capital? And afterwards, how do you judge whether your reasoning was sound? Rather than simply whether the trade made money.
[26:54] The series draws on historical parallels and the PRISM framework behind my research. And I'm writing it for people managing capital professionally and serious, independent traders who want to understand how these markets fit together.
Understand the capital, follow it into the physical economy, and then trade with discipline. That's the thread connecting all three volumes and the thinking behind this podcast.
Research, Resources, and Sign-Off
[27:20] Okay, so you'll find our work at qfmi.substack.com, where the weekly outlook, market pulse, and in-the-spotlight articles are all free. The subscription research, including the data centre report I mentioned last week, is moving to the QFMarkets website at quantumfields.ai. We're currently in the process of developing this and aiming for a November launch. That will also become the home for research and insights from my market intelligence platform and Prism model, bringing everything together in one place. I'll keep you posted as we get closer to the launch. If there's a question you'd like me to explore or a topic you think deserves a future episode, I'd be very interested to hear from you. You can find the email in the show notes. Okay, so that's a wrap for this week. If you found this episode useful, please follow the podcast and share it with someone trying to make sense of the capital flows, physical constraints and commercial returns shaping the AI super cycle. 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.