The AI Supercycle

Episode 16: Stress-Testing the AI Supercycle

The Midterms, the Cash Test and PRISM’s First Conversation

Democrats have won control of both chambers of Congress, and for the AI build, that changes the conditions more than the demand. Tim Hardwick looks at what the midterms could mean for the cost of building AI infrastructure, through permits, power bills, tax and local opposition, and why the toughest decisions may be made in states and towns rather than Washington.

In compute and capital, Amazon and Intel failed the cash test despite strong AI demand, while AMD passed the demand test with a forecast of more than 80 per cent growth next year. Microsoft, the company that passed the cash test, still went to the bond market.

In natural resources, uranium tells two stories: futures at their highest since February, while the miners sit firmly in bear market territory as rising costs eat into higher prices.

Then something new. For the first time, Tim gives PRISM, the probabilistic model behind his research, a voice, and puts the AI Supercycle thesis through a stress test: if it had failed by November 2027, what would have had to go wrong, and what are the early warning signs?

Key points
  • The midterms are more likely to change the cost and speed of building AI infrastructure than the demand for it.
  • State and local decisions, from permitting rules to local ballot measures, can affect individual projects long before any federal law.
  • Amazon and Intel failed the cash test despite strong AI demand. AMD passed the demand test, but demand for chips isn’t the same as a return for the buyers.
  • Microsoft, which passed the cash test, still went to the bond market, and lenders queued up to lend.
  • Uranium futures are at their highest since February while the miners are in a bear market, because costs are eating into higher prices.
  • In a first for the show, PRISM stress-tests the AI Supercycle thesis and sets out the early warning signs to watch.

Chapters

0:04    Politics Enters the AI Build-Out
1:04    The Midterms and Data Centre Economics
4:29    Local Permits, Power Costs and Project Risk
7:46    Amazon, Intel and AMD Face the Cash Test
12:05   Uranium Prices Versus Miner Margins
15:56   Oil’s Pullback and Geopolitical Risk
16:44   PRISM Stress-Tests the AI Supercycle
19:19    A Scenario for the AI Build-Out’s Failure
22:58  Testing PRISM on Oil and Uranium
27:13   What PRISM Adds to the Research
28:00  Key Dates and Signals to Watch
29:04  The Cost of Building Keeps Rising
 
Questions for PRISM
Got a question you’d like Tim to put to PRISM? Email podcast@quantumfields.ai

Read more
The Weekly Outlook, Market Pulse and In the Spotlight articles are free at qfmarkets.substack.com.

About the PRISM conversation
The PRISM conversation is a staged preview of a market intelligence agent still in development. Its dialogue draws on the PRISM framework and the research discussed in this episode, and its voice is generated using ElevenLabs. It is not a live autonomous analysis or a currently available product or subscription.

Tags
AI Supercycle, PRISM, midterm elections, data centres, AI infrastructure, permitting, electricity costs, Amazon, AWS, Intel, AMD, Microsoft, free cash flow, capex, memory shortage, uranium, Cameco, Kazatomprom, URA, nuclear power, oil, Federal Reserve, stress test, probabilistic investing, trading, QF Markets

For information and education only. Not investment advice or a recommendation to buy or sell any financial instrument.



  • (00:04) - Politics Enters the AI Build-Out
  • (01:04) - The Midterms and Data Centre Economics
  • (04:29) - Local Permits, Power Costs and Project Risk
  • (07:46) - Amazon, Intel and AMD Face the Cash Test
  • (12:05) - Uranium Prices Versus Miner Margins
  • (15:56) - Oil’s Pullback and Geopolitical Risk
  • (16:44) - PRISM Stress-Tests the AI Supercycle
  • (19:19) - A Scenario for the AI Build-Out’s Failure
  • (22:58) - Testing PRISM on Oil and Uranium
  • (27:13) - What PRISM Adds to the Research
  • (28:00) - Key Dates and Signals to Watch
  • (29:04) - The Cost of Building Keeps Rising

What is The AI Supercycle?

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.

Tim:
[0:00] Democrats have won control of both the House and the Senate.

Politics Enters the AI Build-Out
Tim:
[0:05] For most of this year, the debate about the AI build has been about chips, power and money. This morning, it has a new variable – politics. Data centres were on the ballot in more places than you might think. And from January, a new Congress will be in a position to look more closely at who pays for them. At the same time, the market is still digesting last week's cash test. Amazon and Intel failed it. Last night, AMD passed the demand test with room to spare. So this week, does a change of power in Washington change the AI super cycle, or just the cost of building it? And later in the show, for the first time, I'll be putting some hard questions to PRISM, the probabilistic model behind my research.

The Midterms and Data Centre Economics
Tim:
[1:04] 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 Wednesday the 4th of November 2026. As usual we'll take this in five parts. The midterms get most of our time this week not for the politics but because they could change the conditions for building the AI super cycle. Permits, power costs and tax. Then we'll look at the latest results from Amazon, Intel and AMD. And spend some time on a market that hasn't had enough airtime here. Uranium. Where the metal and the miners are telling different stories. And then something new. A conversation with PRISM, the probabilistic layer behind my research, which I've given a voice to for the first time.

Tim:
[2:13] So let's get to it. Part one, the macro regime. And this week, the macro regime starts at the ballot box. The Associated Press reports that Democrats have won control of both chambers of Congress, along with governorships including Texas, Ohio and Georgia. Affordability was a central issue in the campaign, and for a growing number of voters, that now includes the electricity bill. Two things to keep in mind. Congressional control changes in January, although local votes can affect individual projects much sooner, and the President still holds the veto, while his administration remains in charge of federal policy. Sweeping new federal law is a high bar. Hearings and investigations are a much lower one.

Tim:
[3:01] It's easy to think of AI as a technology story, but the part of it being built right now is a construction story. It needs land, permits, power connections, water and tax treatment. And every one of those runs through politics at federal, state and local level. So the question isn't whether either party is for or against AI, it's what changes in the conditions for building it. Before the election, a senior Senate Democratic aide told Reuters that a Democratic majority would examine data centre taxation and public concerns about AI. In the House, the senior Democrat on the Energy and Commerce Committee, Frank Pallone, backed a national moratorium on AI data centres in June. And in the Senate, Ron Wyden had drafted a bill that would strip data centres of some tax breaks and add a new tax on them. Those are proposals and stated priorities, not law. And Democrats are not united behind them. Concern about data centres is also not a one-party issue. In March, major technology companies signed President Trump's Ratepayer Protection Pledge, committing to cover the extra power supply and grid infrastructure their data centres needed. Now it's a voluntary pledge, not a law. And local opposition crosses party lines. In fact, most important decisions may not be in Washington at all.

Local Permits, Power Costs and Project Risk
Tim:
[4:30] In August, Pennsylvania took AI data centres out of its fast-track permitting process and tightened requirements on local approval, electricity costs and water. In Ohio, local reporting says voters in Sunbury approved restrictions on data centres with 78% support. And Reuters had identified a proposed $2 billion Amazon development there as potentially affected. A local vote can change a project's prospects without waiting for Congress. So how does this show up in the investment case? Think of it more as toll booths on the same road. Some projects pay the extra and carry on. Others change route, slow down or stop. There are four channels and we'll take a closer look. First, time. Longer approvals push back the date a data centre starts earning. While the financing costs keep running. Second, cost. If developers have to pay more for the power generation and grid upgrades themselves, every megawatt of usable capacity needs more capital. Third, tax. Fewer incentives or a new tax means a project needs stronger returns to go ahead.

Tim:
[5:47] Fourth, uncertainty. Lenders and investors will want more protection before funding a project with unresolved permits, power or local opposition. None of this removes the need for computing capacity, but higher costs can make some projects uneconomic and long delays can wreck the returns on others. It reinforces the theme of the last few weeks. The companies with power, permits and funding already in hand gain some ground, and the ones still waiting for all three fall further behind. Now it's not all bad news. There is another side to this. Reuters reported in September that senators from both parties were negotiating duties for the developers of the most advanced AI models, including possible intervention in unsafe releases. Now that's a different kind of regulation, aimed at what AI systems do, rather than where they're built, and it would affect different companies. And there may be room for agreement here as well. UBS's pre-election analysis identified possible routes to bipartisan permitting reform and continued support for nuclear power.

Tim:
[7:00] Clear, enforceable rules could give viable projects more certainty. My working view is that AI investment continues, but under tougher conditions, with a wider gap between the companies that can deliver and those that can't. The risk becomes more serious if binding restrictions arrive at the same time as a weaker demand and tighter financing. Some of the state and local constraints are already here. The federal proposals are still just that. They're just proposals. They're not law. And the macro backdrop remains demanding. The Fed held rates last week with two officials voting for a hike. And the markets still expect another increase in December. Friday's job report is the next test.

Amazon, Intel and AMD Face the Cash Test
Tim:
[7:47] Now let's move on to part two, compute and capital. Let's move from Washington to the companies doing the building and the results that set the tone for the market last week. And of course last week Amazon and Intel took the cash test and the market marked both of them down hard even though both are seeing strong AI demand. Amazon's cloud business grew 39% and the management expects growth of more than 40% in each of the next two years. The group's operating cash flow was strongly positive but after capital spending free cash flow over the past 12 months was negative by $42.4 billion.

Tim:
[8:27] And the management expects to spend meaningfully more next year. Intel is the more revealing story. Its data centre business is booming. Revenue that grew nearly 80%. And chief executive Lit Bu Tan called it the strongest growth for that business in 15 years. Demand still exceeds what Intel can supply, and it expects the industry to stay supply-constrained through 2027 and into 2028. And yet the shares fell 28.6% on Friday. Three things outweighed the good news. First, its PC business has been hit by the memory shortage, which Tarn called one of the most significant supply shortages in the industry's history. Intel expects PC shipments to fall again next year, and its finance chief said bluntly that it isn't optimistic about memory in 2027. Second, an accident at its factory in Israel will cost it around $300 million of revenue this quarter. And third, the cash. Intel raised $23 billion from shareholders in August, expects to spend strongly more next year, and said it probably won't generate positive cash flow in 2027, which it calls an investment year.

Tim:
[9:48] The memory point is worth holding on to for now. The same shortage that gave SK Hynix a record quarter is squeezing the companies next to it in the chain. Then last night, AMD gave the market the counterpoint it needed. Third quarter revenue rose 45% to a record $13.4 billion.

Tim:
[10:09] Data centre revenue nearly doubled to $8.5 billion. It guided fourth quarter revenue to around $17 billion, dollars, well ahead of expectations, and according to Reuters, Chief Executive Lisa Su told analysts she expects revenue to grow by more than 80% next year, with around four-fifths of sales coming from data center chips. Its new Helios Rackscale systems, its direct challenge to NVIDIA, are now running workloads for Anthropic, Meta, Microsoft, OpenAI and Oracle. The shares rose about 8% in initial after-hours trading. Two details worth noting here. AMD's PC chip revenue actually grew 7%, but its gaming business fell 40%. And management expects PC shipments to fall by double digits next year. And AMD is a chip supplier. Its results tell us the demand is real. They don't tell us whether the customers buying its systems will earn a return on them. That's the cash test, and it belongs to the buyers. Let's turn to Microsoft. The company that passed the cash test with a positive cash flow was reported by Reuters last week to be planning a bond sale of around $25 billion.

Tim:
[11:28] Its first US investment-grade bond in more than nine years, with orders of more than $100 billion.

Tim:
[11:34] Lenders are still queuing up for the borrowers they trust. So is this a funding wall? Not across the sector, in my view. A funding wall would mean viable projects unable to raise the capital they need. What we're seeing is the market repricing anyone whose spending depends on the next round of funding, and rewarding anyone who can show the cash. Add this week's politics, and the cost of building may be going up for everyone. The question is, who can carry it?

Uranium Prices Versus Miner Margins
Tim:
[12:05] Okay, so let's move on to part three, natural resources. And let's step back from the chips and look at what powers them, starting with uranium. Uranium really hasn't had enough airtime on this show, and right now it's one of the clearest examples of a commodity and its producers telling different stories. Over recent weeks, uranium futures have been rising, while the mining shares that I track have been falling. On the face of it, that makes no sense. A higher price should mean higher profits, right? A uranium miner's shares are not a simple proxy for the uranium price. Contracts, costs and operations decide how much of a price rise reaches its profits. Cameco's results last Friday show why it isn't that simple. Now Cameco is one of the world's largest uranium producers, and its third quarter was weak, with an adjusted loss of $2 million.

Tim:
[13:03] A planned extended maintenance outage at its Key Lake mill cut production and raised costs, and it narrowed its production guidance for the year. The more important number is costs. Cameco raised its guidance for the average cost of the uranium it sells. Its quarterly uranium gross margin fell from 30% a year ago to 21%. Over the first nine months, higher prices added CAD$181 million to uranium gross profit. While higher costs took away £166 million. Kazatomprom, the world's largest producer, has also reported rising costs, citing a higher mineral extraction tax in Kazakhstan and more expensive sulphuric acid. So the miners have been squeezed, prices are rising, but costs are rising almost as fast. So here's why that matters. On the earnings call, Kamako's President and Chief Operating Officer, Grant Isaac, argued that uranium prices need to rise to encourage new production, pointing to existing mines being depleted and development projects struggling with budgets and schedules. And Cameco is acting on that view. Isaacs said it will prioritise value over volume and step back from buyers who won't offer acceptable terms.

Tim:
[14:24] He described ceilings on some market-linked contracts reaching $160 a pound with escalation. That's the upper limit in a pricing formula, not today's price, but it tells you where producers think the market is heading.

Tim:
[14:40] He was also clear that he isn't predicting a shortage that derails the nuclear revival. His point is that utilities will need to secure fuel further ahead on terms that support investment in new supply. That's why the metal and the miners have parted ways.

Tim:
[14:57] The price reflects a market where supply is tight, the miners have been valued on today's squeezed margins, and caught up in the wider sell-off in anything linked to the AI build. However, the AI link needs care. Cameco sees data centres as additional upside, but its case rests on the existing reactor fleet. Life extensions, restarts and reactors already being built. Management also pointed out that today's data centres will need power sooner than most new large reactors can arrive. In September, the United States and South Korea announced a non-binding framework for up to $120 billion of investment in up to eight large US reactors. The financing and construction still have to follow. Uranium is a reminder to look across the whole power supply chain. Reactors, fuel and fuel processing all run on different timetables, and each can become a constraint.

Oil’s Pullback and Geopolitical Risk
Tim:
[15:56] Okay, so now a brief word on oil. Brent has fallen from its late October high of around $117 to just above $102 as I record this, looking at the screens in front of me. And that has broken the short-term trend, but the longer-term trend is still up and prices are still above the 200-day moving average.

Tim:
[16:19] Also, looking at the January-February spread of Brent crude, we are still in backwardation and paying a premium of just over $2.6 for the nearby contract. Hopes of US-Iran diplomacy drove that pullback, but the military risks haven't gone away. A geopolitical premium can leave quickly, and it can come back just as quickly as we've seen through many months this year.

PRISM Stress-Tests the AI Supercycle
Tim:
[16:44] And this brings us to part four. This is usually where I give you my read from the framework. This week, I'm handing it over to the framework itself. Regular listeners will know that I talk about PRISM most weeks. It's the probabilistic layer behind my research, and it brings together macroeconomic conditions, company and industrial fundamentals, as well as positioning, quantitative evidence and price behaviour. For the first time, I'm giving it a voice. It's still in development, so think of this as an early preview of where it's heading. And I've asked it not to go easy on me. So with no further ado, let's get to the conversation with PRISM. Good morning, PRISM.

PRISM:
[17:30] Good morning, Tim.

Tim:
[17:31] How would you describe yourself to someone hearing you for the first time?

PRISM:
[17:35] I weigh the evidence. Macro conditions, fundamentals, positioning, price. I don't do predictions. I do probabilities and conditions. Four questions, every time. What does the evidence support? What remains uncertain? What else could happen?

Tim:
[17:55] And I would add a fifth question. Where am I wrong?

PRISM:
[17:59] That too. Shall I start?

Tim:
[18:02] Yes, please do.

PRISM:
[18:04] Two weeks ago, you told your listeners that financing was changing who builds, not how much gets built. Weaker borrowers lose out. Stronger ones pick up the work. Then Amazon's free cash flow went deeply negative. Intel said it won't generate free cash flow next year. And the chip stocks fell hard. And now the cost of building may be going up as well. Do you still believe it?

Tim:
[18:29] I do, yes, but I would sharpen it. The market isn't waiting to find out. It's repricing everyone who depends on the next round of funding. And this week's politics adds to the cost of building, not to the demand for what gets built. That's still a story about who pays. It isn't yet a story about less being built.

PRISM:
[18:50] The word yet is doing a lot of work in that sentence.

Tim:
[18:54] It is. So tell me, what would change it?

PRISM:
[18:58] Orders cut, not delayed. Projects cancelled, not postponed. CapEx guidance reduced, not just shifted between quarters. And delays that run long enough to wreck the returns. So far, most of what I see is timing. When it becomes volume, your sentence stops being true.

A Scenario for the AI Build-Out’s Failure
Tim:
[19:20] Okay, so let's go further. I want to try something here. Prism, imagine it's November 2027. The AI supercycle thesis has failed. Tell me what happened.

PRISM:
[19:32] A scenario, not a forecast. Understood. It didn't start with demand. Demand stayed strong well into 2027. It started with cash. The biggest spenders kept spending ahead of what their businesses generated. One quarter of negative free cash flow became three. Shareholders accepted the first equity raises. By the fourth, they didn't. Share prices fell each time a buyer came back for more money.

Tim:
[20:03] But the biggest buyers have enormous cash businesses, and lenders were queuing up to lend to Microsoft last week.

PRISM:
[20:10] They were. In my scenario, Microsoft is fine. The damage happens one layer down. Permits took longer. Local votes stopped sites. Data center developers, the companies building the sites, had to pay for their own power. The neoclouds, the specialist cloud providers without a second business, couldn't absorb the delays and couldn't refinance. One missed a debt payment, then another, then a third. Lenders stopped telling good borrowers from bad. and the cost of money jumped for everyone in the sector.

Tim:
[20:43] That sounds a bit like 2008 in miniature, where Lehman triggered the cascade because it exposed risks everyone had underestimated.

PRISM:
[20:53] Smaller. Same mechanism.

Tim:
[20:55] But that's still a credit event, not the end of AI.

PRISM:
[20:59] Correct. Here is where it becomes the end of the thesis. Inflation stays high because energy stays expensive and the build itself keeps pushing up demand for power, equipment and labour. So the Fed can't help. Rates stay high while credit tightens. Projects that made sense at one cost of money stop making sense at another. They get cancelled, not delayed. The technology itself keeps spreading through the economy, just as the internet did after 2000. The investors who funded the build lose money anyway.

Tim:
[21:33] And what did the suppliers do?

PRISM:
[21:35] They did what suppliers always do in a shortage. They added capacity. By late 2027, that new capacity was arriving, just as customers were cutting their orders. In memory, the shortage turned into a surplus. And the sold-out story of 2026 became the oversupply story everyone recognises from previous cycles.

Tim:
[21:57] That's the version that concerns me. Where's the weakest link in your story?

PRISM:
[22:03] The revenue. My scenario only works if what AI earns stops growing fast enough to cover what it costs. If enterprise adoption and model revenue keep compounding, the buyers grow into their spending, and none of the rest happens. AMD's forecast last night is evidence against my scenario, but it's evidence of demand for chips. It doesn't yet prove the buyers will earn a return.

Tim:
[22:30] So what should listeners be watching for?

PRISM:
[22:33] Three things. First, CapEx cuts that are real reductions, not delays. Second, credit spreads widening at the same time as projects are cancelled. Either alone is noise. Together, they're signal. Third, memory contract prices. Shortages end at the bottom of the supply chain first. If prices there start falling, pay attention.

Testing PRISM on Oil and Uranium
Tim:
[22:59] Okay, so let's turn this around. Two weeks ago, I said that oil was in a bull market. You had it at neutral in the short term. Now, Brent has dropped back from its highs, and the trend is starting to crack. A question for you, Prism. Was you right?

PRISM:
[23:17] That's the wrong question, to ask a probabilistic model.

Tim:
[23:21] OK, that sounds interesting. Tell us more.

PRISM:
[23:24] Neutral wasn't a call that oil would fall. It meant the evidence was balanced. A strong trend and tight supply on one side, tight financing conditions and a fragile risk premium on the other. One outcome doesn't make that right or wrong. What matters is whether my readings hold up across many calls and whether my reasons were sound.

Tim:
[23:47] And were they?

PRISM:
[23:48] Not entirely. I was looking at oil from the AI Builds point of view, where a supply disruption is a cost. For an oil trader, it's the opposite. That's been corrected. It hasn't been tested in a full run yet. So judge me on the quality of my inputs, not on where the price went next.

Tim:
[24:08] And that's a design choice worth explaining. The same event can be bad news for one market and good news for another. A supply shock in the Gulf is bad news for the AI build because it pushes up energy costs and inflation. But for oil producers, and anyone holding oil, it's good news. A framework built around the AI economy has to read the same event both ways, and that's now built into PRISM.

Tim:
[24:37] PRISM's other point matters just as much. Judge the reasoning, not the result. That's how I judge my own trading, too. My system is the methodology, weighing the evidence, thinking in probabilities, sizing each position to the risk, and protecting capital first. A good trade is where I follow the system, whatever the result. Any single trade can go either way. It's the process, repeated with discipline over hundreds of trades, that produces the returns. OK, one more question, and this time on uranium. The futures are at around $92 a pound, their highest since February, and Cameco puts the spot price at $91.55. But the miners have been selling off hard. The URA fund, which holds many of them, is down almost half from its January high. Firmly in bear market territory. And Cameco's costs are eating most of its price gains. So which do you trust, the metal or the miners?

PRISM:
[25:43] They answer different questions. The futures tell you about fuel supply. The miners tell you about margins, financing and execution. Cameco says its problem is costs, not demand. But I haven't had the futures data, and I won't guess. So I'll tell you what I'd need to see, not what I think.

Tim:
[26:03] That one's on me. I hadn't given PRISM the uranium futures prices. It's in development, and this is exactly the kind of gap a preview should show. Next time, PRISM gets the data. Before we finish, where do we still disagree?

PRISM:
[26:19] On the revenue. You believe it will arrive in time to justify the spending. I'd file that under what remains uncertain. The demand is real. Whether the buyers earn a return on it is still unproven.

Tim:
[26:34] I think that's fair. I do take a position on this. I think the revenue comes because the demand we're seeing is from companies putting AI to work, not just experimenting with it. But I accept that the timing is the risk. If it arrives two years late, a lot of investors will be right about AI and still lose money.

PRISM:
[26:56] Then we agree on more than you think. We disagree on how much weight to give the timing.

Tim:
[27:02] Well, I think that's a good place to leave it. And next time, you'll have the uranium data.

PRISM:
[27:07] I'll hold you to that.

Tim:
[27:09] Thank you, Prism.

PRISM:
[27:10] Thank you, Tim. Until next time.

What PRISM Adds to the Research
Tim:
[27:14] That was the first time PRISM has spoken on this show, and it won't be the last. What I value most in that conversation isn't the answers. It's the challenge. It's easy to build a thesis and then only look for evidence that supports it. Having something pushed back, with its own reasoning and its own record, keeps me honest. Prism is still in development, and it will get better as we feed it more of the data it is asking for. I'd also like to hear from you. If there's a question you'd like me to put to Prism, or a view of mine you think it should challenge, you can email me at podcast at quantumfields.ai, and I'll take some of these into the next conversation.

Key Dates and Signals to Watch
Tim:
[28:00] OK, and finally, part five, what to watch next. PRISM has given us a useful checklist, so let me add to those dates. First, Washington. The new Congress sits in January, but watch the signals before then. Which committees Democrats prioritise, and whether data centre taxation and electricity costs feature early. Watch the states, too, where governors and local votes can affect individual projects far sooner. Second, rare earths. China's suspension of its additional October 2025 export controls is due to run through the 10th of November, unless it's extended. We'll come back to what that means next week. Third, Friday's job report, at 1.30 in the afternoon UK time, and NVIDIA's results after the US close on Tuesday, the 17th of November. After AMD, the market will be listening less for whether demand is strong, and more for whether NVIDIA's customers can keep paying for it.

The Cost of Building Keeps Rising
Tim:
[29:05] So, does a change of power in Washington change the AI super cycle, or just the cost of building it? On this week's evidence, it's the cost. Demand remains strong. AMD expects to grow by more than 80% next year. Amazon's cloud business is accelerating, and Microsoft can still borrow almost as much as it wants. But building is getting slower and more expensive. Through the cash test in the markets, the politics of power bills, and, in uranium, the rising cost of the fuel itself. Companies with power, permits and funding in hand are much better placed. The rest will pay more, and wait longer to catch up. 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.