The AI Supercycle

Episode 18: Nvidia’s Next Chapter

 Leadership, Capital and the Consequences Across the AI Stack

Most of the coverage has asked whether Nvidia can cope without Jensen Huang. Last night’s results suggest it can: revenue more than doubled to about $116 billion, Vera Rubin delivered the fastest product ramp in the company’s history, and the outlook was raised again. So this episode asks a different question. What happens to everyone who relied on the role Huang played?

For years, Huang acted as an informal clearing house for the AI build-out. When chips were scarce, Nvidia decided who got them. When customers needed capital, Nvidia’s money and his backing helped them raise more. Now that support rests on contracts, conditions and a queue that rewards whoever is ready. Tim works through the AI stack layer by layer to see who gains bargaining power and who loses it, looks at what kind of leader Nvidia needs next, and sets out the framework view. The episode closes with the latest news: Anthropic’s record-breaking IPO, a reported SEC investigation into Nvidia, and how AI is reshaping cybersecurity.

Key points
  • Nvidia’s physical clock is running fast; Huang’s departure is a financial clock event
  • Capital priorities have shifted: shareholders first, suppliers second, customer financing last
  • Bargaining power is moving down the stack to powered, permitted and funded sites
  • Around $105 billion of Nvidia’s $118 billion in guarantees supports a single campus
  • Falling token prices and rising GPU rents squeeze whoever sells AI output
  • The CEO search now has a national security dimension, and prediction markets have swung from insiders to outsiders
  • Anthropic’s filing describes the financing risk in the AI chain in its own words
Chapters 

00:00 Cold open
00:57 Introduction
02:18 What the earnings told us
05:38 What Nvidia really loses
08:20 What the market is telling us
10:25 Through the stack, layer by layer
19:00 What kind of leader comes next?
21:31 The framework view
23:23 What matters next
24:10 Where I’ve landed
25:00 Latest news: Nvidia
26:38 Latest news: Anthropic’s IPO
28:35 Latest news: AI and cybersecurity
29:55 Wrap-up 

Links
  • Articles and research: https://qfmarkets.substack.com
  • Last week’s episode: AI’s Battle for Power
  • Questions for a future episode: podcast@quantumfields.ai
Tags
Nvidia, Jensen Huang, AI Supercycle, AI stack, circular financing, AI infrastructure, data centres, semiconductors, Vera Rubin, Colette Kress, Anthropic IPO, cybersecurity, Palo Alto Networks, export controls, AI investing, trading

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

  • (00:00) - Cold open
  • (00:57) - Introduction
  • (02:18) - What the earnings told us
  • (05:38) - What Nvidia really loses
  • (08:20) - What the market is telling us
  • (10:25) - Through the stack, layer by layer
  • (19:00) - What kind of leader comes next?
  • (21:31) - The framework view
  • (23:23) - What matters next
  • (24:10) - Where I’ve landed
  • (25:00) - Latest news: Nvidia
  • (26:38) - Latest news: Anthropic’s IPO
  • (28:35) - Latest news: AI and cybersecurity
  • (29:55) - Wrap-up

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.

[0:00] For years, the AI build-out had one man at its centre. Jensen Huang didn't just sell chips. When they were scarce, he decided who got them. And NVIDIA's money, and his backing, helped customers raise the capital to buy them. Some call that circular financing. Others call it building a market. Now he's gone, and last night's results tell us NVIDIA itself is doing remarkably well. Most of the coverage is asking whether NVIDIA copes without him. I think that's the wrong question. The real question is what happens to everyone who relied on that support? And who, if anyone, steps into the role he played? Today, we follow that question through every layer of the AI stack.

[0:57] 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 18th of November 2026. Last week in AI's battle for power, I looked at the difference between owning operational capacity and still needing to finance, permit and build it. Hold on to that distinction because it turns out to be central to NVIDIA's next chapter. We'll start with what the earnings actually told us, then look at what NVIDIA really loses with its founder and how the market has read that. After that we'll work through the stack layer by layer, consider what kind of leader comes next, then the framework view and what I'm watching from here. But first, a quick update on the book. Editing on Volume 1 of Trading the AI Super Cycle is now complete. So we're into the final phase, which is proofreading, graphics and final production. So we're getting closer, and I'll keep you posted on when we publish.

[2:18] So let's start with the numbers. NVIDIA reported roughly $116 billion of revenue for the quarter, more than double a year earlier, and comfortably ahead of what Wall Street expected. Its new Vera Rubin platform brought in nearly $20 billion in its first quarter of production shipments, which management called the fastest product ramp in the company's history. It guided to around $130 billion for the current quarter, again above expectations, and raised its revenue growth outlook for fiscal 2028 from about 70% to about 75% off a higher base. Two qualifications come with that. The quarter ended before Jensen Huang left. So these results describe the business going into the transition, not through it. And next quarter includes an extra week, which flattens the forecast. The margin detail is more interesting than it first looks. Gross margins are expected to bottom next quarter around 71-72%, then recover to 72-73% next year. Kress explained why. NVIDIA is honouring orders already taken at the old prices, while absorbing higher supply costs, and moves to new pricing in the new year. Read that the right way around, and it isn't a company losing pricing power. It's a company that expects to pass higher costs on.

[3:44] The moment that mattered most for me came on customer financing. Colette Kress, now interim chief executive as well as chief financial officer, put the emphasis on discipline. But when an analyst asked whether the newly clarified conditions on NVIDIA's commitments amounted to a change in policy, her answer was blunt. Is it new? No. Did we change anything? No. But listen to how she ranked the uses of NVIDIA's cash, and a quieter shift appears.

[4:17] Alongside the results came a new $150 billion buyback authorisation, which NVIDIA described as the largest in its history, and a plan to return all excess free cash flow to shareholders. Crest put shareholder returns first. Next came suppliers, who need prepayments to expand capacity. Customer financing comes last, and here she was more cautious. She said NVIDIA will be selective and look at each case in detail, noting that a lot has changed over the last quarter. So there's a distinction worth drawing here. The terms on the commitments NVIDIA has already made haven't changed. What looks different is its appetite for new ones. And when customers sit third in the queue for NVIDIA's cash, That tells you where its next dollar is most likely to go. Here's how I'd put it together. On what I call a physical clock, the pace at which systems are designed, built and shipped, NVIDIA looks very healthy. Demand is strong and the new platform is shipping. Jensen Huang's departure is a financial clock event. It touches how quickly deals get financed, how much trust gets extended and who takes the risk. Those two clocks don't always run at the same speed, and much of this episode is about what happens when they drift apart.

[5:38] Which brings us to the man who isn't on the call anymore. The board announced Jensen Huang's resignation on the 12th of November. Cress stepped in on an interim basis, and Harvey Jones became an independent chair. Last night, Jones confirmed the board has no intention of reinstating Jensen Huang. A search firm has been appointed, and both internal and external candidates will be considered. Now I'm not going to revisit the personal allegations. What matters for us is the role that Jensen Huang played, and it was much bigger than running a company.

[6:13] He ran NVIDIA with something like 60 direct reports and very few layers, so that technical problems, customer needs and commercial decisions sat close together. But his reach went well beyond NVIDIA. Larry Ellison has told the story of a dinner where he and Elon Musk were, in effect, asking Jensen Huang for more GPUs, an account Reuters revisited recently. Reuters has also described Jensen Huang helping startups well beyond the investment check. Looking at their technology, making introductions, and sitting in on fundraising conversations. Put those together and you get a very different picture of the job. Jensen Huang acted as an informal clearinghouse for the AI build-out. When chips were scarce, NVIDIA decided who got them. When customers needed capital, NVIDIA's money and his endorsement helped them raise more from others. And when a market was unproven, his conviction was itself a signal that other investors followed.

[7:16] None of that appears on a balance sheet, and all of it lowered the cost and the time of getting things built. When a company loses that kind of figure, it usually replaces the person with process, allocation criteria, contract conditions, investment committees. That's exactly what Cress described last night. Process is more predictable, and arguably better governance. But it's slower, and it can't lend its personal credibility to a customer's next fundraising round. Cress closed the call with a simple line. We've got this. On operations, I believe her. It's the coordinating role that's harder to replace. Think of a Formula One team losing its team principal halfway through the season. The car is still fast. The drivers, engineers and strategists still know their jobs. And they make the detailed calls on tyres and pit stops anyway. What's gone is the person who aligned the organisation, secured the resources and handled the big relationships. And that matters most in the negotiations happening away from the track. So how has the market read all of this? I think the sequence is the most revealing part. Last week, Nvidia fell 13% while AMD rose about 19% and Broadcom about 18%. That was the market treating this as NVIDIA's problem. Sell the company that lost its founder, and buy its rivals.

[8:44] This week, through Tuesday's close, the story has changed. The chip funds fell together. The two main semiconductor funds lost between 5% and 5.5%, and the memory fund more than 8%, against about 3.5% for the NASDAQ 100%. Intel and Micron fell harder than Nvidia. Every one of the 11 major sector funds was lower, with staples and utilities losing least, and technology losing the most. In other words, within a week, the market moved from an NVIDIA-specific story to a question about the whole spending cycle. Memory falling hardest fits that reading, because memory is one of the parts of the chip industry most sensitive to how much gets spent. That's a price signal, rather than a measure of fund flows. But it lines up with the coordination argument. If the clearinghouse is weakened, the risk isn't really market share. It's how smoothly the whole build-out gets financed. Then, after the results, NVIDIA rose more than 4% in extended trading. Encouraging, but let's keep it in perspective.

[9:51] From its high of about $243 on the 6th of October to Tuesday's close around $155, the stock has fallen roughly 36%. In early November, it broke through the support level it had held through September and October. Around $209, and the selling since has come on heavy volume. The trend is down. Until the price starts making higher lows, a bounce after good results is a bounce inside a downtrend, not a change of direction. So let's work through the stack, layer by layer. The chips, the physical build-out, the financing, and the models and applications that sit on top.

[10:34] At each layer, I'm asking one question. Who gains bargaining power from here? And who loses it? Okay, that's two questions. Okay, we'll go with two questions. Let's start with the chips. CUDA, NVIDIA software platform, didn't resign. Neither did the installed base, the engineering teams or the manufacturing agreements. The big buyers already had custom chip programs and alternative suppliers in the works. And a leadership shock may give those plans more attention. But turning attention into deployed capacity takes years of software work, testing and procurement. A customer testing an alternative chip shows interest. A customer placing repeat orders for it, at scale, is a real shift. And that's what I'm watching for. Last night's call also pushed back on the idea that NVIDIA's pricing power is crumbling. Management said token prices have fallen nearly 60% since May, while GPU rental prices are rising.

[11:35] Older hopper systems have been recontracted at up to twice their original price, and rates for the newer B300 are up more than 60% since launch. The logic is that as systems squeeze more useful output from the same equipment, the price of that output can fall, while demand for the equipment stays strong. NVIDIA says its installed compute is fully used across every hyperscaler and NeoCloud it serves. Now that's management's view, rather than a market survey. But it does matter, and we'll come back to it at the top of the stack. The other chip names don't line up as simple winners from NVIDIA's loss. TSMC makes chips for everyone, so its exposure is to total volume rather than who wins. Micron supplies high bandwidth memory and has announced HBM4 production for Vera Rubin, so it gains alongside NVIDIA, not instead of it. And Intel is both a competitor and a partner, with NVIDIA's $5 billion investment agreement in the mix. That's why this week's selling hit all of them. They share the same spending cycle, more than they compete for NVIDIA's share. The call also made clear where the real constraint sits. Demand still outruns supply. NVIDIA has been locking in memory, substrates and advanced packaging, and it's prepaying suppliers to expand.

[12:59] The bottleneck is upstream, and that's where NVIDIA is now putting its money. So at the chip layer, I don't think bargaining power is moving sideways to rivals, at least not yet. I think it's moving down the stack. Because here's the thing, a megawatt doesn't care whose logo is on the chip, but a financing agreement might. Cress was very clear about how NVIDIA allocates systems, land, power and buildings ready to receive them, capital in place, and confidence in the customers who'll use that capacity. Read that list carefully, because it tells you where the scarcity now sits.

[13:37] Under Jensen Huang, relationships and conviction could move a customer up the queue. Under a process, the queue rewards readiness. Cress added a detail that's easy to miss. Everyone gets the same pricing. So the advantage isn't a better price. It's getting the equipment at all and getting it first. Imagine two operators advertising the same capacity. One has its financing sorted, a grid connection, and a customer ready to start paying. The other still needs permits, and another funding round. Same headline number, completely different position. The first can take equipment that the second can't receive. And readiness is worth a great deal. NVIDIA says a 1MW AI factory can generate revenue approaching $50bn a year. So reaching first output a single quarter earlier could be worth more than 10 billion. That's management's arithmetic, but it shows why time on the physical clock has become one of the most valuable things in the stack. Now that's the distinction from last week's episode, now with a sharper edge. Powered, permitted and funded sites get bargaining power. Projects that exist mainly as announcements lose it. The build-out can keep growing in total while individual projects are delayed or cancelled. So for utilities, cooling specialists and equipment suppliers, what matters is the funded order book and the quality of the customer, not the headline spending forecast.

[15:13] Which brings us to the layer where Jensen Huang's absence may matter most. Financing. Here's a simple, made-up example of circular financing. A chip supplier invests in a cloud operator. The operator adds borrowed money, buys a supplier's systems, and rents the capacity to an AI company. The supplier might also agree to buy some capacity back, or offer a limited guarantee. The same few companies end up as investor, supplier and customer in one connected chain. There can be sound reasons for that, but the test is where the cash ultimately comes from. Are end customers paying enough for useful AI to cover the infrastructure and its financing costs? Or does the chain need another funding round before the bills fall due? It also helps to keep the prices separate. An equity stake, a loan, a capacity purchase, and a guarantee are different commitments. And adding every headline figure together makes the web look bigger but harder to understand. And, of course, the sums involved are very large. NVIDIA reported total guarantees of $118 billion. And $105 billion of that is credit support for land, power, and buildings at a single campus, to be leased to OpenAI for 20 years. NVIDIA has also around $390 billion of purchase commitments to suppliers, and more than $100 billion in leases and cloud commitments.

[16:43] They're different kinds of obligation, but one thing stands out. Most of the guarantee exposure rests on one site and one customer. The contracts may not have changed, but something subtler has shifted. Under Jensen Huang, part of NVIDIA's support was reputational. Lenders and investors drew comfort from his involvement, on top of whatever the contract said. Now the support is only as strong as the contract, and contract conditions can bite late, because the customer may need to raise its own debt and equity, finish the facility and deploy the equipment before any backstop takes effect. That's where the two clocks collide. Interest arrives on a schedule. Customer revenue can arrive late. For a heavily borrowed operator, the danger isn't that AI demand disappears. It's a timing gap. That's how a financing problem can travel through the stack, even while usage keeps growing. So in our research, we trace each relationship separately. Who provides the money? Who owes it back? Who buys the equipment? And who promised to buy the output? And what happens if the start date slips?

[17:58] The aim isn't the biggest possible total. It's to see where the same risk turns up in several places at once. So at the top of the stack, those pricing numbers come back into play. If token prices are falling while the rent for the equipment is rising, someone is caught in between, and it's whoever rents, compute and sells tokens. Cheaper AI is great for users, but for the businesses selling it, lower prices only turn into profit if usage grows fast enough to outrun the falling price per token and the rising cost of capacity, or if the product is different enough to hold its price.

[18:35] Value tends to flow to whoever owns the bottleneck, and right now, that's still the equipment and the ready sites, not the token. So more tokens processed, more agents deployed, and more users signed up are encouraging, but they're not a substitute for paying customers retention and cash generation. And this layer matters to every other one, because in the end, it's the revenue here that has to pay for the chips, the power, and the debt underneath. Okay then, so here's a question that's been speculated about in great detail in the media. So who takes the wheel? It's tempting to assume an ex-CFO means caution, or that an internal appointment means nothing changes. I don't think either of those assumptions holds. The board really faces two separate choices.

[19:24] One is, how much of NVIDIA's existing way of operating to keep? The other is how much financial support to keep providing to the ecosystem. One path keeps both. The same organisation and the same appetite for financing. Now that preserves speed, but it also keeps the risks of a tightly connected customer base. The other path keeps the teams, but gets more selective with money. Protecting the roadmap while pushing weaker projects to find capital elsewhere. Or the board could reorganise, with or without the financing ambition. Each path has a different trade-off, and we'll learn which one we're getting from decisions, not from a CV.

[20:05] On Names, Reuters has reported analysts and investors discussing Kress and Ian Buck internally, and arms Rene Huss as an outside possibility. Now that's not a confirmed shortlist, but there's been plenty of speculation on the prediction markets, and interestingly, they've swung sharply from insiders to outsiders since the resignation. In the first hours, Buck and Kress were the favourites. This morning, Rene Huss leads. These markets are thin, and a single trader can move them, so I wouldn't read too much into the odds themselves. The direction is more telling, and it fits another clue. Alongside the results, the board has set up a new committee to oversee national security and export controls. NVIDIA has stopped shipping its licensed hopper products to China, and new export controls arrived this week. The chair said that the next chief executive will have to navigate national security dynamics very carefully. The job description has changed. The next leader may need to be as fluent with governments as with engineers. And the deeper question is whether the next leader can rebuild the coordinating role, or whether NVIDIA decides it shouldn't play that role at all. For now, last night's call points to continuity in operations and tighter scrutiny of commitments.

[21:31] So where does all of this leave the framework view? Well, my working view is that NVIDIA's physical clock is running strongly, and its financial clock has been disrupted. The earnings did strengthen the demand case. They didn't settle the leadership question. And they didn't remove the wider financing concerns the market moved on to this week. And I'm keeping those in separate boxes. Strong product demand can coexist with higher execution risk. A valuable asset can sit on a fragile financing structure, and a sound company can still be a poor trade if you don't get the timing right. My constructive scenario is that the roadmap keeps delivering. Customers receive working systems on time, key people stay in the company, and credible projects keep finding funding.

[22:21] The harder scenario is that governance uncertainty drags on, and weaker customers struggle to finance their deployments. That would slow the conversion of demand into revenue, without any collapse in enthusiasm for AI. A genuine demand problem would look different again. Customers cutting actual deployments, cancelling orders or reporting weaker economics from their AI services. As a trader, I also need the price action to cooperate. My preference remains mean reversion, within an established trend. And right now, NVIDIA's established trend is down. So a sharp rebound is evidence to weigh, but not a signal to chase. I want to see higher lows, and strength spreading beyond a handful of names. And I want the trade and the thesis on compatible timeframes. A multi-year opportunity doesn't tell me this week's sell-off is over. Of course it doesn't. And one bad week doesn't invalidate a multi-year opportunity.

[23:23] And finally, what matters next? There are four things that I'm watching. First, whether last night's response holds through the regular trading, and whether suppliers and infrastructure names join in. Second, customer execution. Sites ready, systems installed, and contracts turning into revenue. Third, the financing deals. Who puts in the money, whether outside lenders take part, and whether any existing projects are delayed or restructured. And fourth, the leadership appointment. How NVIDIA handles the new export controls, whether key people stay, and whether anyone steps into the coordinating role Jensen Huang played. Together, they'll tell us whether the investment case is improving across the whole stack, or only in one layer of it. So here's where I've landed. NVIDIA's physical clock is running fast, The demand is real, and the systems are shipping. What's changed is the financial clock. Support that once rested partly on Jensen Huang's personal backing now rests on contracts, conditions, and ranking of priorities. And also a queue that rewards whoever is ready. That's why the question was never simply whether the AI trade still works. It's which business, in which layer, financed on what terms, and at what price.

[24:50] For years, the AI build-out had one man at its centre. NVIDIA's next chapter will show us how much of that was the company, and how much was him. OK, before we wrap up, a quick update on the latest news. I am recording this part on Thursday morning, the 19th of November, because a lot has happened since the main episode, and three stories connect directly to what we've been talking about. First, NVIDIA. Earlier I said we'd need to see whether the after-hours bounce held. It didn't. NVIDIA closed almost 3% lower on Wednesday, and Reuters now puts the fall from its October record close at around 36%. But look at what the analysts did. Their average estimate for NVIDIA's earnings next year went up, while their average price target came down. In other words, they think the business is stronger than they did on Monday, but they're willing to pay less for it. That's the two clocks in a single data point. The physical clock speeding up, and the financial clock demanding a bigger discount for uncertainty. As one strategist at Saxo put it, strong earnings have bought management time.

[26:06] A few other items. The board has confirmed it won't reinstate Jensen Huang, despite, according to Reuters, an offer from him to return. And the Wall Street Journal has reported that the US Securities and Exchange Commission is investigating the company. There are no details yet, so I won't speculate on that. But it adds weight to the governance question. And on China, one investor summed up the export control issue very neatly. Demand is not the same as the legal ability to ship.

[26:38] Second, the big one, Anthropik's IPO. Anthropik, the company behind Claude, has raised $81 billion in the largest stock market debut on record. Add a further $26 billion from a private placement at the same price, and the total deal comes to just over $107 billion. It priced above its range, increased the size of the deal, and is valued at around $1.5 trillion on a fully diluted basis. As I record, it hasn't yet started trading. It debuts on the Nasdaq today under the ticker ANT.

[27:15] And the timing is remarkable. The day after an episode asking who pays for the gap between cheap AI and expensive infrastructure, part of the answer turns up. Public markets. Investors have just written one of the largest checks in history to the model layer. And the filing reads like a case study on this episode. Anthropik's annualised revenue reached $81 billion at the end of October, and third quarter revenue grew 47% on the previous quarter. But its computing costs are rising faster, which cut the revenue returns for every dollar of compute, and it swung to an operating loss. That's the squeeze at the top of the stack. It has also committed to nearly $900 billion of computing spending with partners including Amazon and Google. And in its risk factors, it warns that those partners depend on outside financing and credit support. If that support becomes unavailable or more expensive, it says contracted capacity may be delayed, repriced, reduced or not delivered. Now that's the financing layer, described in the company's own words. The money raised this week makes Anthropic itself much stronger. The question is whether that strength reaches down the chain to the operators building its capacity.

[28:35] Third, cyber security. Palo Alto Networks raised its annual forecast and agreed to buy Applied Compute for $4.5 billion to help customers train and run open models securely. Applied Compute was one of the partners NVIDIA named on its call, so you can see security becoming part of the AI stack itself. Now that news lands alongside a run of technology outages in the US over the past week, touching email, air travel, and even the National Security Agency. Officials say the incidents are unrelated and there's no evidence of an attack. And I'd be careful about joining dots that may not be there. But one thread is worth watching. The US Director of National Intelligence linked the NSA outage to patches for a vulnerability that was discovered using AI. As one analyst put it, AI tools are servicing vulnerabilities faster than companies can patch them. AI is speeding up both sides of security, finding the weaknesses and forcing the fixes. And that's a topic that we'll come back to in a future episode. Finally, Nasdaq futures are up nearly 2% this morning, as the market digests all of this. Next week, we'll dig into what Anthropik's numbers tell us about the economics at the top of the stack.

[29:55] Okay, so that's a wrap for this week. If you missed last week's episode, AI's Battle for Power, it sets up the distinction between operating capacity and the projects still waiting to be built. You'll find it, along with our wider research, at qfmarkets.substack.com. 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. And if you've got a question for a future episode, you can email me at podcast at quantumfields.ai.

[30:32] Just before we close, a quick disclaimer. Everything discussed here is for information and education only. It isn't investment advice or a recommendation to buy or sell any financial instrument. Thank you for listening to the AI Supercycle podcast. See you next week.