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] Last week, I asked a simple question. Who can afford to wait?
Who Can Afford to Wait?
[0:05] This week brought some answers. Meta's investment spending overtook the cash generated by its business for the quarter. It then turned to shareholders for fresh funding. Elon Musk said lenders were becoming much less willing to finance companies whose revenue depends on AI data centres. And Anthropic published the prospectus for its planned stock market listing. Then, overnight, memory chip maker SK Hynix reported quarterly operating profit more than seven times its level a year earlier. It says, signed deals and discussions with customers would effectively fill its memory capacity through 2029. So demand for AI infrastructure still looks strong. But strong demand does not guarantee that every project can raise the money it needs, or earn enough, to justify the investment. Who can fund the next stage, and who stands to benefit.
[1:11] 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 28th of October, 2026. The market picture includes Tuesday's US close and this morning's Asian session. The Fed announces its decision at 6 this evening, UK time. Alphabet and Microsoft report after the US close, followed by Amazon and Intel on Thursday, and Exxon and Chevron on Friday.
[2:02] Most of those will be out by the time you hear this, so I'll focus on the questions they need to answer. Last week, we looked at why strong demand and a nervous stock market can exist side by side. A data centre may take years to build, while interest payments and investor expectations have to be met along the way.
[2:22] This week, we'll start with the borrowing costs and the Fed. Then we'll look at the company's buying and supplying computing capacity, before turning to natural resources and geopolitics. I'll finish with my market view and what I'm watching next. And just a quick heads up on two things coming at the end of today's show. I'm recording this before the Fed decision later today and the key earnings after the close. But given how much they matter this week, I'll add a latest news update just before I publish. I'll also give you a preview of something I'm working on. A conversation with Prism, our AI market intelligence agent. I'll be challenging its analysis and giving it room to challenge mine. So more on that later.
Borrowing Costs and the Fed
[3:08] Okay, so let's move on to the first part, where we discuss the macro regime. Government borrowing costs have eased from earlier this month, but investors are demanding more compensation for lending to riskier companies. The US 10-year government bond yield fell to just under 5% last Thursday, and has been around 5% since. That is some relief for businesses whose expected profits lie many years ahead. Meanwhile, the extra yield investors demand on riskier corporate bonds has widened to around 3.5% above comparable government debt. Last week I used the example of your credit card. The underlying cost of money can fall, but your lender may still keep your rate high because of the risk of lending to you. We're seeing that tension in corporate debt. Government yields have eased, while the extra charge for credit risk has increased. For a company trying to finance a new data centre, lower government yields alone may not bring much relief.
[4:10] Now inflation is part of the problem. US consumer prices were 3.7% higher in September than a year earlier. Excluding food and energy, annual inflation was 2.6% up from August. Energy was the clearest source of the increase, but electricity and computer equipment prices were also higher. Does that mean AI is feeding inflation? AI investment may be adding pressure to construction, equipment and electricity demand, But these figures measure consumer prices, not the cost of building and running data centres, so they cannot tell us how much. My concern is that the AI investment boom is competing for resources while the economy is also absorbing higher energy costs. That could make inflation harder to bring down and leave borrowers facing high interest rates for longer. The Fed raised its policy rate in September to a range of 3.75 to 4%. Markets strongly expect it to hold that range today, while favouring higher rates by December. Officials have been in their pre-meeting blackout, so their recent silence tells us little about whether they agree with the market. Chair Kevin Walsh has been sceptical of giving detailed promises about future policy.
[5:30] My reading is that he will leave room to respond to the evidence rather than commit to a December decision. We'll know more after the press conference. And more evidence arrives almost immediately. Thursday brings the first estimate of third quarter US growth and the Fed's preferred inflation measure. Friday brings the employment cost index which tracks wages and benefits. If the Fed leaves its next move open, those releases could move interest rate expectations sharply. That matters for companies arranging large amounts of funding. This is also the last Fed decision before next Tuesday's US midterm elections. Elsewhere, the Bank of Canada decides today, the European Central Bank on Thursday, and the Bank of Japan on Friday. Here in the UK, today brings the budget.
Funding the AI Build-Out
[6:21] Okay, so let's move on to part two and talk about computing capacity and capital. As you know, there is no single AI trade, and I've been talking about that for many months now. A chip supplier, a cloud operator, and a model developer can face very different business conditions. This week makes that particularly clear. Let's start with SK Hynix. Its third quarter operating profit rose to about 83 trillion won, more than seven times the level a year earlier. Revenue was more than four times its level a year ago. The outlook matters even more. Hynix says demand for high bandwidth memory, the fast memory used alongside AI processors, is roughly twice what it can supply.
[7:08] The company describes its memory capacity as effectively sold out through 2029. That includes both signed deals and business still under discussion. More than half of its output for 2027 and 2028 is already covered by long-term agreements. Now that is really strong evidence of demand, though not a guarantee of every future sale. Hynix expects memory shortages to persist through 2030 and plans to increase investment significantly next year, Its shares rose about 11% today helping South Korea's stock market to a strong recovery, There is still competition to watch though Reuters reports that China's CXMT, plans to begin limited production of advanced AI memory within the next two months though only a small proportion of its chips currently meet the required standard.
[8:04] That is a longer-term challenge, rather than an immediate end to the shortage. ASML, which supplies the machines used to manufacture chips, reported a similar picture earlier this month. It said its planned expansion still would not fully meet customers' expected demand in 2027 and 2028. And some customers are already discussing orders for 2029 and 2030. The suppliers are giving us evidence of a substantial pipeline of spending. The next question is how their customers pay for it. And as we've seen in the news this week, Meta are raising more money. Meta's revenue rose 26% in the third quarter. Its core business is still generating substantial cash. But it generated about $33 billion from operations and spent about $35 billion on investment, including principal payments on equipment leases. That left free cash flow negative by just over $2 billion.
[9:07] Now this needs perspective. Free cash flow was still positive over the first nine months, and Meta ended September with more than $82 billion in cash and marketable securities. It has not run out of money. The issue is the scale and pace of its future spending. Meta has now priced $37.5 billion of equity financing. That includes ordinary shares sold through arrangements that allow Meta to receive the cash later, and securities that pay a dividend and eventually convert into shares. Separately, Meta plans a share sale programme next year that could raise up to another $60 billion.
[9:47] Now that programme has not yet been established, so this is not all cash arriving today. The funding spreads the load beyond the cash generated by the business. It also means existing shareholders face more shares being issued. Meta shares have fallen sharply and remain below the short-term trend measure that I follow. The financing adds a dilution concern to the debate about whether its spending will earn an adequate return.
[10:15] Now there is another connection worth watching here. NVIDIA has announced a $3 billion investment in Meta's offering, and AMD has announced $1 billion. Broadcom has also expressed an interest. Suppliers investing in a customer does not make that customer's orders artificial, but it links the funding of the buyer more closely to the company's benefiting from its spending. That deserves scrutiny alongside the order numbers. And we can also see that debt is becoming more selective. Over the weekend, Elon Musk said SpaceX had closed $40 billion of debt financing for AI processors due to arrive in the first half of next year. He credited its investment-grade status and revenue from Starlink and the launch business. Reuters said it was not clear whether the debt had been priced. So I would treat the details as Musk's account. He also said debt markets had almost entirely shut off for companies without substantial revenue, independent of AI data centres. Now that is a broad claim from someone who has a competitive interest in the outcome. But there is evidence of pressure. The market yield on one of CoreWeave's existing bonds reached around 15% earlier this month. CoreWeave is one of the specialist providers that rent out AI computing capacity, often called neoclouds.
[11:38] That yield does not mean all of its debt suddenly costs 15%. It does show investors demanding a much larger return to hold that bond. My reading is that lenders are becoming more selective. Strong cash generation and dependable customers matters a lot more when investors start questioning how the next stage of construction will be financed.
[12:01] Anthropic published its prospectus on Monday for a proposed NASDAQ listing under the ticker A&T. Its first half revenue was about $16 billion. Its headline net loss was around $81 billion. Now that loss needs explaining. It was dominated by accounting adjustments linked to financing arrangements. It was not $81 billion of cash leaving the business. The operating loss was much smaller, at just under $700 million, and cash flow from operations was positive at almost $4 billion, helped by the timing of payments to suppliers and receipts from customers. For the wider AI build, its future commitments matter. Anthropic has agreed to pay for substantial computing and infrastructure capacity, over many years, and some contracts require payment whether or not it uses all of the capacity. The prospectus leaves the broad aggregate commitment figure blank, so I'm not going to give a single total here. The question is whether revenue growth, existing funds and future financing will cover those obligations. A successful listing could strengthen Anthropik's funding position. It would not, on its own, show that every infrastructure contract will earn a good return.
[13:20] So what does the comparison tell us? Hynix is also planning substantial returns to its shareholders, around $75 billion for the year. But that includes earlier buybacks and a special dividend linked to gains on its Kioxia investment. It's not all new money from this quarter's chip sales. So the contrast with Meta is useful, but it needs a bit of care. The evidence shows strong business conditions for a key memory supplier, alongside enormous funding requirements among buyers of computing capacity.
[13:54] Financing may first change who can build as a weaker operator loses a project to a stronger one. But that transfer takes time, and delays can reduce how much capacity actually gets delivered. So I'm watching for projects failing to secure funding, construction being postponed, orders being reduced, and customers struggling to meet their commitments. None of that adds up to a broad contraction investment just yet. However, it does give us reason to look beyond headline orders. So what are the charts saying? Well, the semiconductor shares are recovering, but most of the markets on my watch list remain below their 21-day moving average. Hynix is one of the exceptions. So this average gives more weight to the recent prices and helps me judge the short-term trend. It does not tell me why a share price is moving, and crossing it does not guarantee a lasting recovery. For now, the business news looks stronger than the share price trends. I want to see more companies joining the recovery, before treating this as a resumption of the uptrend.
Energy, Copper, and Constraints
[15:02] So now let's turn our attention to part 3 and natural resources. Now I appreciate that natural resources might sound like a change of subject, but it isn't. The AI build is a physical build-out and it runs on energy and materials. The link works in two ways. Directly, data centres need electricity. Natural gas is a major source of that power in the US, and the grid connections behind them depend on copper. Indirectly, energy has been the main driver of the latest rise in inflation, as we heard earlier. That keeps borrowing costs higher for the companies financing the build. So for each market, I'm asking two questions. What it tells us about the physical constraints, and what it means for the cost of money.
[15:52] Brent made highs of over $117 per barrel last Friday. However, it has pulled back this week from those highs, but remains over $111 as I record this. Oil for December delivery was trading around $5 above January delivery last week. Now that premium for earlier delivery points to concern about near-term availability. US commercial crude stocks also fell by about 3 million barrels in the latest weekly report. Refining margins, however, have softened on the measures I follow. A strong crude price does not automatically mean improving margins for the companies turning oil into fuel. Natural gas presents a different picture on both sides of the Atlantic. US gas storage is slightly above its five-year average, and US gas prices have been weak. European gas prices are a different story, and are much higher than earlier in the year. So not every energy market is suffering the same shortage. And plenty of gas at a national level does not solve the local problem of getting electricity to a particular data centre. Copper is holding near its recent highs, although shares in copper miners have been much weaker. Tech's Highland Valley operation is running at a reduced capacity following strike action, which can support copper prices while hurting the company losing production.
[17:15] China's economy grew 4.5% from a year earlier in the third quarter, but that headline does not by itself establish a recovery in construction demand, which matters for copper. The wider point is that a commodity and the shares of its producers can behave differently. Costs, production problems and financing all affect what shareholders actually receive. For the AI build, the message is this. Oil feeds into inflation and borrowing costs. While gas and copper bear on where the power and grid connections arrive on time. And those prices can move quickly when geopolitics gets involved.
Geopolitics and Supply Risks
[17:55] OK, let's move on and talk about geopolitics. Because the risk of disruption in the Gulf remains an important part of the oil story. The Houthis claimed a strike on Saudi Arabia's Ras Tanura refinery on Tuesday. Saudi authorities had not confirmed it in the reporting available this morning. A claim of an attack, confirmed damage and a measurable loss of oil supply are three different things. We need to keep them separate. Diplomacy could ease the market's fears before physical conditions change, but to judge whether the supply risk is improving, I'm watching production, exports and shipping. A confirmed loss of output or exports would add upward pressure to prices. Evidence of stable supplies and reduced tensions could take some of the premium away. Now lower oil prices would help ease inflation, but they could also reduce producer revenues. Which is relevant as Exxon and Chevron report on Friday.
[19:00] In China, the Communist Party's fifth plenum is taking place this week, with party governance its announced focus. For the AI supply chain, the more immediate question is Rare Earth's export controls. China's suspension of its additional measures runs to the 10th of November, less than two weeks away. We should not assume the broader trade truce, which extends into January, settles what happens to those measures.
A Cautious Market View
[19:27] Okay, so let's move on to part five and take a look at the framework view. As you know, PRISM is the research framework that I use to bring together the economic backdrop, company evidence and market behaviour. Now my overall stance remains cautious. Borrowing costs are high, investors are asking for more compensation for credit risk, and many infrastructure shares have not yet established a recovery. Now that is not a forecast of a collapse. It means I'm looking for stronger evidence before becoming more positive across the group. Memory is an important distinction here. The business outlook remains strong, but I can be positive about demand for memory chips and still be patient about buying the shares. The business outlook and the timing of a trade are different questions.
[20:16] Now oil presents a different challenge. The framework's near-term reading is more cautious than the price trend, while its monthly view remains positive. It does not yet fully capture geopolitical supply risk, so I need to assess that separately. In uranium, I'm keeping the metal and the mining shares separate. Their price signals are different and I don't want to treat one as a substitute for the other. Across these markets, disagreement between business conditions and share prices deserve attention because it can reflect financing concerns, valuation or other risks. I don't think the chart alone can tell us which. So my risk management point is simple. Owning several AI-related shares does not necessarily give me several independent investments. Many of them are highly correlated. Chips, data centres, networking and the largest cloud companies share exposure to the same spending cycle. And I need to position size for that shared risk. So I'm watching four things. First, borrowing conditions after the Fed decision. Do expectations for higher rates by December hold up? and does the extra yield demanded on riskier corporate debt stabilise or keep rising?
[21:31] If credit conditions improve, some heavily sold shares could benefit, though businesses with weaker finances may continue to struggle. Second, the largest cloud companies' cash generation and spending. How much cash are they bringing in? How much are they spending on infrastructure and equipment leases? And is the capacity arriving on schedule? One quarter of negative free cash flow can reflect timing. A persistent shortfall, combined with weaker demand or more expensive funding, would be more concerning. Third, the breadth of the recovery in semiconductor shares. I'd like to see more companies and the main sector funds moving back above their short-term averages and holding those gains. That would give me more confidence in the trend, though it would not prove the financing risks have disappeared. Fourth, actual energy supply. Confirmed changes in gulf production, exports and shipping will help us judge whether the oil market's concerns are being realised. Beyond those immediate events, Anthropik's proposed listing will test investors' appetite for funding AI model developers. Now the price matters, but so does how much fresh capital the company itself receives.
[22:50] So what would change my view? I would become more positive if companies secure funding on acceptable terms, credit conditions stabilise and the recovery in infrastructure shares broadens. I'd become more cautious if funding pressure leads to delayed projects, cancelled orders or customers struggling to pay. And a sustained reduction in computing purchases by a major cloud company would make me revisit the outlook for its suppliers. So, is the AI build running short of money, or changing where the money comes from? This week's evidence points to funding becoming more selective, while demand for capacity remains strong. Meta's offering shows equity taking a larger role in its funding mix. SK's Hynix results show the strength of its position in memory. And Anthropik's prospectus gives us a clearer view of both rapid growth and substantial future obligations. The question is whether those buying the capacity can keep funding it and ultimately earn enough from using it. Last week I asked who can afford to wait. Companies with cash generating businesses, strong balance sheets or dependable access to new funding have more room to do so. Those relying on the next financing round have less.
The Fed and Earnings Update
[24:10] As mentioned at the start, I recorded the main part of this episode before the Fed decision and last night's earnings. So here's the latest news, and what it means for the story we've been following. The Fed held rates at 3.75 to 4%, as expected. But two officials, Governor Lisa Cook and Cleveland Fed President Beth Hammock, voted for a hike, after a unanimous vote in September. In its statement, the Fed said the economy is growing strongly, but that inflation remains elevated. Walsh didn't commit to December, but he left the door open to further increases, if inflation doesn't come down. He was also clear that politics played no part, with the midterm elections next week. Markets were quite volatile during the press conference, and futures markets still see a December hike as the most likely outcome. So we didn't learn much about the Fed's next move from Walsh's comments. What we did learn is that the committee is more divided. AI also came up more than once in the questions, with reporters asking whether the build-out is driving inflation. The pressures are the ones we've discussed earlier. Energy, tariffs, strong consumer spending, and demand from the AI build itself. Now that loop between AI spending, inflation, and the cost of money is now part of the Fed conversation. So the data, starting with Thursday's inflation figures, matters more than anything the Fed says.
[25:38] In oil, US crude stocks fell again last week by almost 2 million barrels, when no change was expected. Distillate stocks, mainly diesel, also fell by more than expected, while gasoline stocks rose. So the tightness is in crude and diesel, which fits the strength in oil prices.
Which Companies Can Fund Growth?
[25:57] Then the results, and they answer this episode's question almost directly. Microsoft showed growth and cash together, Azure grew 48% and demand still exceeds the capacity it can bring online. It's spent about $45 billion on capex, including leases, yet free cash flow stayed positive at around $18.5 billion. Some third-party capacity has arrived late, pushing spending into the next year rather than cutting it. Share prices rose after hours. Alphabet showed the other side. Google Cloud more than doubled, helped by sales of its own AI chips. But CapEx also more than doubled, to over $50 billion for the quarter. However, free cash flow was negative by about $9 billion. Advertising growth disappointed, and it expects to spend significantly more next year. The share price fell around 10% after hours.
[26:55] And also overnight, Samsung added to the memory story. Profit in its chip division rose 15-fold, to around 106 trillion won, and it says the AI chip shortage is getting worse. It plans to raise chip investment by more than 60% this year, with another sharp rise next year, but the shares fell about 6%. Investors had hoped for a shareholder return plan to match SK Hynix, and it didn't get one. So even among the suppliers, the market wants to see the cash coming back, not just more of it going out. So this doesn't change the argument, it actually sharpens it. Demand isn't the question. The market is rewarding companies whose cash keeps pace with their spending, and marking down those whose spending runs ahead of it. For this quarter, Alphabet joins Meta in that second group, and Nasdaq Futures, which briefly pushed above 30,000 around the results, gave that back overnight and are lower again this morning, Still below their short-term average. The charts haven't confirmed the recovery just yet.
[28:02] The question from this episode still stands. Can the buyers of capacity keep funding it? Microsoft says yes, for now. Alphabet says not without a harder test from shareholders. Amazon and Intel report after Thursday's close. And that's where we'll pick up next week.
Putting PRISM to the Test
[28:22] And before we finish, as I mentioned at the start, I'm working on something a little different for the podcast. A conversation with Prism, our AI market intelligence agent. PRISM is the probabilistic model we use for market analysis and trading here at QF Markets, and we've given it a voice so it can talk through its analysis, and answer back when I challenge it. The voice is created with 11 labs, but the analysis is PRISM's own, and I intend to put it under some pressure. If PRISM makes a case for the AI supercycle, I want to know what could break that case. If the evidence looks compelling, I want to know what we might be overlooking. And if one of my questions rests on a flawed assumption, Prism will have room to challenge me too. The first question I have in mind is this. What would actually prove the AI supercycle thesis is wrong? Now that should give us plenty to discuss, and I'll share the first part in a coming episode. Okay, so you can find more details about QF markets on Substack, where the weekly outlook, market pulse and in the spotlight articles are free. The subscription research is moving to the QF Markets website at quantumfields.ai which we're aiming to launch in November. So again, I'll keep you posted on that as we get closer.
[29:43] 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.