The AI Supercycle

Episode 11: The Price of Building AI
Why Rising Borrowing Costs Are Splitting the AI Build-Out in Two

Demand for AI looks strong. The cost of building the power, computing capacity and infrastructure to meet it is rising fast. Last week those two stories began to pull apart, and the market started paying for AI's story while selling the plumbing that makes it possible.

In the first episode of the new weekly format, Tim Hardwick works through five parts:

1. The macro regime. Long-term borrowing costs rose sharply as the US ten-year Treasury yield moved to post-crisis highs. The episode looks at why: strong business surveys, heavy debt supply, and energy costs driven by the US-Iran war and record diesel prices. It also covers what interest-rate futures now expect from the Federal Reserve in October.
2. The AI Supercycle. From Meta's Muse agent at the customer end, through Anthropic's cloud commitment to Akamai and a record wave of hyperscaler borrowing, to Oracle's force majeure notice on the Jupiter data centre campus in New Mexico. Power delivery has become a contract risk, local opposition is holding up projects, and uranium, copper and Starship's first orbital flight all show the gap between strong fundamentals and cautious prices.
3. Natural resources and geopolitics. Brent and European gas in a short-term downtrend now under test after Iran's Hormuz offer was rejected, and the difference between crude supply and the fuel that actually reaches customers. Also covered: the two-month rare earths truce and what it does and does not settle.
4. The PRISM view. Tim's base case in plain language: the cost of capital, not demand for compute, is setting the tone, and so far it has changed nothing structural. The confidence, triggers and evidence that would change his mind.
5. What matters next. Four tests to watch: the rate path, memory demand, data centre delivery, and whether supply and safety agreements become working processes.

The episode closes with late news: Anthropic's IPO prospectus, which reveals the company's growth, costs and more than half a trillion dollars in infrastructure commitments, and which warns investors that advanced AI could pose existential risks to humanity. OpenAI, meanwhile, is seeking to raise money privately instead. Episode 12 will go through the filing in detail and return to AI safety.

Chapters

 0:04   Macro Costs and AI Risk
 1:58    Rising Yields Hit AI Builds
 6:12    Consumer Agents and Cloud Deals
 10:11   Power Delays and Data Centres
 16:14  Energy, Rare Earths, and Geopolitics
 19:56  Framework for the AI Cycle
 24:31  Anthropic Filing and Safety Warnings
 28:28 OpenAI Raises More Capital 

Read more

Articles, reports and the free Market Pulse and In the Spotlight reports: qfmi.substack.com

Subscriber research, including this week's Weekly Outlook, The Two-Speed Market: qfmi.substack.com
 

Tags: AI, AI infrastructure, data centres, power, grid, bond yields, interest rates, Federal Reserve, oil, diesel, natural gas, rare earths, uranium, copper, Oracle, Blue Owl, Anthropic, OpenAI, IPO, AI safety, SpaceX, Starship, capital allocation, macro

For information and education only. Not investment advice or a recommendation to buy or sell anything.

• (00:04) - Macro Costs and AI Risk

Episode 10: Who Checks the Machines?

Why Proving AI Is Safe May Be the Next Bottleneck

AI’s leaders asked the industry to slow down. Markets wobbled, then shrugged. But the real story may not be whether frontier AI slows—it may be what it costs to prove that it has.

In Episode 10, Tim Hardwick examines Anthropic chief executive Dario Amodei’s call to pace frontier AI, the three-step assurance framework behind it, and the divided response from the major laboratories, Europe, Washington and Beijing.

The episode adds a sixth potential constraint to the AI Supercycle: assurance, the work of proving that models are safe and reliable enough to deploy. Model capability is advancing in months, while the supply of qualified evaluators grows far more slowly. If different jurisdictions impose incompatible requirements, the same thin pool of specialists may have to evaluate the same models repeatedly.

Tim separates catastrophic tail risk from observable operational risk, follows the impact through depreciation, time to revenue and return on invested capital, and explains why stronger assurance could unlock adoption in banking and insurance.

The QF Markets base case is that pacing fragments rather than coordinates: higher costs, slower deployment, little near-term reduction in physical AI infrastructure spending, and a possible bridge between frontier capability and enterprise revenue. Tim closes with three conditions that would change that assessment and five signals worth watching.

In This Episode

• What Anthropic’s proposed pacing framework actually involves
• What the semiconductor sell-off did, and did not reprice
• Why assurance could become the sixth constraint on the AI build-out
• The limited supply of qualified frontier-model evaluators
• The challenge presented by open-weight models
• Competing assurance regimes in Europe, the United States and China
• Why catastrophic tail risk and operational AI risk require different treatment
• How the assurance clock affects depreciation, revenue and returns
• Why stronger assurance could unlock deployment in banks and insurers
• How future frontier-lab IPOs could make assurance costs more visible
• The QF Markets base case, invalidation conditions and five signals to watch

Chapters

1. AI Slowdown Warning
2. The Podcast Returns
3. What “Pace the Frontier” Proposes
4. What Markets Actually Repriced
5. Assurance: The Sixth Constraint
6. The Strongest Objections
7. Europe, America and China Diverge
8. Tail Risk Versus Operational Risk
9. Four Clocks, One Balance Sheet
10. How Assurance Could Unlock Enterprise AI
11. IPOs Could Make the Constraint Visible
12. Base Case: Fragmented Assurance Regimes
13. Five Signals to Watch

Also in This Episode

Tim provides an update on his latest book: Trading the AI Supercycle - A Cross-Asset Framework for Big Tech Infrastructure, Energy and Critical Materials.

QF Markets

Quantum Fields AI is a decision-intelligence company focused on the strategic, economic and investment implications of the AI Industrial Economy. We connect both sides of the AI investment equation: how enterprises convert AI expenditure into measurable business value, and how investors and capital allocators understand the capital, infrastructure, constraints and returns shaping the wider AI economy.
Our work brings together enterprise AI strategy and economics, independent capital intelligence and domain-specific applications of decision intelligence.

Read the free Market Pulse and In the Spotlight reports, and explore the subscriber research, at:

https://qfmi.substack.com

Tags: AI, semiconductors, HBM, memory, energy, nuclear, uranium, copper, rare earths, data centres, photonics, capital allocation, macro, AI infrastructure

• (00:05) - AI Slowdown Warning

Episode 9: Physical AI: The Race for Embodied Intelligence

The AI Supercycle has been built in data centres and financial models so far. In Episode 9, Tim Hardwick moves the thesis into the physical world: robots, factories, materials, and the question of whether artificial intelligence can actually get a machine to do useful work, reliably, in the real world.

The episode opens by benchmarking five humanoid platforms against a single standard, not how good the demonstration looks, but whether the robot can complete the same task safely, repeatedly, and at a cost that earns a return. Tesla Optimus, Boston Dynamics' Electric Atlas, Figure 03, Agility Robotics' Digit, and AgiBot A2 each represent a different route into embodied intelligence, from vertical integration to mechanical heritage to state-backed industrial scale.

Inflated headline figures are separated from the real numbers: robotics venture funding is measured in the tens of billions, not the hundreds, and NVIDIA's fifty-trillion-dollar framing describes the addressable economy, not addressable revenue. A real industry disagreement, between claims of a "ChatGPT moment" for robotics and the blunter reality that lab performance regularly halves in real-world deployment, sets up the sector's binding constraints: dexterity, power, industrialisation, safety, and rare-earth materials.

The second half works through the CFO and COO questions that will actually decide enterprise adoption, the entire physical AI value chain from magnets to orchestration software, and physical AI's emerging role beyond Earth, in orbital maintenance and lunar infrastructure. The episode closes with a three-horizon framework for investors and the QF-MI base case: not a flood of humanoids into every factory and warehouse, but a slower, more uneven build, with the number to watch being the gap between company-reported production and independently verifiable fleet utilisation. This is also the final episode in the current run of solo episodes, with Tim taking a break for the summer before inviting guests on the show to discuss the AI Supercycle.

All reports are published at qfmi.substack.com

The Market Pulse and In the Spotlight articles are free, and always will be. The Weekly Outlook, the Weekend Debrief, and the Monthly Strategic Research Report sit behind a paid subscription. Subscribe and you get the full picture.

Chapters

0:10 Introduction

0:40 Business Update: PRISM and the Book

3:29 Introducing Physical AI

5:25 From Artificial Intelligence to Physical Intelligence

8:32 Why Humanoid Robots?

10:02 Tesla Optimus

12:30 Boston Dynamics Electric Atlas

15:30 Figure 03

18:40 Agility Robotics Digit

20:43 AgiBot A2 and the Chinese Ecosystem

23:00 The Existing Robotics Economy

24:45 Following the Money

27:00 The Physical AI Stack

28:55 The CFO and COO Test

31:52 The Binding Constraints

37:17 From the Factory to Orbit

38:50 What Should Investors Monitor?

43:20 The QF-MI Base Case

46:10 Conclusion

48:55 Close and Forward Look

Tags:  AI supercycle, physical AI, embodied intelligence, humanoid robots, Tesla Optimus, Boston Dynamics, Figure AI, Agility Robotics, AgiBot, vision-language-action models, industrial robotics, robotics investment, NVIDIA, rare earth magnets, robotics-as-a-service, orchestration layer, delivery gap, China robotics, space robotics, lunar robotics, enterprise adoption, capital formation 

• (00:10) - Introduction

The Enterprise AI Payoff: From Tokenmaxxing to Value per Token

The AI infrastructure build-out only matters if enterprises can turn compute into durable economic value. In Episode 8, Tim Hardwick moves from the supply-side story of GPUs, data centres and power to the harder demand-side question: is enterprise AI spending actually paying off.

The episode opens by drawing a sharp line between activity and value, tokens generated, users provisioned and hours saved don't count until they reach the P&L. Strong hyperscaler results from Microsoft, Alphabet and Amazon confirm enterprise demand for AI capacity is real, but are shown to be evidence of commitment, not proof of return.

Conflicting survey findings from PwC, McKinsey, Deloitte, Google Cloud and EY are reconciled: the disagreement itself reveals how immature enterprise AI measurement still is, and a concentration effect (20% of companies capturing 74% of the value) suggests returns are polarising rather than spreading evenly.

The second half sets out a practical framework: what makes a credible AI business case, a three-level scorecard connecting technical, operational and financial measurement, and the shift from tokenmaxxing toward disciplined token economics, selecting the right model, controlling architecture, and measuring cost per successful outcome. The episode closes with the dashboard of signals worth tracking, the case for and against the current build-out, and the QF-MI base case: not a spending collapse, but a shift toward selective scaling under real financial discipline.

All reports are published at qfmi.substack.com

The Market Pulse and In the Spotlight articles are free, and always will be.

The Weekly Outlook, the Weekend Debrief, and the Monthly Strategic Research Report sit behind a paid subscription. Subscribe and you get the full picture.

Chapters

0:19 AI Value Chain Begins
3:02 From Compute to Revenue
7:57 ROI Surveys Diverge
12:25 Capturing Real AI Value
19:10 Measuring Across Three Levels
22:22 Token Economics Shift
26:03 Optimizing for Outcomes
29:29 Efficiency and Demand Rebound
32:03 Tracking the Key Signals
34:45 Optimistic Case, Rising Demand
36:23 Selective Scaling Ahead
39:47 Closing Thoughts on the Cycle

Tags:
AI supercycle, enterprise AI, AI ROI, token optimisation, tokenmaxxing, token economics, value per token, FinOps, AI FinOps, Microsoft Copilot, Azure, AWS, Google Cloud, hyperscaler capex, agentic AI, model routing, inference cost, enterprise adoption, business case, benefit realisation, unit economics

• (00:19) - AI Value Chain Begins

Episode 7:  The Nervous System of the AI Supercycle

Capital Formation and the Race to Fund an $805 Billion Build-Out

For six episodes, this show has tracked the physical stack of the AI supercycle. The chips. The power. The materials. Most recently, the photonics connecting it all, and the possibility of taking infrastructure into orbit. But before any of that gets built, somebody has to raise the money.

In this episode, we turn to capital formation: not a new layer in the stack, but the nervous system running through every layer already covered. Hyperscaler capex is now guided toward roughly $805 billion in 2026, climbing toward $1.1 trillion in 2027, and the way that spending gets financed has shifted fast, from internally funded cash flow to a credit market that is starting to ask harder questions.

We trace that shift through three stages, place it against the closest historical parallel (the year-2000 telecoms fibre boom), and unpack the parts of this build-out that don't show up cleanly on any balance sheet: special purpose vehicles, private credit exposure, and this week's live example of circular financing involving Nvidia, SK Group and OpenAI.

Finally, we present the QF-MI base case, and what a more selective, more expensive capital market could mean for the pace of the AI build-out over the next 12 to 18 months.

In This Episode

• Why capital formation sits above the physical stack as the constraint that funds all the others
• The scale of hyperscaler capex, and what Alphabet's latest earnings reveal about the pace of spending
• Comparing today's build-out to the year-2000 telecoms fibre boom
• The three stages of AI financing: internal cash, external credit, and capital crowding
• What a falling bond coverage ratio actually signals, and why it moves before spreads do
• The rise of off-balance-sheet financing through special purpose vehicles
• Who is really holding the risk: Blackstone, Blue Owl, Apollo and Pimco's growing exposure
• Circular financing explained, and why Nvidia's SK Group and OpenAI commitments matter
• The private equity and IPO story: OpenAI, Anthropic and the test still to come
• Where this sits against a Federal Reserve giving markets no forward guidance
• The sceptic's case, and the QF-MI base case for the next 12 to 18 months

Follow QF-MI on Substack: https://qfmi.substack.com The Market Pulse and In the Spotlight research series are free to read. Subscribers also receive the Weekly Outlook, Weekend Debrief, and the Monthly Strategic Research Report, providing institutional-grade analysis of the capital flows and physical constraints shaping the AI industrial economy.

Chapters

0:19    Capital Formation Emerges
2:04   The Financing Layer
4:23   Telecom Bubble Comparison
7:03    Debt Markets Take Over
10:29  Demand Weakens for Bonds
13:20  Off-Balance-Sheet Leverage
16:12   Circular Financing Risks
19:39  Private Funding Boom
22:09  Capital as the Constraint
24:47  Fed Risk Returns
27:30  The Skeptics Case
29:49  Base Case Outlook
33:00  Nervous System of AI

Tags: AI, capital formation, hyperscalers, financing stack, bond markets, private credit, special purpose vehicles, circular financing, Nvidia, capital allocation, macro, AI infrastructure

• (00:19) - Capital Formation Emerges

Episode 6: Photonic Interconnects – The Next AI Bottleneck

For the past several years, the AI industry has been focused on one constraint: compute. More GPUs. Larger clusters. Faster processors.

But that bottleneck is changing.

As AI systems continue to scale, the limiting factor is no longer simply how much compute we can build, but how quickly data can move between processors, servers, racks and entire AI factories. Increasingly, the constraint is the network itself.

In this episode, we explore photonic interconnects and explain why the future of AI infrastructure will be built on light rather than copper.

Using a four-layer framework, we examine where photonics fits across the AI Continuum, from connections inside processor packages, through hyperscale data centres, all the way to laser communications between satellites in orbit. We also analyse one of the least understood strategic materials in the AI supply chain: indium phosphide, the foundation of modern optical communications and an emerging geopolitical bottleneck.

Finally, we present the QF-MI base case, highlighting where we believe the greatest investment opportunities, and risks, are likely to emerge over the next several years.

In This Episode

• Why the AI bottleneck is shifting from compute to data movement
• Understanding the I/O Wall and why GPUs are waiting for data
• Why copper is approaching its physical limits
• How co-packaged optics could transform AI hardware
• The four layers of the photonics ecosystem
• The rapid growth of 800G and 1.6T optical networking
• Why orbital computing depends on laser communications
• Indium phosphide: the hidden material underpinning AI infrastructure
• China's strategic position in the optical supply chain
• The QF-MI investment framework and key indicators we're monitoring

Follow QF-MI on Substack: https://qfmi.substack.com

The Market Pulse and In the Spotlight research series are free to read. Subscribers also receive the Weekly Outlook, Weekend Debrief, and the Monthly Strategic Research Report, providing institutional-grade analysis of the capital flows and physical constraints shaping the AI industrial economy.

Chapters

0:19 AI Supercycle Begins
2:33 The Photonics Bottleneck
6:27 Inside the Chip Limits
9:08 Co-Packaged Optics Arrive
10:51 Transceivers Power the Network
14:03 Laser Links in Orbit
15:53 Indium Phosphide Risk
19:08 Investment Base Case
21:52 Light Connects the Stack

Tags: AI, semiconductors, HBM, memory, energy, nuclear, uranium, copper, rare earths, data centres, photonics, capital allocation, macro, AI infrastructure

• (00:19) - AI Supercycle Begins

Episode 5: The Orbital Compute Thesis: Trading Terrestrial Constraints for Orbital Ones

Space does not bypass constraints. It trades them. After three episodes documenting the terrestrial binding constraints on the AI Supercycle, memory, power, and critical materials, this episode reaches the top of the AI Continuum and examines what happens when serious capital proposes moving compute off the planet.

The episode tests the engineering claims rigorously, from orbital solar physics and thermal management to the latency gap between training and inference workloads. It reviews the key players: SpaceX's million-satellite FCC filing, Google's Project Suncatcher, Blue Origin's Project Sunrise, Starcloud's GPU in orbit, Axiom Space's operational data centre nodes, Nvidia's Space One module, and China's Three-Body Computing Constellation.

The sceptics' case, led by SoftBank's Masayoshi Son, gets equal weight. The economics rest on one variable: launch cost per kilogram. The base case: directionally correct, but on a longer timeline than proponents suggest, with the near-term investable opportunity in the picks and shovels, not orbital compute itself.

Chapter Marks

 [00:00] Introduction and production note
 [01:32] The three binding constraints recap: memory, power, materials
 [02:06] Space trades constraints, it does not bypass them
 [03:05] The AI Continuum: from underground mines to orbit
 [05:52] Terrestrial constraints compounding: power, water, land, materials
 [08:06] What space offers: solar power, thermal management, and the engineering reality
 [10:37] Constraints that space introduces: radiation, latency, debris, maintenance
 [11:49] The players: SpaceX/xAI, Google Suncatcher, Blue Origin, Starcloud, Axiom, Nvidia
 [16:52] China's Three-Body Computing Constellation
 [17:45] The economics: launch cost per kilogram and the path to cost parity
 [21:36] Which workloads suit orbital compute: inference, not training
 [22:29] The sceptics' case: Masayoshi Son and the decisive years argument
 [23:10] Latency: distance, bandwidth, and why training in orbit is unworkable
 [24:39] Space debris, maintenance, and the Starship dependency
 [26:18] Governance: the regulatory void and the SpaceX concentration question
 [29:22] The QF-MI base case: 40% probability of cost parity within five years
 [31:17] Global Launch Intelligence Database: 150 orbital launches tracked in 2026
 [32:15] The AI Continuum: from the mine to the antenna
 [35:43] Production note and sign-off

Links

QF-MI research and subscriptions: qfmi.substack.com

Over the past two episodes we've explored the first two binding constraints shaping the AI Industrial Economy: high-bandwidth memory and delivered power. This week we move further down the supply chain. To the ground itself.

The AI revolution doesn't begin inside a data centre. It begins in copper mines, uranium deposits, rare earth refineries and the global supply chains that underpin every transformer, GPU, cable and power station.

In this episode we examine why critical materials may become the next major bottleneck in the AI Supercycle, and why markets may still be underpricing the scale of the challenge.

We also explore the latest developments shaping the investment landscape, including:

•  Micron's record earnings and what they reveal about structural shortages 
•  Apple's price increases as supply constraints begin to reprice the value chain 
•  Escalating tensions in the Middle East and implications for energy markets 
•  The G7 Critical Minerals Resilience and Production Alliance 
•  Proposed US tariffs on refined copper 
•  China's export controls on critical materials 
•  The long development timelines that make mining fundamentally different from semiconductors or power generation 
•  Why copper, uranium, rare earths, gallium, germanium and tin all matter to the future of AI infrastructure 

The episode concludes with my current base case for critical materials over the next three to five years and explains why memory, power and materials should be viewed as one interconnected system rather than three separate investment themes. The AI Supercycle isn't simply a software story. It's becoming one of the largest physical industrial build-outs in modern history.

Chapters

00:19 - AI Supercycle Overview

02:30 - The Underground Constraint

05:32 - Iran, Energy Markets & Macro Update

08:40 - The G7 Critical Minerals Alliance

10:18 - Copper Tariffs and Industrial Policy

13:04 - Mining Runs on Geological Time

15:08 - Why Copper Matters

18:07 - Uranium and the Nuclear Supply Chain

21:02 - The Wider Critical Materials Complex

23:52 - China's Strategic Leverage

25:47 - Base Case Outlook

28:48 - Connecting the Three Constraints

About The AI Supercycle

The AI Supercycle follows the capital flows, infrastructure investment and physical constraints shaping the AI Industrial Economy.

From semiconductors, hyperscale data centres and power grids to critical materials, orbital compute and embodied intelligence, each episode examines where capital is being deployed, where bottlenecks are emerging and what this means for investors, businesses and the global economy.

• (00:19) - AI Supercycle Overview

Episode 3: The Power Constraint: When Demand Meets Reality

This week on The AI Supercycle, we move from high-bandwidth memory to the second major bottleneck shaping the AI industrial economy: delivered power. Generating electricity is not enough. The real constraint is getting reliable power to AI factories at the right voltage, in sufficient quantity, and on a timescale that can support hyperscale growth.

In this episode:

•  Why the grid, not generation capacity, has become one of the defining bottlenecks of the AI Supercycle. 
•  The US-Iran agreement, energy markets, and why lower oil prices do not solve the power problem. 
•  SpaceX's first week as a public company and what its acquisition of Cursor tells us about AI infrastructure. 
•  Apple's agreement with Intel and the growing reshoring of semiconductor manufacturing. 
•  The "Two-Clock Problem" and why AI demand grows faster than power infrastructure can respond. 
•  Natural gas, nuclear, renewables and the longer-term possibility of orbital compute. 
•  Why transmission infrastructure, interconnection queues and transformer shortages matter. 
•  The investment implications for utilities, independent power producers and nuclear energy. 
•  The QF-MI base case: why power could become the primary binding constraint on the AI Supercycle by 2027. 

The most valuable asset in the AI industrial economy may not be a chip. It may be a power station.

Topics discussed

• AI infrastructure
•  Power grids and energy markets
•  Data centres and hyperscalers
•  SpaceX IPO and AI capital formation
•  Intel, Apple and semiconductor reshoring
•  Nuclear energy and utilities
•  Orbital compute
•  Critical materials and the AI Supercycle

Subscribe

📰 Substack: https://qfmi.substack.com

💼 LinkedIn Newsletter: The AI Supercycle

Chapters

 00:19 - New Name, New Cycle
 01:21 -  The Delivered Power Constraint
 02:09 - The Gulf Deal and Oil Flows
 05:56 -  SpaceX, Intel and Capital Flows
 10:29 -  Why Power Is Different
 16:27 -  Four Paths to More Power
 20:40 - The Grid Bottleneck
 23:34 -  Investing in Power
 25:12 -  QF-MI Base Case
 29:26 - Power Becomes the Bottleneck

Next Episode

Episode 4: Critical Materials - copper, uranium, and the physical inputs underpinning the AI industrial economy.

Tags

AI Supercycle

QFMI

capital flows

physical constraints

semiconductors

data centres

power grids

energy markets

delivered power

electricity demand

• (00:19) - New Name, New Cycle

Episode 2: The Memory Constraint: Why HBM Has Become AI's New Oil

Oil Semiconductor Bifurcation, Strategic Partnerships and the Fifth Binding Constraint

This week Nvidia locked up future memory supply through a multi-year strategic partnership with SK Hynix, while Alphabet reportedly ordered more than three million TPUs from Intel Foundry, signalling that hyperscalers are diversifying fabrication away from TSMC as demand overwhelms a single manufacturing ecosystem.

Tim Hardwick examines High Bandwidth Memory as the first binding constraint on the AI Supercycle. The supply chain runs through three countries and a handful of companies. Demand is accelerating from three directions: more memory per chip, more chips, and more inference workloads. HBM is sold out for the rest of this year.

The episode addresses the semiconductor correction, arguing that the sell-off reflects crowded positioning and leveraged ETF amplification rather than a change in the fundamental thesis. The PRISM framework assigns eighty per cent probability to a mid-cycle correction rather than a market top. The sell-off is bifurcated: ASML made all-time highs the same week Nvidia corrected over thirteen per cent, suggesting the market is repricing the most crowded expressions of the AI trade, not the thesis itself.

The QF-MI base case is that HBM remains structurally tight through 2027. Supply will grow but demand is likely to outpace it. The principal risks are deteriorating ROI on hyperscaler spending, tighter financial conditions, or geopolitical disruption around Taiwan or Korea. The episode also flags capital formation as a potential fifth binding constraint and previews Episode 3 on delivered power.

Chapters

0:17 HBM and the AI Supply Chain 
5:54 Capital Becomes the Constraint 
10:05 The Three HBM Producers 
12:52 Demand Is Accelerating 
15:57 Market Rally, Then Correction 
19:30 Correction or Top? 
23:16 The HBM Base Case 
26:58 Watching IPOs and Capital Flows

Tags: AI supercycle, HBM, high bandwidth memory, semiconductors, SK Hynix, Nvidia, TSMC, ASML, CoWoS, Intel Foundry, SpaceX IPO, capital formation, semiconductor bifurcation, memory constraint, data centres, inference, Blackwell, Rubin, capital allocation, macro

• (00:17) - HBM and the AI Supply Chain