DeepSequel (June 2026)
The Capital Cycle is the official podcast that Marathon Asset Management (the London firm founded in 1986) launched in 2024, hosted by financial historian Edward Chancellor, who interviews Marathon's investors about each Global Investment Review letter — applying the firm's long-term, contrarian "capital cycle" supply-side approach.
In plain words
This report shows how Chinese companies are copying or beating global giants at a fraction of the cost—DeepSeek's AI model costs 90% less, and Chery's hybrid SUV sells for half the price of a Land Rover. For regular investors, this means stocks like Samsung and Micron, which rely on high profits from near-monopolies, might be riskier than they look. Chinese rival CXMT is rapidly expanding in low-end DRAM chips, threatening their pricing power. Worth a read because it warns that even hard-to-replicate advantages (like ASML's chip-making machines) can erode over time with enough capital and patience.
In January 2025, China's DeepSeek launched the R1 large model with a training cost of only approximately $6 million (rivals spent billions), with per-million input/output token pricing at $0.55/$2.19, causing Nvidia's market capitalization to evaporate nearly $600 billion in a single day. Similarly,
Theme and Background
This chapter discusses how China’s low-cost capacity expansion challenges seemingly impenetrable technology moats. As seen in the cases of DeepSeek and Chery, China is replicating and even surpassing competitors’ products at extremely low costs and rapid speeds. The market may be underestimating the impact of this trend on the profit pools of global oligopolies.
Core Argument
The author argues that the current pricing power and valuation premiums enjoyed by the DRAM triumvirate (Samsung, SK Hynix, Micron) and ASML rest on a fragile assumption — that no new entrant can challenge their position at low cost. CXMT’s capacity expansion in China is proving this premise invalid. The counterintuitive point is that the market views AI-driven HBM demand as a structural tailwind, but this may instead become a trigger for disruption in the low-end market, as the giants voluntarily cede low-end share to focus on high-end capacity.
Key Arguments and Data
- DeepSeek case: R1 model training cost ~$6 million (competitors billions), pricing $0.55/$2.19 per million tokens, with performance comparable to rivals but at one-tenth the cost.
- Chery Jaecoo 7: Plug-in hybrid range surpasses the Range Rover Evoque, priced at half the latter; became the UK’s best-selling car in March, reaching one million sales in three years.
- Solar history: Prices fell 90% over a decade to 11 cents per watt, undergoing two brutal shakeouts.
- Humanoid robots: Unitree G1 priced at $16,000, competitors at ~$150,000.
- DRAM current status: The three giants’ share prices have risen 80%-240% year-to-date, with operating margins exceeding 60%. CXMT expanded monthly capacity from 100,000 to 300,000 wafers, revenue surged over 700%, and is expected to capture 15% of global DRAM share this year.
- EUV catch-up: CXMT currently cannot access ASML’s EUV equipment, but recent estimates put the EUV gap at only 5 years (previously over 10 years), and it has already planned to mass-produce HBM using DUV alternatives by the end of this year.
| Item | DRAM Trio (Samsung/SK Hynix/Micron) | CXMT (China) |
|---|---|---|
| Current DRAM share | ~95% (oligopoly) | Expected to reach 15% this year |
| Capacity direction | Shift to high-end HBM, actively reducing low-end | Focus on low-end, rapid capacity expansion |
| Operating margin | >60% | Loss-making/low profit (subsidy-driven) |
| Core risk | Rapid loss of low-end share, pricing power collapse | Unable to enter HBM profit pool in near term |
| Valuation | Part of trillion-dollar market cap | Unlisted |
Companies/Assets Involved
- Samsung, SK Hynix, Micron: Bearish view. The author believes their high profits are based on the premise that Chinese competitors have not yet entered in scale. Once CXMT impacts the low-end market, it could wipe out tens of billions in market cap.
- CXMT (ChangXin Memory Technologies): Bullish view. Monthly capacity tripled, revenue up 7x, representing the largest marginal change in the global DRAM industry.
- ASML: Bearish view. EUV monopoly is considered the most solid hardware moat, but China, with capital and patience, is closing the gap — estimated at only 5 years remaining, and has already found DUV alternatives.
- Sany Heavy Industry: Bullish view. One of the world’s largest construction machinery manufacturers, over 60% of revenue from overseas, 28% gross margin leaves very low profit room for new entrants. Pioneer in electrification, secured orders from global clients like Holcim. Price-to-sales ratio <2x.
- BYD, CATL, DJI, Unitree: Background cases demonstrating Chinese companies’ cost advantages and technological leadership across multiple sectors (CATL holds nearly 40% of global battery share, BYD surpasses Tesla, etc.).
Investment Implications
1. Avoid low-end DRAM oligopolies: Investors should reassess the valuation safety of Samsung, SK Hynix, and Micron. CXMT’s capacity expansion could cause low-end DRAM prices to plummet, thereby compressing the overall profit margins of the three giants. Even if AI-driven HBM demand is real, the erosion of low-end business may still severely drag down profits and market cap.
2. Focus on Chinese "survivors": Investment should target Chinese manufacturers that have survived China’s brutal capital cycles and possess global cost advantages (e.g., Sany), rather than global oligopolies relying on moat assumptions. Sany’s globalization, electrification, and low valuation (<2x price-to-sales) provide downside protection.
3. Beware of overpriced scarce assets: Seemingly irreplicable technological barriers like ASML are also facing challenges. Capital and patience can buy time, ultimately breaking through the boundaries of "genius and physics." Investors currently paying high premiums for these assets may face valuation restructuring risk.