AI eats software? (February 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 uses the story of the Jacquard loom to explain how AI coding tools are flooding the software industry with excess capacity and competition. Even though stocks like ServiceNow have dropped over 50%, their valuations remain high—not a bargain. The author warns against value traps and advises avoiding weak-moat SaaS companies that charge per seat. Worth reading because it applies capital cycle logic to show why software stocks may still have further to fall.
This report draws an analogy between the capital cycle story of the Jacquard Loom and the current software industry, exploring the disruption of AI on software companies. Core view: software production capacity is increasing, competition is intensifying, and pricing power is weakening. Since the end
Theme and Background
This chapter introduces the capital cycle story of the Jacquard Loom as a parallel to the AI disruption currently facing the software industry. The author points out that software companies have fallen from their status as market darlings "eating the world" to a lagging sector. Since late July 2025, the S&P 500 Software Industry Index has underperformed the S&P 500 by 35 percentage points. The core context is that AI coding tools are significantly boosting software production efficiency, leading to industry capacity expansion and intensified competition.
Core Thesis
The author's core investment thesis is: The software industry is in the "overcapacity" phase of the capital cycle, and investors should remain cautious rather than blindly bottom-fishing due to falling stock prices. Counterintuitive judgments include:
- Software companies' stock prices have plunged sharply (e.g., ServiceNow down over 50% in the past year), but valuations remain elevated and have not entered a margin of safety.
- Market concerns about "vibe coding" and non-programmer application development may be exaggerated, but the more fundamental threats are capacity expansion driving intensified competition and weaker pricing power.
- Although some software companies (e.g., Autodesk) are relatively safer due to high technical barriers, the overall industry faces structural risks.
Key Arguments and Data
1. Signals of Capacity Expansion
- Data from mobile analytics firm Appfigures shows iOS app releases in 2025 increased 24% year-over-year, reaching the highest level since 2017, with the acceleration coinciding with the launch of agentic coding tools in early 2025.
- Anthropic CEO Dario Amodei stated that his engineers no longer write code and predicted that "within 6-12 months, AI will do most or all of the work of software engineers."
2. Valuations Remain High
- The S&P 500 Software Index trades at 10x 2025 revenue estimates. After Marathon's adjustment (treating stock-based compensation as an expense), the P/E ratio exceeds 37x.
- Despite a share price decline of over 50%, ServiceNow still trades at 8x 2025 revenue and 62x GAAP P/E (Marathon believes stock-based compensation should be expensed).
3. Weakening Competition and Pricing Power
- Increased capacity will lead software companies to encroach on each other's feature sets, and startups can launch replicas faster. Historical capital cycle patterns show that more capacity means more competition and weaker pricing power.
4. Business Model Risks
- If AI causes the front-end UI to be intermediated by intelligent agents, the per-seat pricing model may shift to a consumption-based model, which, in an environment of falling technology/computing costs, would actually compress SaaS revenue.
Comparative Data Table:
| Metric | S&P 500 Software Index | ServiceNow |
|---|---|---|
| 2025 Revenue Multiple | 10x | 8x |
| P/E Ratio (GAAP) | >37x (adjusted) | 62x |
| Stock Price Change (Past Year) | - | Down over 50% |
Companies/Assets Involved
- ServiceNow: Bearish signals are evident. The stock has fallen sharply but valuations remain high (8x revenue, 62x P/E). The author implies it has yet to fully price in heightened competition and business model risks.
- Autodesk: Relatively safer. It provides highly technical, mission-critical design software with high cost of failure, engineer training dependence, and switching costs forming a moat. The author holds this stock in client portfolios.
- Lovable (tool): Mentioned as a representative of vibe coding, allowing non-programmers to quickly build applications, but it is not an investment target.
- Anthropic: The CEO's views are cited, but the author cautions that he is not neutral (has an IPO intention).
Investment Implications
- Avoid pure-play software companies with high valuations and weak moats, especially those relying on per-seat SaaS models.
- Be wary of "value traps" after stock price declines: ServiceNow's case shows that even after a 50% drop, valuations may still lack appeal.
- Focus on capacity expansion-driven intensifying competition: Pricing power across the software industry will systematically weaken in the future.
- Favor software with high technical barriers, high switching costs, and mission-critical use cases (e.g., Autodesk-type). Such companies are better positioned to withstand capacity shocks.
- Overall strategy: Remain cautious during the capacity expansion phase, avoid rushing to bottom-fish, and wait for opportunities after the capital cycle turns.