AI’s "Free Software" Boom Just Showed How the Bubble Dies

Executive Summary

The video examines two seemingly conflicting assumptions about the AI market:

1. AI will make software easy to copy and therefore less valuable.
2. AI companies may not generate enough cash to repay the debt financing their growth.

Its central distinction is that cheaply reproducing code is not the same as delivering dependable software that customers can use in established workflows. Software’s defensible value may lie in reliability, compatibility, accumulated user habits, and reproducible results—not merely in its visible interface or code.

Product, Strategy, and Competitive Dynamics

• AI-built software clones are arriving faster and at lower cost. Developer Brendan Thomas, a former Square engineer, released seven free, open-source apps intended to replace Adobe products. He said he used Anthropic’s Claude Opus 5.5. Another developer reportedly built a Photoshop competitor for about $2,000 in AI costs.
• The clones remain incomplete. Thomas’s development plan estimated that PhotoCraft covered 25–35% of a professional’s daily needs. Basic editing, shortcuts, and speed still needed work, and content-aware fill had not been implemented. In one reported test, PhotoCraft reproduced Photoshop’s smart-filter results correctly in only 5 of 30 cases.
• Product parity involves more than feature lists. Users reported issues with Photoshop files and smart objects, slow performance, errors, and FilmCraft freezing on large clip imports. As of October 11, five of the seven apps were still marked as in development.
• Reliability and workflow integration are the moat. Professionals pay for files that open and render consistently, familiar shortcuts, compatibility with clients and teams, and continuity with years of work. Switching costs include retraining and disruption across an organization’s tools and collaborators.
• Free is not automatically a winning go-to-market strategy. Free Photoshop alternatives have existed for decades, yet Adobe continues to sell to professionals. LibreOffice reportedly had 200 million users in 2018, compared with about 1.2 billion for Microsoft Office.
• Freemium can still support growth. Adobe said its free plan contributed to a strategy that had brought it more than 1 billion monthly users, with the aim of converting some to paid products.

Financing and Value Capture

• AI-related borrowing fell from $113 billion in June to $23 billion in September, according to Morgan Stanley. One explanation is that companies borrowed heavily earlier in the year, but lenders’ pricing also suggests increased caution.
• The video cites an estimated $466 billion in AI-related debt by this point in 2026, compared with $101 billion at the same point a year earlier. A separate $60 billion financing package involving Broadcom and Anthropic was still pending.
• Credit risk indicators worsened:
  • Oracle’s credit-default-swap cost reached $2.88 per $100 of debt annually, a record.
  • CoreWeave’s reportedly rose from about $3.60 to over $10 in nine days; its 2032 bonds traded at about $0.77 per dollar.
  • Oracle-related project loans for its New Mexico data-center project reportedly fell from about $0.90 to $0.85 on the dollar.
• The core infrastructure question is whether leasing computing capacity earns enough to service the debt used to build data centers. Demand for AI chips may be real without guaranteeing that every infrastructure provider can earn an adequate return.
• The video identifies three possible destinations for AI-generated value:
  1. Model developers capture it through paid access to their models.
  2. Users capture it through cheaper or free software.
  3. Intermediary infrastructure companies and lenders bear the downside if prices fall or revenues fail to cover debt.

Metrics and Examples

• Adobe: Record quarterly revenue of $6.76 billion, up 13% year over year. Its creative-professional business reached $4.65 billion, also up 13%. Despite that quarter, its shares were down about 30% for the year at the time discussed.
• Anthropic: Reported revenue rose 45% in Q3 to $16.8 billion, from $11.6 billion in Q2 and under $5 billion in Q1. The speaker noted that growth had slowed from more than doubling in a prior quarter.
• TSMC: Reported its sixth consecutive record quarter, with profits up 69%—evidence cited for strong underlying chip demand.
• Adobe legal example: The subtitles say Adobe settled a U.S. government case over hidden cancellation fees for $150 million in March, without admitting wrongdoing. The example highlights how cancellation friction can damage customer trust and invite regulatory scrutiny.

Business Playbook and Execution Implications

• Treat AI prototypes as a starting point, not proof of product substitution. Test real customer workflows, edge cases, file compatibility, performance, and repeatability before claiming parity.
• Compete on the complete job to be done. Reliability, interoperability, support, and preservation of existing work can be more defensible than feature imitation alone.
• Measure switching friction and customer value. Track successful workflow completion, error rates, time to complete tasks, retention, and conversion from free to paid—not just downloads or feature counts.
• Stress-test infrastructure economics. For compute and data-center businesses, evaluate utilization, lease revenue, operating costs, debt service, and sensitivity to lower prices or slower demand.
• Be cautious about assuming value capture. A technology can make production cheaper while directing the savings to customers rather than the software vendor or infrastructure provider.

Presenters and Sources Mentioned

• Presenter: Unnamed narrator of House of El: AI.
• Sources and people referenced in the subtitles: Brendan Thomas and his development plan and code; Morgan Stanley; Adobe; a Chinese tech outlet’s PhotoCraft testing; LibreOffice; Microsoft Office; Anthropic; a Financial Times–quoted credit investor; TSMC; Broadcom CEO Hock Tan; and an SMBC analyst.
• The video also includes a sponsored segment for Incogni.