Link to presentation below:
"AI eats the world"
Key takeaways:
Generative AI (GenAI) Platform Shift & Market Ambiguity
• Next Platform Shift: Generative AI is identified as the most recent platform shift, occurring every 10–15 years (following PCs, the Web, and smartphones). These shifts create new gatekeepers and new, greater value capture within the technology secto
• Uncertainty and Failure: The successful method for deployment remains unclear. Furthermore, an MIT report suggests that 95% of generative AI pilots at companies are failing
• Unknown Ceiling: Unlike previous platform shifts where physical limits for improvement were generally known, the extent of future improvement for Large Language Models (LLMs) is unknown
• Model Commoditization: Three years into the GenAI boom (as of November 2025), moats and clarity on product/value capture are lacking. Models are converging and leaders change weekly on general benchmarks, suggesting that the underlying models may become commodities.
• Engagement Gap: User data from June 2025 indicates an "engagement gap," showing that significantly fewer people use GenAI chatbots daily than those who use them occasionally (weekly or monthly)
Massive Capital Investment and Constraints
• FOMO-Driven Capex: Fear of Missing Out (FOMO) among Big Tech is driving a massive capex surge. The four largest hyperscalers are planning to spend approximately $400b in2025 alone
• Infrastructure Crisis: The investment cycle is constrained by supply issues. US data center construction is overtaking offices, but Nvidia and TSMC are unable to meet demand, and US power backlogs are becoming a major issue.
• Circular Financing: Companies like OpenAI, which lack their own cashflows or platform moats, are making ambitious infrastructure commitments (e.g., 30GW+ capacity, estimated at $1.4t) via complex, circular financing deals involving equity swaps and partnerships with entities like Nvidia and Oracle.
Adoption, Automation, and Value Capture
• Deployment Pattern: New technologies typically deploy in four steps: Absorb (make it a feature), Automate (obvious use-cases), Innovate (new products), and Disrupt (redefine the question).
• Early Use Cases: The most successful initial use cases are those that "absorb" existing tasks, particularly coding, marketing, customer support, and general automation. AI coding is viewed as a "new step change reduction in software creation costs" .
• Infinite Interns: AI is compared to providing "infinite interns" . This invokes the Jevons paradox: the question is whether companies will do the same amount of work with fewer people, or exponentially more work with the same number of people.
• Shifting Value: If models become commodities, value capture will shift away from the model itself toward factors like the best user experience (UX), proprietary vertical data, distribution, and solving the user's ultimate problem .
• The Disappearance of Automation: When automation is successful and widespread, it tends to disappear from conscious notice, similar to the disappearance of elevator attendants after automatic elevators (the "Autotronic") were introduced in 1950
*Summary provided with the help of Google NotebookLM.
- Todor
------------------------------
Todor Kostov
Director
------------------------------