- The question “Is AI a bubble?” is considered to be beside the point; the real issue is that AI consists of multiple different bubbles, with varying risks and collapse timelines.
- The AI ecosystem is divided into 3 layers: wrapper companies, foundation models, and infrastructure.
- The layer most prone to collapse is wrapper companies: startups that merely wrap OpenAI’s or competitors’ APIs, add an interface, and charge a monthly fee.
- Jasper.ai once reached about $42 million ARR in its first year, but this model faces three major risks: being “swallowed” by big platforms adding features, commoditization, and zero switching costs.
- When Microsoft, Google, or Salesforce integrate AI directly, wrapper products can lose value overnight.
- Most wrappers do not possess proprietary data, core workflows, or the ability to lock in users; Cursor is a rare exception thanks to deep integration into the coding workflow.
- It is forecast that from late 2025 to 2026, many wrapper companies will go bankrupt or sell off.
- The foundation models layer, such as OpenAI, Anthropic, and Mistral, has a stronger “technical moat” but still faces a valuation bubble.
- OpenAI is cited as an example with about $1 trillion in committed infrastructure investment, while projected revenue is only around $13 billion.
- Future competitive advantages lie not just in model training but in inference optimization, memory management, token speed, and infrastructure efficiency.
- The 2026–2028 period is forecast to see strong consolidation, with only 2–3 major players remaining.
- The infrastructure layer (chips, data centers, cloud, memory, storage) is considered the least “bubbly.”
- Global AI spending in 2025 has exceeded $600 billion and could touch $1.5 trillion, yet infrastructure retains long-term value.
- Nvidia’s revenue for Q3 of fiscal year 2025 reached about $57 billion, with the data center segment alone accounting for $51.2 billion, reflecting real demand.
- The “bubble bursting” process will occur in a sequence: wrappers collapse first, foundation models consolidate later, and infrastructure adjusts but survives long-term.
- Advice for builders: the biggest risk is not building a wrapper, but remaining just a wrapper; one needs to own the workflow, data, and distribution channels.
📌 AI is not a single bubble but multiple stacked bubbles. The AI ecosystem is divided into 3 layers: wrapper companies, foundation models, and infrastructure. Wrapper companies are at risk of collapse within 18 months, foundation model companies will consolidate within 2–4 years, while infrastructure companies, despite short-term excess, will retain long-term value. Advice for builders: the biggest risk is not building a wrapper, but remaining just a wrapper; one needs to own the workflow, data, and distribution channels.
