The AI Bubble Isn’t One Bubble – It’s a Cascade. Here’s What Comes Next.
The breathless hype around artificial intelligence has led to a single, looming question: are we in an AI bubble? The more pertinent question, as VentureBeat recently highlighted, is which AI bubble are we in, and when will each one inevitably burst? The ecosystem isn’t a monolith; it’s a layered structure, each with its own risk profile and timeline. Ignoring these distinctions is a recipe for disaster.
The Looming Collapse of the “Wrapper” Layer
The most immediate danger lies with companies simply repackaging existing AI models – the “wrapper” companies. These businesses take powerful APIs like OpenAI’s GPT and add a user-friendly interface, often charging a monthly subscription. While some, like Jasper.ai, initially saw explosive growth, their long-term viability is questionable.
The threats are multifaceted. Large tech platforms can easily absorb these features into existing products. Microsoft could integrate AI writing tools into Office 365, Google into Gmail, and Salesforce into its CRM – effectively eliminating the need for standalone solutions. Furthermore, as foundation models become more commoditized and pricing drops, margins for wrappers shrink to nothing. Switching costs are virtually non-existent; users can effortlessly move between platforms or directly to the underlying API.
Pro Tip: If your business model relies solely on reselling access to a large language model, you need a serious plan B. Focus on building unique value *around* the AI, not just *on top of* it.
Cursor, a developer-focused tool, offers a rare example of defensibility. By deeply integrating into developer workflows and creating proprietary features, it’s evolved beyond a simple wrapper. However, Cursor remains an outlier. Expect significant failures in this segment between late 2025 and 2026.
Foundation Models: Consolidation is Coming
The companies building the large language models (LLMs) themselves – OpenAI, Anthropic, Mistral – are in a more secure position, but not immune to risk. Richard Bernstein of Richard Bernstein Advisors points to OpenAI’s massive investments ($1 trillion in deals, including a $500 billion data center buildout) against relatively modest revenue ($13 billion) as a clear sign of bubble-like dynamics.
The key to survival in this layer won’t be simply building a large model, but optimizing its performance and scalability. Engineering breakthroughs in inference optimization – specifically, addressing the “memory wall” through techniques like extended KV cache architectures and faster token throughput – will be crucial. Companies that can deliver economically viable AI inference at scale will thrive.
Did you know? Nvidia’s revenue in Q3 of fiscal year 2025 reached $57 billion, a 62% year-over-year increase, largely driven by demand from AI infrastructure. This demonstrates real investment, not just hype.
Expect consolidation between 2026 and 2028, with 2-3 dominant players emerging through acquisitions and market share gains. Smaller model providers will likely be absorbed or forced to shut down.
The Infrastructure Layer: The Surprisingly Solid Ground
The most resilient part of the AI boom is the infrastructure layer – Nvidia, data centers, cloud providers, and specialized memory systems. While AI-related spending is projected to exceed $1.5 trillion globally in 2025 (Gartner), this investment is fundamentally different from the speculative frenzy surrounding many applications.
Like the fiber optic cables laid during the dot-com bubble, the infrastructure being built today will retain value regardless of which AI applications ultimately succeed. The demand is real and driven by tangible needs. Nvidia’s continued growth is a testament to this. Modern AI infrastructure isn’t just about storage; it’s about the entire memory hierarchy, from GPU HBM to DRAM to high-performance storage optimized for AI workloads.
Some short-term overbuilding is possible in 2026, but long-term value retention is highly probable as AI workloads continue to expand.
The Cascade Effect: A Phased Collapse
The AI boom won’t end with a single, dramatic crash. Instead, we’ll see a cascade of failures:
- Phase 1 (Now – Late 2026): Wrapper and white-label companies face margin compression and feature absorption. Hundreds of startups will fail or be acquired for minimal returns. Over 1,300 AI startups are currently valued at over $100 million, many of which are overvalued.
- Phase 2 (2026 – 2028): Foundation model consolidation. Expect 3-5 major acquisitions as tech giants consolidate their positions.
- Phase 3 (2028 onwards): Infrastructure spending normalizes, but remains elevated. Some data centers may experience temporary underutilization, but overall demand will continue to grow.
What This Means for Builders: Move Upstack
The biggest risk isn’t being a wrapper; it’s *staying* a wrapper. Owning the user experience is paramount. If you’re building in the application layer, you need to move upstack:
- From Wrapper to Application: Don’t just generate outputs; own the entire workflow before and after the AI interaction.
- From Application to Vertical SaaS: Build execution layers that lock users into your product. Create proprietary data and deep integrations.
- Focus on Distribution: Your competitive advantage isn’t the LLM; it’s how you acquire, retain, and expand your user base.
FAQ: Navigating the AI Landscape
- Q: Is all AI investment doomed?
A: No. The infrastructure layer is likely to provide long-term value. However, many application-layer companies, particularly wrappers, are at high risk of failure. - Q: What should investors look for?
A: Focus on companies with strong engineering capabilities, defensible moats, and a clear path to profitability. - Q: Will AI still be transformative?
A: Yes, but the transformation will be more gradual and selective than current hype suggests.
The AI revolution is underway. But navigating this landscape requires a clear understanding of the underlying layers and the risks associated with each. Don’t get caught in the wrong bubble.
Want to learn more about building resilient AI applications? Explore our resources on data infrastructure and AI engineering.
