Bold prediction: 99% of AI startups will be dead by 2026.

That’s according to Srinivas Rao, host of the Unmistakable Creative podcast, in his recent Medium article. His reasoning? Most AI startups are building on shaky foundations that won’t survive because their product and underlying technology aren’t defensible enough. Let me share how Rao has come to this prediction and why he believes that focusing on infrastructure is the answer.

The Three Fatal Flaws

1. OpenAI Dependency – Look around and you’ll see countless startups using AI to automate presentations, meeting notes, or video creation. Scratch beneath the surface and most are simply reskinning OpenAI’s data model. When you’re entirely dependent on someone else’s API, you’re not building a company—you’re building a feature.

2. Renting Intelligence, Owning Distribution – Rao calls this “the wrappers problem.” The startups that he deems to be at risk are all using APIs from OpenAI or Anthropic and only control product distribution. They’re essentially middlemen in the AI supply chain, which historically isn’t a defensible position.

3. No Infrastructure Ownership – Without owning core infrastructure like memory layers or workflow engines, these companies lack any real moat. Meanwhile, NVIDIA sits pretty as the kingmaker—powering GPT-4, Claude, and Gemini through their chips and deployment infrastructure.

The Infrastructure Test

Infrastructure companies win because others build on them and become dependent on them. Think AWS, Snowflake, or Stripe—the invisible layers that only matter when they break.

Rao references Peter Thiel’s seven questions every enduring company should answer:

1. Engineering — Can you create breakthrough technology?
2. Timing — Is now the right time for your business?
3. Monopoly — Do you have a big share of a small market?
4. People — Do you have the right team?
5. Distribution — Can you deliver your product effectively?
6. Durability — Will your position be defensible in 10-20 years?
7. Secret — Have you identified a unique opportunity others miss?

The Brutal Reality Check

In this “wrapper economy,” nobody’s asking these questions. Here’s the test: Would anyone rebuild your product if it disappeared? Would anything break if it went away?

If the answer is no, you’re not infrastructure—you’re just noise.

The survivors will be businesses that provide durable infrastructure that can’t simply be deleted or replaced by the next API update.

Main learning point: As product managers, we need to think beyond the AI hype and ask whether we’re building something defensible. Simply wrapping someone else’s intelligence in a prettier UI or just creating an automated version of a Job To Be Done – without an underlying infrastructure or intelligence layer – is likely provide a lasting answer to Peter Thiel’s seven questions.