Date: 2026-05-18 (Monday)

Pillar: Spicy take + Funding (combo)

Topic: 89% of AI startup revenue goes to OpenAI/Anthropic. Sierra's $950M raise at $15.8B is the contrarian counter-example. Three things Sierra got right that 99% of AI wrappers got wrong.

Why this story today

• Continues the running thesis (Vapi funding → Mira Murati interaction layer → Sierra agent platform). The third beat in the same narrative: the value sits one floor above the model.
• Two huge data points in one post: 89% revenue concentration (The Information) AND Sierra's $950M / $15.8B (May 4 raise).
• Bret Taylor is OpenAI's Board Chair. The implicit narrative "OpenAI's own chair just raised $950M for the layer OpenAI doesn't own" lands hard.
• Actionable framework for AI entrepreneurs: pick a $100B legacy industry, replace the labor, sell to enterprise.
• Spicy enough to drive replies. Specific enough to drive saves.

Source list

• TechCrunch — Sierra raises $950M as the race to own enterprise AI gets serious
• CNBC — Bret Taylor's Sierra raises nearly $1B in latest AI capital push
• The AI Insider — Sierra Secures $950M at $15B Valuation
• The Decoder — AI startup revenue hits $80B but Anthropic + OpenAI take 89%
• CNBC — Anthropic teams with Goldman, Blackstone on $1.5B AI venture

LinkedIn post

89% of all AI startup revenue goes to OpenAI and Anthropic.

That stat dropped last week. The wrapper era is supposed to be dead.

Then Sierra raised $950M at a $15.8B valuation. Bret Taylor's AI agent company. Eight quarters from launch. $150M ARR. 40% of the Fortune 50 already paying.

The wrapper era is not dead. The bad wrappers are.

Here's what Sierra got right that 99% of AI wrappers got wrong.

1. They sold to Fortune 50, not freelancers.
No $20-a-month indie users. Insurance, banks, telcos paying seven figures a contract. Different go-to-market. Different valuation math.

2. They built on top of every model, not one.
Sierra is model-agnostic. When GPT-5 ships or Claude Opus 4.7 drops, Sierra swaps the backend. The customer never sees the price hike, the rate limit, the new bug. That makes Sierra the platform, not the dependency.

3. They picked one painful workflow and replaced the entire function.
Not a Slack bot for support. The whole customer service department. The math closes the deal in one meeting.

The pattern is clear.

OpenAI and Anthropic are taking the model layer.
The 11% of revenue outside them is going to companies who pick one $100B legacy function, replace the labor, and sell to the Fortune 500.

If you're building an AI startup right now, run your idea against that filter.

Pick a $100B industry.
Replace the labor, not the software.
Sell to enterprise.

That is the only wrapper playbook that survives.

What's the next Sierra-shaped opportunity you see?

@RAHIL JAIN

X / Twitter post

89% of all AI startup revenue goes to OpenAI and Anthropic.

The wrapper era is supposed to be dead.

Then Sierra raised $950M at $15.8B.
8 quarters from launch. $150M ARR. 40% of the Fortune 50.

3 things they did that 99% of AI wrappers got wrong:

1. Sold to Fortune 50, not $20/mo indie users.
Seven-figure contracts with insurance, banks, telcos. Different math.

2. Stayed model-agnostic.
GPT-5 ships, Claude Opus 4.7 drops, Sierra swaps the backend. Customer never sees the price hike or the rate limit.

3. Replaced the whole customer service department, not the chat widget.
Not a Slack bot for support. The whole function. Closes the deal in one meeting.

The pattern:
OpenAI and Anthropic take the model layer.
The 11% goes to companies who pick one $100B legacy function, replace the labor, sell to the Fortune 500.

Pick a $100B industry. Replace the labor. Sell to enterprise.

That's the only wrapper playbook that survives.

@rahilnjain

Media: Text-only post today (same gap as last run, flagging for next iteration). Future: 3-slide carousel — (1) 89% / 11% stat split, (2) Sierra stats card, (3) the 3-step filter.