Back
OpenAI Case Study - FutureSearch
webfuturesearch.ai·futuresearch.ai/openai-case-study/
This FutureSearch case study applies AI-assisted forecasting methodology to OpenAI, potentially useful for understanding how prediction markets and structured forecasting can inform AI safety and governance assessments, though content details are unavailable for deeper evaluation.
Metadata
Importance: 30/100blog postanalysis
Summary
A case study by FutureSearch examining OpenAI as a subject of AI forecasting and analysis, likely exploring predictions about OpenAI's trajectory, capabilities development, or safety practices. FutureSearch is an AI-assisted forecasting platform that applies probabilistic reasoning to questions about AI development.
Key Points
- •Applies structured forecasting methodology to analyze OpenAI's development and decisions
- •Demonstrates use of AI-assisted forecasting tools for evaluating major AI lab trajectories
- •May assess predictions around OpenAI's safety commitments, governance changes, or capability milestones
- •Serves as an example of how forecasting platforms can be applied to AI governance and safety questions
Cited by 1 page
| Page | Type | Quality |
|---|---|---|
| FutureSearch | Organization | 50.0 |
Cached Content Preview
HTTP 200Fetched Mar 20, 20264 KB
[← Back to Research](https://futuresearch.ai/research/)
## Key Takeaways
- **Widely cited reporting from The Information badly overestimated how big OpenAI's API business was in early 2024**, claiming 1BARRasofMarchwhenourfigureof1B ARR as of March when our figure of ~1BARRasofMarchwhenourfigureof500M was much more accurate given later disclosures
- **Our ChatGPT Plus estimate of 7.7M subscribers was accurate**, much more so than other public estimates at the time
- **We overestimated ChatGPT Teams subscribers** (900k estimate was way too high, likely ~100k actual)
## From Estimates to Validation: Three Months Later
Shortly after reports of **OpenAI's $3.4B ARR** came out on June 12, [FutureSearch released estimates](https://futuresearch.ai/openai-revenue-report) on how much came from ChatGPT Plus/Teams/Enterprise, and the API, based on how many users.
Exactly 3 months later, on September 12, **The Information reported a leak from OpenAI** that an email from COO Brad Lightcap claimed more than 10M ChatGPT Plus subscribers, and 1M on "higher-priced plans."
Taking this leak at face value, what did we get right and wrong in our June numbers?
## What We Got Right
### ChatGPT Plus: 7.7M Subscribers Was Accurate
Our estimate of **7.7M ChatGPT Plus subscribers looked accurate**, much more so than other estimates circulating at the time. The leaked information from OpenAI COO Brad Lightcap claiming "more than 10M" subscribers three months later aligns well with our June estimate, given the rapid (though slowing) growth we observed.
Our methodology involved analyzing data points, modeling growth trajectories, and inferring global subscriber numbers using reference classes like "viral consumer apps" and anchoring to plausible ancillary data.
### API Revenue: ~$500M ARR Was On Target
Our **API revenue estimate of ~$500M ARR looked accurate**, validating our claim that it was significantly lower than was being reported in mainstream news.
We challenged widely reported numbers about API revenue by using a method of bounding growth and back-calculating revenue, while rejecting speculative numbers that didn't align with credible data sources.
## What We Got Wrong
### ChatGPT Teams: 900k Was Way Too High
Our estimate of **900k Teams subscribers was way too high**. The actual number was likely much lower, around 100k.
This error stemmed from mistakes in our benchmarking and extrapolation methods. **We neglected important "inside view" perspectives** that would have suggested more conservative adoption rates for the Teams product.
### ChatGPT Enterprise: 1.2M Was Possibly Too High
Our **ChatGPT Enterprise estimate of 1.2M was possibly accurate, but probably too high**. While we don't have definitive data on the actual Enterprise subscriber count, the overall pattern suggests our estimate was on the optimistic side.
## The Value of P
... (truncated, 4 KB total)Resource ID:
c8db1bec51c509e7 | Stable ID: MWY0NzhkZT