# Surge AI

URL: https://aidataforsale.com/companies/surge-ai
Publisher: AI Data For Sale (Marketing Signals Ltd)
Updated: 2026-11-10

Surge AI supplies RLHF, RL environment and language annotation data to AI labs. What it does, the DataAnnotation connection, how it buys data, and alternatives.

A data annotation company focused on human feedback, reinforcement learning environments and language data.

## In short

- Surge AI was founded in 2020 by Edwin Chen and focuses on RLHF, RL environments and language annotation.
- Wikipedia, citing press reports, puts its 2024 revenue at $1.2 billion, reached without outside funding.
- MIT Technology Review reported in October 2026 that Surge buys workplace archives from closed companies to build RL environments.

## Profile

| Field | Detail |
| --- | --- |
| Type | Human feedback, RL environment and annotation vendor |
| Founded | 2020, by Edwin Chen |
| Customers named in reporting | Anthropic, OpenAI, Microsoft, Meta and the US Army |
| Worker platform | Lists DataAnnotation among its subsidiaries |
| Buys data from companies? | Reported to buy Slack, email and calendar archives to build agent training environments |

## What Surge AI does

Surge AI provides data for reinforcement learning from human feedback (RLHF), builds reinforcement learning environments and annotates language data. Its listed customers include Anthropic, OpenAI, Microsoft, Meta and the US Army, and Wikipedia reports 2024 revenue of $1.2 billion with the company bootstrapped by its founder ([Wikipedia](https://en.wikipedia.org/wiki/Surge_AI)). Its worker platform DataAnnotation is listed among its subsidiaries.

## Surge as a buyer of company archives

MIT Technology Review reported in October 2026 that Surge uses purchased workplace data, such as Slack, email and calendar histories, to build reinforcement learning "gyms" where AI agents practice tasks. It looks for varied archives that include disagreements and judgment calls ([MIT Technology Review](https://technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models)). That makes it one of the buyers a data holder may meet, directly or through an intermediary. See [workplace chat data](https://aidataforsale.com/data/workplace-chat-data) and [email archives](https://aidataforsale.com/data/email-archives).

## How it compares with a data broker

Surge is a vendor to AI labs and, increasingly, a buyer of raw material. If you are a data holder, it is a potential counterparty, not an adviser. A broker works for the deal, checks the buyer and helps you set terms. Read [questions to ask an AI data buyer](https://aidataforsale.com/learn/questions-to-ask-an-ai-data-buyer).

## Things to check before you sign

- If Surge is buying your data: whether it will resell or pass data to its lab customers, and on what terms.
- What personal information will be removed before transfer, and who does the removal.
- Whether the license limits use to training environments, or allows any use.
- How long copies are kept, and how deletion is confirmed.

## Alternatives and related companies

- [Scale AI](https://aidataforsale.com/companies/scale-ai)
- [Mercor](https://aidataforsale.com/companies/mercor)
- [SimpleClosure](https://aidataforsale.com/companies/simpleclosure)
- [Agent trajectory data](https://aidataforsale.com/data/agent-trajectory-data)
- [AI Data For Sale: how our brokerage works](https://aidataforsale.com/how-it-works)

## Questions

### Is DataAnnotation part of Surge AI?

Surge AI lists Data Annotation among its subsidiaries, according to [Wikipedia](https://en.wikipedia.org/wiki/Surge_AI).

### Does Surge AI buy company data?

MIT Technology Review reported in October 2026 that Surge buys workplace archives from closed companies to build agent training environments.

## Sources

- [Wikipedia: Surge AI](https://en.wikipedia.org/wiki/Surge_AI)
- [MIT Technology Review: AI labs are buying data from dead startups to train models (October 2026)](https://technologyreview.com/2026/10/13/1145755/ai-labs-buying-data-from-dead-startups-to-train-models)