SourceX Research · public sources
What AI developers say they need: public signals
Quick answer
This page is not a SourceX survey or demand index. It summarizes two kinds of public evidence: formal data-partnership programs that AI developers have announced, and independent research on whether public training data is running out. Together they give a public, verifiable picture of what some AI developers say they want, without any claim about who buys what through SourceX.
By Noah Loul · Updated
This report summarizes published public sources, listed at the end. It does not contain SourceX survey or deal data.
Process
AI developers say
OpenAI's public data partnerships program
In November 2023, OpenAI announced 'OpenAI Data Partnerships,' describing an effort to work with outside organizations to build public and private datasets for training its models, reported by OpenAI and TechCrunch.
OpenAI said it was seeking large-scale datasets 'that reflect human society' and are not already easily accessible online, across any modality including text, images, audio or video, with a particular interest in data that expresses human intention, such as long-form writing or conversations, reported by OpenAI.
Early named partners included the Icelandic government and Miðeind ehf, for Icelandic-language data, and the nonprofit Free Law Project, for legal documents, reported by OpenAI and SiliconANGLE.
Independent research on data scarcity
Epoch AI, an independent research institute, published a 2024 analysis estimating the effective stock of quality-adjusted, human-generated public text available for AI training at around 300 trillion tokens, and projected that if current trends continued, language models could fully use up that stock between 2026 and 2032, reported by Epoch AI.
A related 2024 Epoch AI report on whether AI scaling can continue through 2030 examined data scarcity alongside power availability and chip manufacturing capacity as one of several potential constraints on continued growth in AI training runs, reported by Epoch AI.
Reading these signals together
OpenAI's announcement and Epoch AI's research are separate things: one is a company's own public call for data partners, the other is independent modeling of how much public text data exists. Both point in the same general direction, that developers have publicly signaled interest in data beyond what is easily scraped from the open web, and that researchers have flagged public text as a potentially limited resource on a multi-year horizon.
Neither source tells us what any specific company pays for data, what formats are in highest demand today, or which AI developers are actively buying from U.S. businesses. Those details are not public.
SourceX and named companies
SourceX is not affiliated with OpenAI or Epoch AI. This page summarizes public reports; it does not mean OpenAI buys data through SourceX, and it does not mean any AI developer named here works with SourceX.
What this means for a business
If a U.S. company has years of its own records and is wondering whether there's outside interest, the honest starting point is that public programs like OpenAI's confirm some AI developers do seek data partnerships, but the details of what gets accepted, how it's vetted, and what's paid are set deal by deal and are not publicly standardized. A business considering this should expect to check its own rights to the data and to confirm who de-identifies it before it is used or shared (under a SourceX agreement, SourceX does that, through approved processing routes), regardless of who is asking.
Frequently asked questions
Does SourceX run a demand index for AI companies?
No. SourceX has no survey or demand index of its own. This page reviews public announcements and independent research only.
Does this mean OpenAI buys data through SourceX?
No. SourceX is not affiliated with OpenAI. This page summarizes OpenAI's own public announcement; it does not mean OpenAI buys data through SourceX.
Is AI training data actually running out?
Independent researchers at Epoch AI estimated the usable stock of public text data could be exhausted sometime between 2026 and 2032 under current trends, but this is a projection, not a certainty.
What kinds of data has OpenAI publicly said it wants?
OpenAI's 2023 announcement said it sought large-scale data across any modality that reflects human society and is not already easily accessible online, with particular interest in material expressing human intention, such as long-form writing or conversations.
Sources
Public sources, checked 2026-10-09. SourceX is not affiliated with the organizations named.
- OpenAI announced its Data Partnerships program in November 2023 and described the kinds of data it is seeking openai.com
- Description and context of OpenAI's Data Partnerships announcement techcrunch.com
- Early Data Partnerships participants included the Icelandic government, Miðeind ehf, and the Free Law Project siliconangle.com
- Epoch AI's 2024 estimate that public text data stock (~300 trillion tokens) could be exhausted between 2026 and 2032 epoch.ai
- Epoch AI's 2024 report examining data scarcity alongside power and chip constraints on AI scaling through 2030 epoch.ai
See if your company qualifies
A short company assessment. No data uploads are needed.