Data prices · Semiconductors & electronics
What AI labs pay for semiconductors data
Process runs, test results and yield history from fabs and test houses are some of the hardest industrial data for AI labs to find. No public prices exist, and export control rules decide what can be shared before price comes into it.
What buyers want
Records your team already keeps
01
Process runs
Recipes, tool settings and the measurements that followed, run by run.
02
Test and yield data
Wafer test results and yield over time, linked to process changes.
03
Excursions
What went wrong on the line, how it was traced and what fixed it.
Price signals
What public deals and research show
~$20k-$300k
Price range for working replicas of systems that agents train in, from a simple app to a complex product. Fab and test systems are candidates.
Epoch AI, 2026We found no public sale price for fab, test or yield data. NIST has written about the need for scaled data sharing in semiconductor manufacturing, which signals demand, but buyers set prices deal by deal.
The wider market
$203M
Data licensing contracts Reddit disclosed in early 2024, with terms of two to three years.
TechCrunch$250M+
Reported value of News Corp's five-year content deal with OpenAI, including cash and credits.
Press Gazette$10k-$100k
Per-company payouts for shut-down startups' Slack messages, tickets and email archives, across nearly 100 deals.
Forbes, April 2026$200-$2,000
Typical price per reinforcement learning task, according to Epoch AI's interviews with people who build them.
Epoch AI, 2026Pricing factors
What moves the price
- Export classification
- Data that counts as controlled technology under US or EU rules needs a review and possibly a license, or it stays out.
- Links to outcomes
- Process changes tied to yield results are worth far more than either alone.
- Node and process
- Rarer processes and equipment sets are worth more.
- Years of history
- Longer runs of records show change over time and how outcomes played out, which a single year can't.
- Rights you can prove
- Buyers pay for data you own or are allowed to license. Data held for customers without their permission won't sell, so our calculator cuts the estimate when rights are unclear.
Our calculator
How we estimate semiconductors data
Our indicative model starts from a base value per system for about five years of history at an 11-50 person company, then applies a sector weight of 1.8x, your history, team size and rights. It is an estimate we confirm on a call, not a market price.
| System | Base value |
|---|---|
| Process, test & fab data | $80k |
| ERP & operations | $25k |
| Docs, wikis & SOPs | $10k |
Recurring
Paid again every month
Every lot run adds new process and test records. A refresh term pays for each month's batch once export screening is in place.
How recurring revenue worksRules that apply
We screen out export-controlled process data before offering anything to a buyer. US controls on semiconductor technology depend on the data, the destination and the end user.
Source: Congressional Research ServiceValuation
What is your data worth?
No access to your systems needed. We confirm the number on a 30-minute call.
Questions
Semiconductors data, answered
Is our process data export controlled?
Will competitors see our data?
What leaves the fab?
Keep reading
Selling semiconductors dataWhat makes data valuable to AI labs
Labs pay most for records they can't scrape or generate: real decisions, real outcomes, long history and fresh data, with clear rights to use them.
5 min read
Episodes, records and environments: the three things AI labs buy
Records are de-identified rows, episodes are human actions with outcomes in order, and environments are working replicas. Each sells at a different price.
6 min read
How to prepare an export for an AI data buyer
Export complete history with its audit trail, keep IDs that link tables, and scrub on your own machine before anything leaves. A field-by-field checklist.
6 min read
Prices in other industries