Part of: this research — view all 4 rows
What is Being Measured and the 2026 Baseline
The core metric is the relative shift in the share of external data-acquisition spend directed toward capturing practitioners' actual paid work, comparing end-2028 to end-2026. The 2026 baseline for this "real-work" share is likely in the low single digits to low teens. It includes programs exchanging training rights for tokens or discounts, such as OpenAI's Data Sharing Program Sharing feedback, evaluation and fine-tuning data, and API ... and Google's recently launched invite-only Gemini Contributor Tier Data logging and sharing - Gemini API. It also arguably includes broker purchases of corporate Slack, CRM, and email archives, like Micro1's $1B data-buying commitment https://www.businessinsider.com/micro1-invests-license-data-ai-model-development-2026-10 and Google's roughly $10M Spirit Airlines purchase https://www.businessinsider.com/google-spirit-airlines-data-privacy-union-deal-2026-11.
Why the Share is Likely to Grow
The median of 1.73x reflects a clear trend: as frontier labs push toward autonomous work agents, demand for realistic, long-horizon production trajectories is surging https://www.businessinsider.com/ai-workplace-data-training-agents-robots https://www.businessinsider.com/ai-training-employee-data-emails-slack-dms-2026-10. Programs that subsidize live enterprise sessions are still in their infancy or restricted to invite-only tiers; by 2028, these mechanisms are highly likely to expand, especially if continual learning begins to yield dividends. Because the 2026 baseline share is so small, even a moderate reallocation of budgets toward capturing live enterprise sessions and corporate data exhaust results in a large proportional increase.
Constraints from "Manufactured" Data and Classification Ambiguities
The primary constraint on this ratio is the explosive, parallel growth of the denominator. Manufactured-for-training data—specifically expert-authored RL environments and synthetic data—is scaling massively. Anthropic has discussed spending upwards of $1B on RL environments Top 10 Data Annotators for AI Labs (2026 Benchmark), while Scale AI notes that nearly half of its new projects fall into this category https://www.businessinsider.com/ai-next-data-grab-work-reinforcement-learning-environments-google-meta-2026-8. However, as AI automates synthetic environment generation , growth in the volume of these environments need not translate into proportionate growth in acquisition spending. Labs will likely reallocate a greater share of their data budgets toward capturing real practitioner workflows, modestly favoring a higher real-work share despite the expansion of manufactured data. Furthermore, there is significant classification ambiguity. It is uncertain whether historical archival purchases (like company Slack records) will be counted as "licensed static content" (manufactured data) rather than real-work capture. Similarly, it is debatable whether token allowances and API discounts will be strictly classified as "expenditure."
Key Uncertainties and Tail Risks
The broad distribution (p10 of 0.82 to p90 of 4.10, with p25 of 1.18 and p75 of 2.70) captures these definitional gray areas and the fact that neither the numerator nor denominator is publicly reported. The left tail below 1.0 accounts for scenarios where the denominator outpaces real-work capture due to enterprise privacy pushback https://www.businessinsider.com/google-spirit-airlines-data-privacy-union-deal-2026-11, regulatory friction, or the sheer scalability and legal safety of synthetic RL environments. Conversely, the heavy right tail accounts for the possibility of labs successfully implementing widespread, cash-equivalent data-for-access subsidies at the enterprise tier, fundamentally transforming how high-value proprietary work is captured.
Set against related questions, this estimate was slightly adjusted upward to reflect that as AI automates synthetic environment generation , labs will likely reallocate a greater share of their data budgets toward capturing real practitioner workflows.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited