Synthetic data generation
Synthetic data can train useful models outside language. Google's TabFM uses generated tables for predictive tasks. The commercial test is whether generated examples improve performance on real data the model has never seen. More examples do not by themselves mean better coverage, better privacy or margins a data supplier can defend.
Research snapshot · updated Nov 12, 2026
State of the art (2026)
Synthetic data is now load-bearing in three distinct markets, not one. In frontier post-training it is mainstream: RLAIF, distillation and self-generated reasoning traces dominate alignment work at Anthropic, OpenAI, Google DeepMind and DeepSeek, with feared model collapse not materialising in practice when synthetic and real data are mixed. In physical AI, NVIDIA released Cosmos 3 in June 2026 as an open world-foundation model generating physics-aware training data for robotics and autonomous fleets. In regulated tabular data, Mostly AI, Gretel, Tonic.ai and Hazy sell privacy-preserving generation into finance and healthcare. The labelling layer consolidated when Meta took a 49% Scale AI stake (valuing it at $29bn) in June 2025 and hired Alexandr Wang.
Core thesis
Synthetic data is used where real data is scarce, sensitive or unbalanced. It does not remove privacy or bias problems automatically. Guidance from the UK Information Commissioner's Office says synthetic data falls outside personal-data rules only to the extent that it cannot be linked to identifiable people, and that this has to be assessed in each case. Synthetic data can also carry over or amplify bias in the data it was generated from. The open commercial question is what margins specialist suppliers can keep as large AI labs generate their own.
Explore the deeper analysis in CanaryIQ
Explore what is covered below. These previews show the structure; log in to read the full analysis.
Signal stack
Evidence stacked leading → lagging
9 signals
Signal stack
Evidence stacked leading → lagging
Technology-native KPIs
Metrics that predict trajectory, tracked over time
6 tracked
Technology-native KPIs
Metrics that predict trajectory, tracked over time
Landscape map
Who builds what, and who depends on whom
97 players · 6 layers
Landscape map
Who builds what, and who depends on whom
Catalyst calendar
Dated events that will move the position
6 ahead
Catalyst calendar
Dated events that will move the position
Technology roadmap
Milestones on the path to maturity
8 milestones
Technology roadmap
Milestones on the path to maturity
Watchlists
Companies, people and papers, each with a condition for removing it
20 · 19
Watchlists
Companies, people and papers, each with a condition for removing it
Decision frameworks
What our view means for your job
Locked
Decision frameworks
What our view means for your job
Thesis changelog
When our view changed, and why
9 updates
Thesis changelog
When our view changed, and why
Change our mind
3 disconfirming conditions
You've read the verdict. The file is much deeper.
The full signal stack, technology-native KPIs tracked over time, the landscape of who depends on whom, the dated catalyst calendar, decision frameworks for every desk, live watchlists and a record of every time our view on Synthetic data generation has changed. All of it is in CanaryIQ.