DeAI market summary

DeAI inference throughput

Theoretical maximum throughput of tracked DeAI hardware for the dashboard reference workload, aggregated across networks with compatible inventory data.

No comparable ratio separate measurements

Separate measures
Tracked DeAI theoretical maximum 642.04B tok/day
OpenAI reported API throughput 21.60T tok/day

DeAI market snapshot

Field-wide totals and coverage across the projects tracked by this dashboard.

B200 Index behind estimate 3.12K

Equivalent compute index, not a physical GPU count.

Comparable trend ▲ 8.4% over 29d

Trailing period with unchanged reporting coverage.

Measurement Theoretical max

Not observed utilization or delivered tokens.

Data coverage 9 of 14

networks with compatible throughput inputs

Aggregate theoretical throughput, tok/day

30 daily aggregate observations.

  1. 2026-10-01: 592174983168.0
  2. 2026-10-02: 645900062688.0
  3. 2026-10-03: 653769284832.0
  4. 2026-10-04: 665871881472.0
  5. 2026-10-05: 655596672480.0
  6. 2026-10-06: 644345044128.0
  7. 2026-10-07: 635179639680.0
  8. 2026-10-08: 600680730528.0
  9. 2026-10-09: 589870302048.0
  10. 2026-10-10: 667260992160.0
  11. 2026-10-11: 674187009696.0
  12. 2026-10-12: 650669597568.0
  13. 2026-10-13: 667247840352.0
  14. 2026-10-14: 615588002496.0
  15. 2026-10-15: 621345251808.0
  16. 2026-10-16: 635455729152.0
  17. 2026-10-17: 680370550560.0
  18. 2026-10-18: 640519529472.0
  19. 2026-10-19: 645427932480.0
  20. 2026-10-20: 634189412736.0
  21. 2026-10-21: 631468094400.0
  22. 2026-10-22: 605254068288.0
  23. 2026-10-23: 644356839456.0
  24. 2026-10-24: 649243800576.0
  25. 2026-10-25: 616227423840.0
  26. 2026-10-26: 618583732416.0
  27. 2026-10-27: 625892663520.0
  28. 2026-10-28: 651900152160.0
  29. 2026-10-29: 639389828736.0
  30. 2026-10-30: 642035303424.0

How to read this

  • DeAI throughput is the theoretical max throughput of tracked networks; OpenAI's number is reported API demand. They are separate measures, not a capacity-share ratio.
  • The reference workload is Qwen 3.6-35B, which requires about 70 GB of VRAM; incompatible GPUs contribute zero throughput.
  • The trend excludes earlier dates whose reporting-network set differs from the latest date.
  • Current totals combine each contributing network's latest collected value, so source timestamps can differ.
Read the full methodology