The Record
Every price we can stand behind, and the receipts for each one. Term blocks only. No on-demand list rates, no spot averages, no simulated fills.
Illustrative rows are our own curve, not trades. They are watermarked and never enter a statistic.
Provenance of the record
Executed on our paper. Signed order form, hash + timestamp, click to verify.
Third party attested this trade to us. Counterparty known to us, banded to you.
Public index or listed rate. Source and date on the row.
We have not yet ingested a single third-party index print into the tape. When we do, it will say so here and it will never be mixed into a desk statistic.
Generated by our own curve. Never included in any statistic.
12 countable prints. 3 illustrative rows excluded from every number on this page — from n=, from the term price, from the spread, from the bucket counts.
Ornn OCPI does not disclose its print count or its contributor list. Neither does Silicon Data disclose which providers it surveys. We publish ours: the n behind every price is on the tile, the rows are on the tape, and every Desk row opens to a hash you can check. We are not claiming our sample is larger. We are claiming you can see it.
Our term price, by chip
Term price is the GPU-hour-weighted mean of our countable prints for that chip, each converted to its 3-month equivalent using the term multipliers in the instrument spec. It is not a settle, it is not an index, and it is not tradable. It is the average of what actually cleared.
The tape
12 rows · sorted by date, newest first| Symbol | Chip | GPUs | Term | Region · size band | Src | Verify | |||
|---|---|---|---|---|---|---|---|---|---|
| 28 Aug 26 | SL-H100-512-IB-US-3M | H100 | 512 | 3 mo | US · 512–1,023 GPUs | 2.740 | -0.23 vs 31 AUG 2026 settle 2.97 | Desk | |
| 24 Aug 26 | SL-H200-256-IB-US-3M | H200 | 256 | 3 mo | US · 256–511 GPUs | 4.180 | -0.24 vs 31 AUG 2026 settle 4.42 | Desk | |
| 19 Aug 26 | SL-H100-1024-IB-EU-6M | H100 | 1,024 | 6 mo | EU · 1,024–2,048 GPUs | 2.520 | -0.45 vs 31 AUG 2026 settle 2.97 | Desk | |
| 14 Aug 26 | SL-B200-512-IB-US-3M | B200 | 512 | 3 mo | US · 512–1,023 GPUs | 5.880 | -0.19 vs 31 AUG 2026 settle 6.07 | Contributed | — |
| 11 Aug 26 | SL-H100-256-IB-APAC-1M | H100 | 256 | 1 mo | APAC · 256–511 GPUs | 3.140 | +0.17 vs 31 AUG 2026 settle 2.97 | Desk | |
| 06 Aug 26 | SL-H100-2048-IB-US-12M | H100 | 2,048 | 12 mo | US · 1,024–2,048 GPUs | 2.310 | -0.66 vs 31 AUG 2026 settle 2.97 | Contributed | — |
| 31 Jul 26 | SL-H200-512-IB-US-6M | H200 | 512 | 6 mo | US · 512–1,023 GPUs | 3.960 | -1.66 vs 31 JUL 2026 settle 5.62 | Desk | |
| 27 Jul 26 | SL-B300-256-IB-US-3M | B300 | 256 | 3 mo | US · 256–511 GPUs | 7.350 | +0.07 vs 31 JUL 2026 settle 7.28 | Contributed | — |
| 22 Jul 26 | SL-H100-512-IB-EU-3M | H100 | 512 | 3 mo | EU · 512–1,023 GPUs | 2.690 | -0.07 vs 31 JUL 2026 settle 2.76 | Desk | |
| 16 Jul 26 | SL-B200-1024-IB-US-6M | B200 | 1,024 | 6 mo | US · 1,024–2,048 GPUs | 5.610 | -0.80 vs 31 JUL 2026 settle 6.41 | Desk | |
| 09 Jul 26 | SL-H100-256-IB-US-3M | H100 | 256 | 3 mo | US · 256–511 GPUs | 2.810 | +0.05 vs 31 JUL 2026 settle 2.76 | Contributed | — |
| 02 Jul 26 | SL-H200-1024-IB-US-12M | H200 | 1,024 | 12 mo | US · 1,024–2,048 GPUs | 3.740 | -1.88 vs 31 JUL 2026 settle 5.62 | Desk |
Basis is a dollar difference against a named, dated settle: the Ornn OCPI settle in force for the month the trade printed. We never publish a percentage against a listed on-demand rate. Listed rates are asks that nobody has to accept, and dividing by one manufactures a discount that was never negotiated.
Our July H200 prints sit 1.66 and 1.88 below the July settle of 5.62. On 31 August the Ornn H200 settle came down to 4.42. We were early, and we are showing you the rows that made us look wrong for six weeks.
What the indices say
H200Showing H200 — the widest disagreement between the two indices that settle real money. Pick a chip above to change it.
Ornn OCPI settles ICE futures and every Kalshi GPU market and does not disclose its print count or its contributor list. Silicon Data settles CME contracts pending CFTC approval and surveys listed on-demand rates rather than trades. getdeploying is a listed median and settles nothing. Three methods, three answers, same chip, same day. The term price is not in any of them, which is the whole reason this page exists.
Term structure of our own prints
No fitted line. A line needs two adjacent buckets with four or more prints each. One bucket qualifies.
Each print is divided by the Ornn OCPI settle in force for the month it printed — 31 Aug 2026 for August prints, 31 Jul 2026 for July prints — so H100 and B300 can sit on one axis without pretending they are the same instrument.
The 3-month bucket says our clearing level ran 2.7% under the transaction index, and the band is wide enough to contain 1.00. We are not claiming a discount. We are claiming a measurement.