Hyperscalers & cloud · Thesis 04 · settles Jan 1 · 66 days · rung 1/3

B200 rental price index, late December?

7 voices · 7 claims2 sides2 machines

The other numbers are hidden until you lock yours. The readers, the machines and the Call are waiting.

The two sides

If no

NVIDIA will not make that statement

Expects softer demand or new chip supply to pull B200 rental rates below the mark.

If yes

NVIDIA will say inference is over half of demand

Expects tight B200 supply or strong demand to keep rental rates elevated into year end.

How it settles

Yes if the average of Ornn's daily B200 price index from December 25 to December 31, 2026, on data.ornn.com, is 9.40 dollars per GPU-hour or higher. Ornn Compute Price Index, B200 daily settlement (published index history)

The exact question and the full rule

Will the Ornn Compute Price Index for the B200 average at or above $9.40 per GPU-hour over its seven daily settlements from December 25, 2026 through December 31, 2026?

Take the daily settlement of the Ornn Compute Price Index for the B200, in US dollars per GPU-hour, as Ornn publishes it at data.ornn.com, for each day from December 25, 2026 through December 31, 2026, and average the settlements. A day Ornn does not publish takes Silicon Data's published B200 rental price index (neo-cloud) for the same day. The answer is yes if the mean is at or above $9.40 per GPU-hour; no if it is below $9.40 per GPU-hour.

If the source is late or silent: A day with neither price is left out of the mean; if three or more of the seven days have neither, the test is inconclusive. Each day's value is the one Ornn's published index history (and, for a fallback day, Silicon Data's index page) shows when the test is read by January 3, 2027; a revision published later does not change it.

The case · 7 voices · 2 minutes
NVIDIA will not make that statement1 voice argues noNVIDIA will say inference is over half of demand6 voices argue yes
Stronger if no · 1 voice

NVIDIA will not make that statement

Kerrisdale Capital Management LLCSep 1

Training workloads require high-end GPUs and can be run remotely, aligning with CoreWeave's model of renting expensive, cutting-advantage capacity on long-term contracts. Inference, by contrast, is steady-state, latency-sensitive, and cost-conscious…

Stronger if yes · 6 voices

NVIDIA will say inference is over half of demand

Aurelion ResearchSep 11

Once models are trained they must run on every use, so rising usage converts directly into recurring compute demand; agentic systems that complete tasks and use tools multiply the number of inference calls per user action.

The Call's side, the strongest voice against it and what would change it appear after you lock your number.

Open the dossierThe thesisLock your number first
The dossier · 5 minutes
The machines' reasons

2 models ran this one. Their numbers and reasons appear after you lock yours.

The thesis

Will Nvidia say inference passed 50% of demand by 2027?

As AI shifts from building models to running them in everyday products and agents, inference calls could outgrow training runs. The question is whether NVIDIA says so on the record.