Question
January 2027: Monthly average compute price of NVIDIA's H200
The resolution metric is the January 2027 arithmetic mean of the Ornn Compute Price Index (OCPI) for the NVIDIA H200. As of mid-June 2026, the H200 spot price is oscillating between $3.47 and $3.78 per hour api.ornnai.com. The forecasted probabilities reflect a tug-of-war between strong structural depreciation and persistent, spike-prone AI compute demand.
The dominant downward force is product-cycle substitution. By January 2027, the H200 will be functionally two generations old. Blackwell already accounts for the majority of NVIDIA Data Center revenue sec.gov, and the subsequent Rubin architecture is in full production with partner availability expected in H2 2026 investor.nvidia.com. As these newer, more cost-efficient architectures absorb premium demand, the H200 will increasingly age into a cheaper, legacy tier. This structural decline anchors the median expectation in the low-to-mid $3.00 range, reflected by the 56% probability for Above $3.00 and 43% for Above $3.50.
Conversely, the primary upside risk is the relentless macro demand for AI compute, which routinely triggers severe spot market volatility. NVIDIA's record Q1 FY2027 Data Center revenue of $75.2B and robust forward guidance nvidianews.nvidia.com underscore this sustained appetite. We have empirical precedent for such volatility: in May 2026, the H200 index spiked to a daily maximum of $7.01 api.ornnai.com api.ornnai.com. This inherent volatility preserves a fat right tail in the distribution.
However, the upper-tail probabilities (10% for Above $6.00; 7% for Above $6.50) are deliberately constrained by the resolution criteria. The market resolves on a full-month average, not a daily peak. For context, despite hitting $7.01 during the May 2026 supply crunch, the monthly average for May was only $4.55 api.ornnai.com. Sustaining an average above $6.00 for an entire month on a depreciating, older-generation asset is highly improbable, heavily discounting the extreme right tail while acknowledging the real risk of episodic capacity crunches.