The World's First Daily GPU Rental Benchmark
The Silicon Data GPU Rental Index tracks the daily spot price for renting high-end NVIDIA GPUs from cloud providers and specialist compute platforms. Published every US business day, it aggregates tens of thousands of data points across clouds, brokers, and compute marketplaces into a single, methodology-driven benchmark.
This is the compute equivalent of an energy spot index. Where WTI crude tracks the marginal cost of oil, SDH100RT tracks the marginal cost of AI compute — the most critical input cost of the modern AI economy. Everything that runs on AI — model training, inference, agents — runs on GPUs rented at prices this index reflects.
Silicon Data is backed by DRW and Jump Trading Group, two large proprietary trading firms and market makers that originate from Chicago's open outcry futures pits.
Scarcity is the Name of the Game
Steve Hou, Head of Research at Silicon Data, characterises the current market directly: based on the tens of thousands of data points Silicon Data tracks, GPU rental prices continue to move modestly upwards. Scarcity remains the dominant theme.
When GPU availability tightens — capacity sold out across providers — prices move up. When new supply comes online from hyperscalers or marketplace entrants, prices soften. The index reflects this in real time rather than relying on anecdotal procurement data or lagged corporate disclosures.
The data is also divergence-aware. Different GPU types and tiers move independently. The H100 spiked 10% between December 2025 and January 2026 while A100 and B200 held flat — targeted demand, not a broad market move. B200 surged 24% in March 2026 while H100 barely moved — reflecting B200's newer, scarcer supply base. These cross-chip signals are not visible from any single company's reporting.
Not All GPU Indices are the Same
This is the single most important thing to understand when reading GPU rental data. In late May 2026, Steve Hou posted a Bloomberg chart comparing Silicon Data's SDH100RT against a competing H100 index (ORNNH100). The difference was stark.
Both indices ended at nearly identical prices (~$2.63–2.64) despite wildly different journeys. The media coverage of "GPU rental prices swinging lower" was tracking the volatile competing index — not Silicon Data's. Hou's response was measured: different data sources and methodologies are allowed to naturally disagree.
Volatile competing indices likely measure a narrower slice of the market — a subset of providers, spot listings prone to noise, or simpler averaging that amplifies outliers. Silicon Data's broad provider coverage and rigorous construction produces a more stable series that better reflects the true clearing price. When media reports suggest dramatic GPU price swings, the first question to ask is: which index?
The analogy to established commodity markets is apt: WTI and Brent sometimes diverge, energy spot and futures can disconnect temporarily, different natural gas indices reflect different regional dynamics. The GPU rental data market is still maturing, and methodology is not yet standardised. Silicon Data is building deliberately toward the rigour of a commodity benchmark — hence the CME Group partnership.
Why We Track the Marginal Price
The index tracks two tiers, and the distinction matters:
Neo-Cloud prices are tracked here because they are where real supply and demand clear in real time. The Neo-Cloud tier is where you see a compute crunch before it reaches the broader market narrative.
Key Signals from the Index
GPU Pricing Is Becoming a Financial Asset Class
CME Group × Silicon Data — First Compute Futures Market
CME Group, the world's leading derivatives marketplace, and Silicon Data announced a partnership to launch the world's first compute futures market, pending regulatory review. Contracts will reference Silicon Data's daily GPU rental indices. CME Chairman Terry Duffy: "Compute is the new oil of the 21st century."
Exchange-traded futures require a credible, manipulation-resistant daily benchmark. CME's endorsement validates Silicon Data's methodology as fit for institutional use — the same standard applied to oil, natural gas, and agricultural commodities before they became mature futures markets.
Futures pricing introduces a forward curve for GPU compute. Steep contango signals expected tightening; backwardation signals expected supply relief. This forward curve — which will eventually be publicly visible — becomes an additional signal layer beyond the daily spot index.
Institutional money will increasingly reference this data. Fund managers with positions in AI infrastructure (NVIDIA, hyperscalers, Neo-Cloud operators) can use daily GPU rental indices to track utilisation and pricing power in near real time — a capability that did not exist before Silicon Data built it.
Source & Methodology
| Field | Detail |
|---|---|
| Provider | Silicon Data (silicondata.com) |
| H100 Ticker | SDH100RT Index (Bloomberg / Refinitiv) |
| B200 Ticker | SDB200RT Index (Bloomberg / Refinitiv) |
| Frequency | Daily, US business days |
| Methodology | Aggregates tens of thousands of rental listings across clouds, brokers, and compute marketplaces. Designed to reflect the marginal clearing price — not a simple average. Filters outliers and transient distortions. |
| Coverage | Neo-Cloud and Hyperscaler tiers tracked separately. H100, A100, B200, MI300X available. |
| Backers | DRW, Jump Trading Group |
| Distribution | Bloomberg, Refinitiv, portal.silicondata.com |
| Chart update | Every Friday 07:00 SGT — posted to #gpu-rental Discord |
| On-demand chart | Type /gpuchart in Telegram bot |