Silicon Data
by Silicon Data
Financial-grade GPU rental price indices and AI compute benchmarks, published daily on Bloomberg
Silicon Data publishes independent, financial-grade price indices and performance benchmarks for AI compute — GPU rental rates, LLM inference token costs, memory prices, residual values and carbon intensity — delivered daily via Bloomberg tickers, a direct API and its own portal. It is built for compute buyers negotiating GPU contracts, data-centre operators and financial institutions that need to price and hedge compute risk.
Silicon Data is an independent benchmarking and market-data company that publishes financial-grade price indices for AI compute, positioning itself as the reference layer the GPU rental and inference markets have lacked. Founded in April 2024 by Carmen Li and headquartered in New York, the company published what it describes as the world's first GPU compute index in 2025 and now runs a portfolio of nine indices covering H100, H200, A100, B200 and MI300X rental prices, GDDR6 memory, blended LLM inference token costs and GPU residual values. The indices carry Bloomberg tickers — SDH100RT for H100, SDA100RT for A100, SDB200RT for B200 and SDLLMTK for token expenditure — and are also delivered through a direct API, the Silicon Data portal and distribution partners including LSEG/Refinitiv and Kaiko. The methodology normalises each rental record to a per-GPU, per-hour rate, applies multi-stage filtering to remove reporting errors, and adjusts for rental type, geography, CPU platform and GPU variant before aggregating across providers; every index is republished once per business day, and the company claims coverage of 95% of neo-cloud GPU providers, 100% of major hyperscalers and 80% of the global rental market across more than 3.5 million datapoints spanning up to eight years. Adjacent products include SiliconMark for physical GPU throughput and latency benchmarking, SiliconSiteIQ for data-centre capacity underwriting, SiliconCarbon for compute carbon accounting, SiliconPriceIQ and a GPU forward curve. In August 2026 Silicon Data closed a $30.5 million Series A led by the Valor Atreides AI Fund, with CME Group, DRW, Jump, Samsung, VanEck, F-Prime, Wintermute, Tectonic and Further participating, following a $4.7 million seed in 2025; CME Group plans to settle a cash-settled GPU futures contract against Silicon Data benchmarks, pending regulatory approval.
Infrastructure and FinOps leaders committing serious capital to GPU capacity, plus the trading desks and lenders now underwriting compute — anyone who has to defend a GPU price or a compute valuation with something other than a vendor quote.
An external, daily-published reference price for GPU-hours and inference tokens, so capacity contracts and compute budgets can be benchmarked against the market instead of negotiated blind.
At a Glance
- Category
- Data & Analytics
- Pricing
- Subscription, Contact for pricing
- Target Market
- CTOs, CIOs, FinOps Leaders, Financial Analysts, Data Center Operators
- Deployment
- Cloud-only, API-based
- Founded
- 2024
- Headquarters
- New York, United States
- Team Size
- 1-10
- Customers
- 1,000+ registered users across semiconductor manufacturers, AI developers and financial institutions
Key Features
- ✓Daily GPU rental price indices
Covers H100, H200, A100, B200 and MI300X rental rates, processed and republished every business day.
- ✓Bloomberg Terminal distribution
Indices carry dedicated tickers such as SDH100RT and SDLLMTK, and also reach users via API, portal, Refinitiv and Kaiko.
- ✓SiliconMark performance benchmarking
Validates real GPU throughput and latency across providers, so price can be compared against delivered performance rather than advertised specifications.
- ✓LLM token price tracking
The Token Expenditure Index, Token Market Pulse and Token Pricebook standardise blended inference pricing across model vendors.
- ✓Forward curve and predictive pricing
GPU forward curve and SiliconPriceIQ project rental prices for hedging, budgeting and procurement negotiation.
- ✓Site underwriting and carbon accounting
SiliconSiteIQ assesses data-centre GPU capacity for underwriting while SiliconCarbon quantifies compute carbon impact for ESG reporting.
- ✓Normalised, filtered methodology
Each record is converted to a per-GPU per-hour rate, filtered for reporting errors and basis-adjusted for geography, rental type and GPU variant.
Capabilities
Use Cases
- •Benchmark a GPU cloud quote
Check a vendor's proposed rate against the daily market index before signing a multi-year capacity commitment.
- •Hedge compute price exposure
Use index-settled instruments to manage GPU rental price volatility once CME's cash-settled compute futures launch.
- •Forecast inference budgets
Track blended LLM token prices across vendors rather than each provider's list rate to model real inference cost.
- •Value GPU inventory
Apply market-calibrated residual values when reselling, insuring or pledging GPU hardware as collateral.
- •Report compute carbon intensity
Use SiliconCarbon measurements across providers and regions to support ESG and sustainability disclosure requirements.
Ideal For
Best For
- ✓CTOs and infrastructure leaders negotiating GPU cloud contracts who need an external price reference
- ✓Financial institutions pricing, hedging or lending against AI compute exposure
- ✓Data-centre and neo-cloud operators benchmarking their own pricing and utilisation against the market
- ✓FinOps and procurement teams forecasting GPU and LLM inference spend
- ✓Investors and analysts tracking the unit economics of the AI buildout
Not Ideal For
- ✗Teams running a handful of GPUs or sitting on a fixed cloud commitment — the data is priced and packaged for institutional use and adds little at small scale
- ✗Buyers who need an independently governed, auditable benchmark today: the methodology is proprietary and the CME futures contract that would give it external validation is still pending regulatory approval
- ✗Anyone looking for a compute marketplace or procurement platform — Silicon Data sells data and benchmarks, it does not broker, resell or provision capacity
Deployment
Market & Ratings
1,000+ registered users across semiconductor manufacturers, AI developers and financial institutions
Market Analysis
Pros
- ✓The only GPU price index distributed on Bloomberg Terminal with its own tickers, which is what makes it usable inside existing institutional workflows
- ✓CME Group both invested and plans to settle a regulated futures contract against the benchmarks — the strongest third-party validation available for a private index
- ✓Methodology is published in outline, explaining per-GPU-hour normalisation, error filtering and basis adjustment rather than presenting the index as a black box
- ✓Broad claimed coverage — 95% of neo-clouds, 100% of major hyperscalers, 3.5M+ datapoints over up to eight years — with daily republication
Cons
- ✗No independent user reviews exist anywhere — G2, Capterra, TrustRadius, Hacker News and Reddit returned nothing as of August 2026, so buyers cannot sanity-check data quality against peers
- ✗The headline use case is not live: the CME GPU futures market remains pending regulatory approval, so index-settled hedging is still prospective rather than available
- ✗The index is vendor-run and proprietary; subscribers must trust filtering and basis adjustments they cannot independently audit or reproduce
- ✗Roughly six employees support a product sold on financial-index standards, which is thin operational redundancy for a benchmark others intend to settle contracts against
- ✗No public pricing beyond a seven-day portal trial, and distribution through Bloomberg or Refinitiv adds a second subscription cost
Pricing
Portal trial
$0
- ✓7-day access to the Silicon Index dashboard
- ✓90-day historical pricing
- ✓Market direction and predictive modelling
- ✓Neo-cloud vs hyperscaler comparison
- ✓Carbon insights
Institutional data & index licence
Contact for pricing
- ✓Direct API access
- ✓Bloomberg and Refinitiv/LSEG distribution
- ✓Full index portfolio
- ✓SiliconMark, SiliconPriceIQ, SiliconSiteIQ, SiliconCarbon
- ✓Alternative datasets
No list pricing is published anywhere. Access starts with a seven-day portal trial covering the index dashboard, 90-day price history, neo-cloud versus hyperscaler comparisons and carbon insights; production use is an institutional data licence quoted through sales. Index data consumed via Bloomberg Terminal, Refinitiv/LSEG or Kaiko carries those platforms' own subscription costs on top, so total cost depends heavily on which distribution channel a buyer already has.
Security & Compliance
Connect
Sources
This page was written from 6 sources, 3 on domains other than silicondata.com.
Stay Ahead of the Curve
Weekly enterprise AI insights for technology leaders. No spam, no vendor pitches—unsubscribe anytime.
SubscribeRelated Products
QueryStory
Agentic data platform that turns plain-English questions into auditable, decision-ready business narratives
turbopuffer
Vector and full-text search built object-storage-first, at roughly a tenth the cost of RAM-resident vector databases
Ellis
AI-native operations platform that reconciles private credit fund data and runs the close, reporting and monitoring
Monte Carlo
Data and AI agent observability for enterprises running production AI