Every mature commodity market in the world has free, public price benchmarks.
Brent crude oil prices stream daily from ICE — free to anyone who wants to read them, underpinning roughly 70% of global oil contracts. Henry Hub natural gas pricing is published by NYMEX and serves as the reference for North American gas. The SOFR rate (which replaced LIBOR) is published daily by the New York Fed and underlies approximately $300 trillion in derivative contracts. None of these benchmarks are paywalled. They're public goods.
AI compute pricing isn't yet. Today, the most comprehensive cross-provider GPU and token pricing data lives behind subscription paywalls at SemiAnalysis, Silicon Data, and Ornn — at $10,000–$50,000 annual subscription tiers that exclude most buyers from accessing the data they need to make rational decisions.
That's the gap Mercatus GPU Index and Mercatus Token Index close. Both are free. Both are public. Both will remain so. This article is the case for why open price indices belong in AI compute, what we built, and how to use it.
Every commodity market has open benchmarks
The pattern is remarkably consistent across global commodity markets: as a market matures, it produces free, public price benchmarks that become the reference for all downstream financial activity：
- Crude oil: Brent dated prices are published daily by Platts and ICE, free to view. The Brent benchmark underpins about 70% of global oil supply contracts. WTI plays the same role for North American crude. Anyone — refiners, hedgers, journalists, students — can see today's price without paying.
- Gold: The LBMA Gold Price is set twice a day in London and made freely available.
- Natural gas: Henry Hub futures and spot prices are published by NYMEX/CME and disseminated freely. The benchmark anchors North American gas markets and serves as the basis for thousands of bilateral supply contracts.
- Electricity: ISO/RTO day-ahead and real-time prices are posted openly by market operators.
- Interest rates: SOFR (Secured Overnight Financing Rate) is published daily by the New York Federal Reserve, free to view. SOFR replaced LIBOR, which was also published as a public rate. The two together have underpinned approximately $300 trillion in derivative contracts.
- Currencies： WMR/Reuters fixing rates, ECB reference rates, FED H.10 — every major FX benchmark is published free and public. Commercial FX trading platforms are paid; the benchmarks are not.
In each case, the benchmark is infrastructure. Open access is a feature, not a missed monetization opportunity. The benchmarks exist to enable trade — and trade in turn generates the commercial revenue that funds the exchanges and clearinghouses producing the benchmarks.
The financial industry's open-infrastructure tradition
Beyond price benchmarks, there's a deeper tradition in financial markets of keeping foundational infrastructure free even when it could be paywalled. Two examples that AI compute should learn from:
Bloomberg Open Symbology (BSYM). In 2009, Bloomberg released its security identifiers (FIGI, formerly Bloomberg Global ID) as an open standard, free and licensable to anyone. Bloomberg could have charged for them — they were already the de facto identifier system used by every major financial firm. Instead Bloomberg made BSYM free, expanded its industry adoption, and continued making money on the high-margin commercial layer (Bloomberg Terminals at $25,000+ per user per year). The free identifiers grew the market that the paid terminals served.
FIX Protocol. The Financial Information eXchange protocol — the messaging standard that virtually every electronic trading venue uses — is an open, free, unowned standard. The FIX Trading Community manages the spec; nobody pays for the protocol itself. Trading platforms compete on execution quality, latency, and ecosystem; the protocol layer is shared infrastructure.
The pattern in both cases: commercial value accrues at the application and trading layers, not at the data and standards layers. Open infrastructure expands the market that paid commerce serves.
This is the same logic that should govern AI compute pricing.
AI compute doesn't have open benchmarks yet (almost)
For all the financial sophistication of AI infrastructure markets in 2026 — billions of dollars deployed, multi-year capacity contracts, cloud GPU spend a top-three line item at major tech companies — there is no widely-adopted free public price benchmark for the underlying compute.
Today, the most comprehensive cross-provider AI hardware and token pricing data is concentrated at three subscription-based research firms:
SemiAnalysis. Premium AI hardware research subscription, with institutional tiers running $10,000–$50,000+ per year. Strong on chip-level analysis, model training economics, and quarterly reports. The most comprehensive single source for AI hardware pricing intelligence — but locked behind a tier most buyers can't justify.
Silicon Data. Subscription pricing intelligence focused on data center hardware including GPUs. Mid-tier institutional pricing. Used primarily by procurement teams at large enterprises.
Ornn. Newer entrant in cross-provider GPU pricing. Subscription model. Growing coverage but still gated.
There's a real reason institutional buyers pay these subscription fees. The data is genuinely useful. But the structure is wrong for the market AI compute is becoming:
1. Price discovery is increasingly time-sensitive. Cloud GPU pricing moves on weekly cycles in 2026. Quarterly research reports are stale by the time they're published. Subscription products that deliver weekly or monthly updates are too slow for procurement teams managing real-time spend.
2. Supply is decentralizing. The 22+ cloud GPU providers Mercatus tracks today will grow to 100+ in three years. Subscription analysts can't cover the long tail at sufficient depth — their economics favor focusing on the top 10 providers and ignoring the next 90. Long-tail providers (which often offer the best pricing) get systematically excluded.
3. Pricing data should not be a competitive advantage of well-resourced buyers. A startup deploying $50,000 of GPU compute should have the same access to cross-provider pricing as a hyperscaler deploying $50 million. Subscription paywalls create asymmetric information that disadvantages smaller buyers and entrenches incumbent advantage.
4. The methodology should be public. Subscription products often disclose limited methodology — buyers must trust the analysts' judgment without being able to verify how prices were aggregated. For pricing data that drives capital allocation decisions, opaque methodology is a real flaw.
A market this large deserves better data infrastructure.
Mercatus GPU Index — what it is
Mercatus GPU Index is a free, public, real-time benchmark for cloud GPU pricing across the global provider ecosystem.
What it tracks:
- Cloud rental pricing for NVIDIA H100, H200, A100, and emerging Blackwell (B100, B200) GPUs
- 22+ providers globally as of May 2026, growing
- On-demand, reserved (1-year and 3-year), and spot/preemptible rates
- Per-region pricing where providers offer regional differentiation
- Cross-provider price spreads (the gap between cheapest and most expensive)
How it's updated:
- Refreshed daily across most providers
- Continuous tracking with historical price trajectory
- Public methodology — see docs.mercatus-ai.com/methodology for the full data dictionary
How to access:
- Web UI at mercatus-ai.com/gpu-index — free, no signup required
- Public API for programmatic access — free at individual tier, institutional tiers available for high-volume usage
- Embeddable widgets for journalists, researchers, and platforms that want to surface the data on their own sites
What it costs: $0 for individual access. Always.
What it doesn't include: owned-hardware secondary market prices (those are tracked separately and less reliably), proprietary enterprise contract pricing (which is contractually private), and bespoke arrangements that aren't publicly listed.
Mercatus Token Index — what it is
Mercatus Token Index is the per-token pricing equivalent of GPU Index. Where GPU Index tracks the cost of renting hardware, Token Index tracks the cleared price of running inference on it across major LLM providers.
What it tracks:
- Per-token input and output pricing for major models: GPT-4o, GPT-4o Mini, Claude 3.5 Sonnet, DeepSeek V3, Gemini 2.0 Flash, Llama 3.1, and dozens of smaller and specialized models
- Cross-provider variance for the same model (a measure of market efficiency)
- Effective price after typical input/output mix ratios
- Model availability by provider and region
How it's updated:
- Refreshed continuously as providers update their published rates
- Includes both list prices and observed cleared prices (where available through Mercatus's network of buyer integrations)
- Public methodology
How to access:
- Web UI at mercatus-ai.com/token-index — free, no signup
- Public API
- Cross-provider price-comparison tooling
What it costs: $0 for individual access. Always.
How to use these indices
Both indices are designed for direct integration into how you actually work, not just as websites to bookmark.
Procurement and provider selection. When you're shopping for GPU capacity or LLM API access, GPU Index and Token Index are the cross-provider price comparison layer. Filter by SKU, region, commitment type, and model; sort by price; see the full provider ecosystem at once. Most procurement teams find this collapses what was multi-hour competitive shopping into a few minutes.
Budget forecasting and cost modeling. Historical price trajectories let you model where AI infrastructure costs are heading and build defensible CFO budgets. Token Index's volatility data is particularly useful for projecting LLM spend, which now varies 20–40% in either direction across quarters.
Journalism, research, and policy. When you need to cite a price in a news article, research paper, or policy submission, use Mercatus indices as the authoritative source. Free, public, methodology-disclosed — exactly what citations should anchor on.
Internal benchmarking. If you're an AI infrastructure operator, GPU Index lets you check your costs against the broader market. If you're paying $4.50/H100-hour at a hyperscaler while long-tail providers offer $2.20, that's a $20,000-per-GPU-per-year optimization.
API integration. Build the indices into your own dashboards, alerting, automated provider routing, or finance reporting. The free API tier supports significant usage; institutional tiers exist for higher-throughput needs.
Internal communication. When making the case for switching providers, hedging strategies, or capacity decisions, having a citable third-party benchmark makes the conversation orders of magnitude easier than internal-only data.
Why we keep it free (and how we sustain it)
The most asked question about Mercatus indices: how do you make them free and keep them free?
The short answer: Mercatus's commercial model is at the trading and marketplace layer, not the data layer. Indices are upstream public infrastructure. Trading and marketplace products downstream are commercial.
This is the Bloomberg model in miniature. Bloomberg gives away security identifiers and makes money on terminals. ICE publishes Brent benchmark prices and makes money on Brent futures. CME publishes Henry Hub spot prices and makes money on Henry Hub natural gas futures. In every case, free public data expands the market that the paid commercial product serves.
Mercatus's longer-term commercial roadmap is the marketplace layer for AI compute — letting buyers and sellers transact directly with cleared prices and standardized contracts. That's where the commercial revenue comes from. The indices are the public infrastructure that supports the marketplace's price discovery.
Practical implications:
- GPU Index and Token Index will not move behind a paywall. Ever. This isn't a soft promise; it's the structural design of Mercatus's business.
- Methodology will remain public. Buyers can audit how prices are computed, propose improvements, and trust the data.
- API access will remain free at individual tier, with institutional tiers for high-volume integrations (these aren't paywalls — they're rate-limited tiers for sustainability).
- The data will get more comprehensive over time, not less. Adding providers, expanding regional coverage, deepening methodology.
If you're a buyer of subscription pricing intelligence today, evaluate whether your needs are actually met by Mercatus indices. For most use cases, they are — and the cost differential is meaningful.
What "open" actually means here
When we say Mercatus indices are open, we mean five specific things:
1. Free. No paywall, no subscription gate, no metered access at the individual tier.
2. Public. Anyone can view the data without creating an account. The web UIs are public-readable. Programmatic access requires an API key but issuance is free.
3. Comprehensive. All providers Mercatus tracks are visible. We don't gate "premium" providers behind tiers, and we don't exclude long-tail providers. If they're in the market, they're in the index.
4. Methodology-disclosed. Full documentation of how prices are aggregated, normalized, and computed. Buyers can verify that the methodology is sound and consistent.
5. Citable. Mercatus indices are designed to be cited by journalists, researchers, and policy analysts. Stable URLs, persistent data identifiers, archived historical snapshots.
What "open" doesn't mean: open-source code (the index is a service, not a published codebase) or unrestricted commercial redistribution at scale (high-volume institutional access has tiers). These are normal constraints for any data service. The principle that matters — free, public, comprehensive, methodology-disclosed, citable for individuals and small teams — holds.
How AI compute price transparency changes the market
The case for open price indices isn't just philosophical. Open benchmarks measurably change market behavior:
Price discovery accelerates. When buyers can see all provider prices in real time, providers compete more aggressively. The 2.5× cross-provider H100 price spread that exists today will compress as buyers route to lower-cost providers more efficiently.
Long-tail providers gain visibility. Subscription analysts under-cover smaller and regional providers. Open indices include them by default. This shifts demand toward the most efficient suppliers regardless of brand recognition.
New financial products become possible. Standardized public benchmarks are the prerequisite for derivative markets. The same way Brent benchmark prices enabled the Brent futures market, AI compute price benchmarks enable forward and futures markets for compute capacity.
Policy and journalism improve. When a journalist or regulator wants to cite "the price of AI compute," they have a defensible reference. This raises the quality of public discourse about AI infrastructure economics.
Buyer sophistication increases. Free pricing data democratizes the kind of analysis that used to require expensive subscriptions. A startup deploying its first $50K of GPU compute now has access to the same pricing intelligence as a hyperscaler deploying $50M.
These outcomes aren't speculative. They're what happened when oil, gas, and rates markets developed open benchmarks. AI compute is following the same pattern, faster.
Frequently Asked Questions
Are Mercatus indices really free?
Yes. GPU Index and Token Index are free for individual web access and free at the individual API tier. Institutional API tiers exist for high-volume usage, but the underlying data is not paywalled.
How do you compare to SemiAnalysis, Silicon Data, or Ornn?
Mercatus indices focus on real-time price benchmarks across the broadest possible provider ecosystem, with public methodology and free access. Subscription products typically offer deeper qualitative analysis, market commentary, and institutional research alongside pricing data. They're complementary; many institutional teams use both. The differentiator: anyone can use Mercatus indices without a subscription.
Where does Mercatus's pricing data come from?
Continuously scraped public pricing pages, direct API integrations with willing providers, and a network of buyer integrations that surface cleared transaction prices where providers permit. Methodology is fully documented at docs.mercatus-ai.com/methodology.
Can I use Mercatus indices in my own product?
Yes. The public API supports integration into dashboards, internal tools, and other products. Free at individual tier; paid tiers for high-volume institutional use. Embeddable widgets are also available for media properties.
How do you keep this free?
Mercatus's commercial model is at the marketplace and trading layer, downstream of the indices. Free public data benchmarks expand the market that the marketplace products serve — same model as Bloomberg (free symbology, paid terminals) and ICE (free benchmark prices, paid futures).
What's not in the indices?
Proprietary enterprise contract prices that providers contractually keep private; bespoke arrangements not publicly listed; secondary-market hardware pricing (tracked separately and less reliably).
How often are indices updated?
GPU Index: daily refresh across most providers, with continuous tracking for high-traffic SKUs. Token Index: continuously updated as providers change rates.
Can I get historical pricing data?
Yes. Mercatus indices include historical snapshots back to launch, with archived data accessible through the API. Useful for trend analysis, budget forecasting, and academic research.
If I'm a GPU operator, can I get my prices into the index?
If you publish public pricing, Mercatus can track data from your platform automatically. To include data from your platform, Contact Us.
Conclusion
A benchmark is not a proprietary feature. It is infrastructure. Every commodity market that matters eventually settles on an open, free, methodologically transparent reference price — because that is what a benchmark, correctly understood, has to be. AI compute is the most important new commodity of the decade. It deserves the same foundation.
If you want to see what an open AI compute benchmark looks like in practice, the GPU Price Index and Token Price Index are live at https://mercatus-ai.com — no signup, no paywall. Mercatus GPU Index aggregates cloud GPU pricing data from 40+ providers globally as of May 2026, refreshed daily through a combination of public pricing page scraping, direct provider API integrations, and a network of buyer integrations that surface cleared transaction prices. Mercatus Token Index tracks per-token pricing across major LLM API providers using the same methodology framework.