> ## Content Index
> Fetch the complete content index at: https://nexi.fund/llms.txt
> Use this file to discover other available public pages before exploring further.

# Compute Pricing Index: GPU Costs Become Wall Street’s Newest Commodity Benchmark
- URL: https://nexi.fund/compute-pricing-index-gpu-2026/
- Published: 2026-07-24T11:00:21.000Z
- Updated: 2026-07-24T11:00:21.000Z
- Description: Kalshi, CME, and ICE are racing to build a futures market for GPU compute. Babbage Index launched five pricing indices covering the full AI infrastructure stack. Here is what the financialization of compute means for investors.
- Author: Nexi.fund Labs
- Tags: AI & Infrastructure, #mode-6, #hook-number, #track-F

🎯

GPU compute now has price indices. Five of them at Babbage Index alone — tracking GPU throughput, CPU cycles, database memory, storage, and LLM inference. Kalshi launched the first CFTC-regulated compute forward curve on July 14\. CME and ICE have announced compute futures. The financial plumbing for a trillion-dollar market is being built in real time.  
  
Compute is the largest new commodity of our time — and it has had no price discovery mechanism. That is changing this year.  
  
This piece maps the four venues racing to price compute, the three settlement indices they depend on, and what the emerging term structure tells investors about AI infrastructure costs through 2027\. 

Five indices. Four exchanges building forward curves. Three settlement benchmarks. One commodity that did not exist as a tradeable asset two years ago. GPU computing power — the silicon that runs every large language model, every inference call, every AI training run — now has the financial infrastructure of a real commodity market. Kalshi, a CFTC-regulated exchange, launched the first market-driven forward curve for GPU compute on July 14\. Babbage Index, founded in London in 2026, publishes daily pricing across five compute layers. CME Group and Intercontinental Exchange have announced compute futures. Architect Financial, founded by former FTX US president Brett Harrison, is building perpetual swap venues. None of these existed 18 months ago. The speed of institutional plumbing being laid says something about the scale of capital at stake. Hyper scalers alone have committed $500 billion to $600 billion in AI infrastructure spending for 2026\. That is capital that needs to be priced, hedged, and financed. And the instruments to do that are arriving — not in pilot programs, but on Bloomberg terminals and exchange dockets.

## The index layer: three benchmarks, one emerging standard

The foundational question for any commodity market is simple: what is the reference price? For GPU compute, three answers have emerged. \*\*Ornn Compute Price Index (OCPI)\*\* is the incumbent. Launched in 2025 and added to the Bloomberg Terminal in April 2026 under ticker ORNNH100, OCPI is built from printed transactions — actual trades, not scraped listings or survey estimates. It raised a $33 million seed round from a16z Crypto, Galaxy Ventures, and others, and its index now powers three of the four venues building compute derivatives. Over 400 data center operators, investors, and AI companies use its platform. \*\*Silicon Data\*\*, backed by proprietary trading firm DRW, publishes daily GPU benchmark indices based on quoted rental rates. CME Group chose Silicon Data as the settlement index for its compute futures, announced May 12\. \*\*Babbage Index\*\* launched in London in 2026 with a different thesis: cover the full compute stack, not just GPUs. Babbage publishes five daily indices — BI-G (GPU), BI-C (CPU), BI-M (managed database memory), BI-S (storage), and BI-T (LLM inference tokens) — each rebased to 1000 at launch. The GPU index stood at 1421.1 on July 19, reflecting a 39.3% surge in dollar-per-TFLOP-hour over the prior 30 days. Each index is auditable by construction: the methodology and components are open, and every headline recomputes from source data in CI on every pipeline commit. The benchmark race is quietly resolving. Kinetic Alpha, a research firm tracking the compute complex, notes that three of the four derivatives venues now anchor on Ornn-family indices. Only CME settles elsewhere, on Silicon Data's quote-based assessment. When the same reference price underlies both the regulated futures and the crypto-style perps, that index becomes the Brent of compute.

## The venue race: four approaches to pricing the same thing

\*\*Kalshi\*\* launched compute forward curves on July 14, 2026\. These are not futures in the traditional sense — they are binary threshold ladders that resolve against Ornn's OCPI print at weekly and monthly expiries. The implied forward price is read from the strike where the probability crosses 50%. Live launch-day data showed the H100 SXM implied forward at $2.52 for the July 31 monthly expiry and the B200 weekly at over $7\. Its edge is regulatory clarity. It is a CFTC-designated contract market, available to US traders. Its chief risk officer, Udesh Jha, spent 16 years at CME Group before joining it to build the compute product — a direct competitive signal. CEO Tarek Mansour has stated publicly that "compute futures will dwarf oil futures." \*\*CME Group\*\* announced compute futures on May 12 in partnership with Silicon Data. The contracts, pending CFTC review, reference the Silicon Data H100 Rental Index and target a late 2026 launch. CME brings institutional liquidity, clearing infrastructure, and the brand that defined commodity derivatives for a century. \*\*Intercontinental Exchange\*\* followed on May 19, announcing GPU compute futures with Ornn as index partner. ICE's contracts are dollar-denominated, cash-settled, and reference OCPI across H100, H200, B200, and RTX 5090 hardware. The New York Stock Exchange's parent company moves deliberately — but when it moves, the product typically arrives. \*\*Architect Financial Technologies\*\* (Brett Harrison, ex-FTX US) is building the crypto-native approach: perpetual swaps with no expiry, tradeable on its AX exchange through a Bermuda-regulated subsidiary for non-US institutions. The perps reference the Ornn indices and are already seeing offshore volume. The result is a four-venue, three-index, two-continent race to define the price of compute. It mirrors the early days of oil futures in the 1980s, when ICE Brent and CME WTI competed to set the global benchmark.

## What the prices actually say

The GPU compute market is fragmented in ways that make oil look standardized. An NVIDIA H100 rents for $1.65 per hour on Vast.ai's marketplace and $14.19 per hour on Google Cloud. That is an 8.6× spread for the same silicon. The hyperscale-to-neocloud spread for H100s is 5.92×, according to Babbage Index. The US-versus-China frontier inference spread is 9.9× — an arbitrage signal that credit desks and treasury teams are already pricing into cross-border compute contracts. B200 pricing tells a sharper story. Ornn's index showed the Blackwell spot rental price surging 48% between mid-February and mid-April 2026, from $2.75 to $4.08 per GPU-hour. By late June it had fallen to $4.22 — still elevated but trending down as new supply entered the market. Kalshi's forward curve for B200 expiring July 17 implied a price above $7, suggesting the market expects near-term tightness despite the recent decline. For investors, these are lead indicators. Compute pricing captures real-time supply and demand at the infrastructure layer, stripped of the narrative trading and multiple expansion that drive equity prices. A declining B200 forward curve is bearish for the "GPU scarcity" trade and bullish for the AI application companies that consume compute as a primary input.

📊

**Key signals to track**  
  
**Index convergence** — if CME's Silicon Data index diverges from OCPI, the basis becomes a tradeable spread. Tracking which index captures more open interest reveals the market's choice of benchmark.  
  
**Forward curve shape** — backwardation (spot above forward) signals immediate GPU scarcity; contango signals expected oversupply. The shape is a leading indicator for hyperscaler capex cycles.  
  
**Babbage expansion** — Babbage Index's five-layer coverage means it may capture institutional compute exposure that GPU-only indices miss. If Babbage attracts derivative listings, the index race gains a third credible settlement layer.  
  
**CME v Kalshi litigation** — CME is currently suing the CFTC to block its perpetual futures. The outcome determines whether compute derivatives develop on regulated exchanges or offshore perp venues. 

## What this means for institutional investors

The financialization of compute creates three distinct investment signals. First, compute pricing indices are a new alt-data class. A quant fund that tracks Babbage's BI-G GPU index alongside its forward curves has a real-time view of AI infrastructure supply and demand that no equity or bond market provides. OCPI on Bloomberg terminals puts this data on the same desktops that trade WTI and 10-year Treasuries. Second, compute derivatives enable direct exposure to AI infrastructure without owning the equities. An investor who believes GPU supply will remain tight in Q4 2026 can long the H200 forward curve or buy CME compute futures when they list — no NVIDIA stock, no cloud provider equity, no multiple expansion risk. Conversely, a bearish position on GPU supply loosening can short the curve directly. Third, the index methodology itself is an investment thesis. Babbage Index is auditable and open — its methodology rejects the "black box" approach that undermined LIBOR. OCPI is transaction-based, built from actual trades. Silicon Data's index is quote-based, assembled from provider rate cards. Each methodology produces a different price, and that difference is where basis trading will emerge. The market will reward the index that best resists manipulation while capturing real transaction activity. The LIBOR-to-SOFR transition taught the financial industry what happens when a benchmark loses credibility. Compute pricing is being built after that lesson, not before it.

## The competitive subplot: Kalshi vs CME

A pointed rivalry sits underneath the market structure. CME Group is currently suing the CFTC to block Kalshi's perpetual futures. Its chief risk officer Udesh Jha spent 16 years at CME before joining the challenger. That is not a normal career move — it is a competitive defection that signals which side of the trade Jha believes will win. It waited four years for regulatory approval before launching its first product. Its compute forward curves are the result of that patience. Jha argues that prediction markets — with their ability to list many contracts simultaneously across GPU grades and tenors — are structurally better suited to a non-standardized commodity than a single-index futures contract. CME and ICE argue that institutional liquidity requires the clearing and settlement infrastructure only they can provide. Both arguments are probably right for different segments of the market. The question is which segment captures the reference price — the single number that the rest of the market trades around.

$500–600B Hyperscaler AI infra spend 2026 ↑ 62% YoY 

#### Capital at stake

Amazon, Microsoft, Google, and Meta will spend a combined $500–600B on AI infrastructure in 2026, primarily on chips and data centers. That capital needs pricing and hedging infrastructure. · *Kalshi, 247 Wall St., Jul 2026*

As we noted in July, the compute cost curve is becoming an institutional asset allocation signal in its own right. The pricing index layer — the benchmarks, the forward curves, the settlement mechanisms — is the infrastructure behind that signal. The July 14 launch is the first liquid, executable, public forward pricing in the compute complex. It predates both CME and ICE futures, which remain pending regulatory review. For investors, the message is clear: the market is already trading. The question is whether it trades on a regulated US exchange or through offshore perp venues. The answer will determine where liquidity concentrates and which benchmark becomes the WTI of compute. The next 12 months will tell us whether compute pricing indices become a standard institutional data feed or a niche product for hyperscaler treasury desks. The early signal — three competing venues, four derivatives structures, and a CFTC-regulated forward curve within 18 months of the first index launch — suggests the market is leaning toward the former. GPU compute has its Brent moment. It remains to be seen whether it gets its WTI as well.

```

COMPUTE PRICING: VENUE TIMELINE
─────────────────────────────────────────────────────────────
  2025 ────  Apr 2026 ────  May 2026 ────  Jul 2026 ────  H2 2026
  🟢         🟢            🟡            🟢            🔮
  Ornn      OCPI on       CME + ICE     Kalshi       CME/ICE
  OCPI      Bloomberg     announce      forward      futures
  launch    Terminal      futures       curves live  (pending)

```

Ornn OCPI launched 2025, added to Bloomberg Terminal Apr 2, 2026\. CME and ICE announced compute futures within days of each other in May 2026\. Kalshi launched live forward curves Jul 14\. Regulated futures target H2 2026.

## Sources

[ Kalshi builds prediction markets for GPU computing power prices Crypto Briefing's coverage of Kalshi's July 14 launch of compute forward curves — the first CFTC-regulated market-driven forward pricing for GPU compute, covering B200, H200, and A100 chips and settling against Ornn OCPI. Crypto Briefing ](https://cryptobriefing.com/kalshi-gpu-compute-price-prediction-markets?ref=nexi.fund) 

Primary source: Kalshi's July 14 product launch with specific pricing data and regulatory context.

[ Compute Is the New Oil: Kalshi Just Launched a Way to Bet on the Future Price of AI Computing Power 24/7 Wall St. analysis covering the competitive dynamics between Kalshi, CME, and ICE, with context on hyperscaler spending and the CME lawsuit against CFTC over Kalshi's perpetual futures. 247 Wall St. ](https://247wallst.com/investing/2026/07/15/compute-is-the-new-oil-kalshi-just-launched-a-way-to-bet-on-the-future-price-of-ai-computing-power?ref=nexi.fund) 

Context on the Kalshi-CME rivalry and the regulatory landscape for compute derivatives.

[ Babbage Index — AI Infra Compute Pricing Index Babbage Index publishes five daily pricing indices covering the full compute stack: GPU (1421.1), CPU (1016.1), memory (960), storage (976.9), and LLM inference tokens (1000.1). Updated daily, auditable by construction. Babbage Index, London ](https://babbageindex.com/?ref=nexi.fund) 

Live source for five-layer compute pricing with auditable methodology — represents the newest entrant in the compute index space.

[ ICE plans GPU compute futures with Ornn index partner TNW coverage of ICE's May 19 compute futures announcement, including the competitive dynamic with CME, Ornn's OCPI methodology, and the oil futures analogy. The Next Web ](https://thenextweb.com/news/ice-nyse-compute-futures-market-gpu-ai?ref=nexi.fund) 

Coverage of the ICE-Ornn partnership, providing context on the exchange-level race to define compute pricing.