Arnold Ventures Co-Chair John Arnold: Computing Power Is Not Yet a Commodity
John Arnold, Co-Chair of Arnold Ventures, believes that computing power is not yet a commodity because the products delivered by one data center are not equivalent to those delivered by another. He proposed two paths to market: either deliver enough standardized products to buyers and sellers, or create a price index recognized by the entire industry.
He pointed out that short-term transactions are already few, and these transactions have even lower transparency. Different chip models, data center designs, and specific load requirements make it difficult to design a physical delivery mechanism and to create a trustworthy index. This statement was highlighted to explain why computing power futures have been stalled at the standardization stage for so long.
Arnold has a background in natural gas derivatives trading, having worked at Enron in natural gas derivatives before founding Centaurus. After exiting the hedge fund in 2012, he turned to philanthropy and policy. He has frequently analyzed AI infrastructure through the energy curve, stating that current near-term computing power is tight, with the curve showing backwardation, and the expected cost of new capacity is lower than the current spot price.
Exchanges are already competing for index entry. CME Group and Silicon Data plan to launch computing power futures based on H100 and B200 leasing indices, while ICE and Ornn have also announced contracts settled by trading-based computing power price indices, all pending regulatory review. Bloomberg has listed some leasing index codes, but the sampling methods and data center stratification are not uniform across firms.
BCG estimates that the AI computing power market is expected to grow from about $360 billion in 2025 to about $2.3 trillion by 2030, while listing the lack of consensus benchmarks as a liquidity barrier. Reports from buyers like Grayscale indicate that large-scale cloud providers still control about 70% of computing power capacity, with spot markets primarily dominated by long-term contracts and customized clusters, and spot short orders are insufficient to support continuous delivery.
In terms of market mechanisms, buyers are model companies and cloud providers looking to lock in training and inference costs, while sellers are large-scale clouds and new clouds holding powered capacity. The event driver is a capital expenditure race colliding with the inability to interchange data center quality. Funds continue to flow to powered cabinets that can sign long-term contracts, rather than to untrusted futures curves; those benefiting are holders who can package and sell chips, networks, and electricity, while those under pressure are trading desks trying to hedge overall market risk with a single index.
Source: Public Information
ABAB AI Insight
Arnold's assessment of computing power is similar to how he viewed natural gas back in the day: first asking whether the delivery item is the same, then questioning whether there are enough short orders to expose the price. During his time at Enron, he relied on pipelines, storage capacity, and nominated delivery points; later at Centaurus, he profited from the curve and basis during the Amaranth collapse, rather than slogans. He applies the same standards to data centers, concluding that H100 hours under different interconnection and cooling conditions are not the same commodity.
Thus, the capital path is stalled at "first create an index, then create futures." Silicon Data and Ornn are competing for pricing power positions like Platts and Argus, while CME and ICE are vying for listing rights. The motivation is not to immediately replace long-term contracts from Nvidia or Microsoft, but to provide cash settlement exits for clusters that cannot be relocated. Whoever's index is written into settlement terms will take away the benchmark rents in the energy market.
The analogy is with electricity node pricing, not Brent crude oil. Electricity cannot be stored, and is also split due to nodes and congestion, ultimately relying on node pricing and differential contracts to convert physical differences into basis. Computing power is still stuck before this step: there is a leasing index, but no widely accepted delivery point. Founders of new clouds like CoreWeave, who also come from commodity trading, openly state that they are temporarily non-interchangeable, indicating that sellers are also unwilling to have their data centers treated as standard barrels.
The structural judgment indicates that the transfer of pricing power is not yet complete. Chip manufacturers and large-scale clouds still price using long-term contracts and bundled services, while futures aim to transfer pricing power to exchange indices. The mechanism is: without enough short orders and deliverable standards, the index is merely a survey price; survey prices cannot force buyers to accept settlements that do not match their data center quality, and commoditization will remain at the promotional level, with real prices locked in customized contracts.
ABAB News · Law of Cognition
Non-interchangeable items cannot become commodities even with an index.
Too few short orders mean futures are just survey questionnaires.
Pricing power lies in settlement terms, not at press conferences.