When will the GPU futures market have to wait longer for the commodification of computing power?

null

Author: Caleb Shack, Alana Levin

Translation: Jiahui, ChainCatcher

At Variant, we are passionate about exploring emerging markets. Emerging asset classes, financial products, asset issuance, expanded market access, and innovative participation methods are deeply rooted in our founding DNA.

Recently, we have been contemplating markets built around computing power.

Acquiring computing power is a vast and continuously growing field, which can be said to already possess conditions for further financialization.

However, the supply and demand dynamics of computing power are highly complex, opaque, and constantly evolving. Many mysteries remain regarding market timing, structure, and even what the specific traded assets are.

In the process of debating and exploring these issues, we want to share an emerging analytical framework as a window into thinking about the computing power market.

The birth of a new futures market typically requires five prerequisites:

Fragmented supply side

Continuous price volatility

Some form of physical settlement infrastructure

Standardized, tradable units

Lack of substitutes for price discovery or hedging

Our framework examines the current landscape of the computing power market along these five dimensions. We use historical analogies to explain the importance of each dimension and predict when the market might reach a critical point of explosion.

Key points summary

A quick glance at this framework reveals that today’s computing power market still lacks the maturity needed to sustain a robust futures market.

(That said, the market is vibrant, with many startups actively working to change the status quo; if you are doing so, please contact us!)

Below are our current ratings of the computing power futures market across the five dimensions:

Fragmented supply side: 🔴 Supply is highly monopolized by mega cloud providers

Price volatility: 🟢 GPU prices are highly volatile

Physical settlement infrastructure: 🟢 OTC brokers already have physical settlement infrastructure

Standardization: 🔴 Computing power lacks standardized, tradable units

Lack of substitutes: 🟡 Vertically integrated suppliers can hedge internally; other participants are forced to go long

1. Fragmented supply side (Computing Power Score: 🔴)

Futures markets are mechanisms for price discovery.

Under monopolized supply, price discovery becomes unnecessary because prices are set by a few large suppliers, eliminating any pricing uncertainty.

Historically, this situation is common.

Oil futures only gained strength after supply-side cartels (like the “Seven Sisters,” the seven multinational giants dominating global oil in the mid-20th century) weakened.

Electricity markets formed after deregulation, breaking monopolistic pricing and allowing independent producers to enter. Fragmentation of supply drove futures markets to become key venues for price discovery.

Looking at today’s computing power dynamics, supply appears relatively concentrated.

The four major cloud giants (such as AWS, Azure, GCP, Oracle) control about 78% of global self-built critical IT power capacity and approximately 69% of H100 supply (assuming 12.4 million H100 units in Q4 2025).

From this, we infer they also dominate the global supply of compute hours. Supply does not appear fragmented.

Nevertheless, we are still considering factors that could change this dynamic.

New cloud providers are emerging. New chip architectures create opportunities for other suppliers to gain market share.

Some long-term contracted capacities from major labs may ultimately go underutilized, meaning these labs could eventually become market suppliers or sellers of computing power.

Therefore, although we are uncertain about future concentration levels, our current judgment is that the development of the supply side will trend toward greater fragmentation than now.

2. Price volatility (Computing Power Score: 🟢)

Ornn H100 Index on Bloomberg Terminal

Another prerequisite for futures markets is that the underlying asset exhibits high volatility.

Without significant price uncertainty, hedgers lack motivation to hedge against fluctuations.

Volatility also attracts speculators, who profit from large price swings. If the market is stable or predictable, speculators will look elsewhere.

This was evident in the oil markets of the 1950s.

At that time, due to oversupply, the Soviet Union set prices below the “Seven Sisters” listed prices. The “Seven Sisters” then lowered prices in the Middle East without informing OPEC or local oil producers.

The resulting chain reaction led to nationalization of Middle Eastern oil, the formation of OPEC, and increased global oil price uncertainty. Later, oil volatility triggered fluctuations in the electricity markets of the 1970s.

Computing power pricing has historically been, and will continue to be, volatile.

The pace of new supply entering the market is uncertain. New chips or data center architectures may improve token efficiency for specific tasks. Demand continues to surge and expand in unpredictable ways.

We are very confident that this prerequisite is now in place.

3. Physical settlement infrastructure (Computing Power Score: 🟢)

For markets to operate efficiently, buyers must be confident they can receive and consume the underlying asset at the specified date and time.

This requires infrastructure support: aggregating supply, ensuring reliable delivery, clearing transactions, handling collateral, and managing settlement mechanisms. These tasks are usually handled by intermediaries or brokers.

In electricity markets, these responsibilities are managed by independent system operators, acting as neutral third parties with quasi-governmental roles.

Currently, the computing power market lacks fully equivalent roles, but our assumption is that: computing power brokers or OTC desks are beginning (and increasingly tending) to assume many of these functions.

Today, brokers are building indices and data aggregation tools around computing power purchase and lease agreements to anchor market prices.

Ornn and Silicon Data have started releasing price data for data center-grade GPUs.

Broker groups are also reaching consensus on contract terms, similar to how SAFE protocols standardized early financing terms. These tools improve the underlying physical settlement infrastructure—before this, much of this coordination was still happening in group chats.

We give physical settlement infrastructure a green score because it lays the foundation for price discovery.

But compared to mature spot markets, it is far from perfect. These purchase activities occur at the infrastructure level, and not all market participants have the right to resell after purchase. We are closely monitoring progress in creating new markets at this layer.

4. Standardization (Computing Power Score: 🔴)

A major challenge for new commodities is the uniqueness and irreplaceability of their units.

Too many variables can disperse liquidity across many markets or cause basis risk to become too high, making them unsuitable for most hedging and delivery needs.

For example, crude oil is measured by density and sulfur content, which vary by origin.

NYMEX found a product-market fit with its WTI index (light, low-sulfur oil) because it locked in a standard that could serve the global upstream market and was even used downstream (e.g., by airlines) for hedging.

Electricity is standardized regionally, considering demand and supply fluctuations caused by temperature, population density, etc.

The computing power market currently lacks a level of standardization that can meet general hedging needs.

The challenge is: one H100 instance is not always equivalent to another.

Factors such as region (and local power input), hardware configuration (components and network), and contract duration (term length) all increase pricing variability.

However, we have seen early signs of standardization, especially when demand comes from the long tail (non-frontier labs).

Unlike training, inference workloads require fewer subtle differences and can run in distributed, non-co-located environments.

If inference supply becomes dispersed among many vendors—for example, open-source weights increasing market share—standardization may naturally emerge.

5. Lack of substitutes (Computing Power Score: 🟡)

This is a subtle but often overlooked point in market formation.

Futures markets are built to serve hedgers. If substitutes with sufficient liquidity and negligible basis risk exist, then substitute contracts will be ignored.

A textbook example is the lack of adoption of jet fuel futures—because WTI and other upstream indices already sufficiently meet demand.

In the electricity domain, temperature-based futures failed because market participants found hedging the outcome of price fluctuations (electricity) more efficient than hedging the cause (temperature).

Today, model providers hedge computing power risk through long-term leasing agreements or joint ventures, often in a “pay-as-you-go” manner, exchanging spot price risk for counterparty risk.

Mega cloud providers typically own their deployed GPUs outright.

Meanwhile, long-tail suppliers lack the bargaining power for favorable leasing contracts and lack the capital to build their own vertical infrastructure, making them most vulnerable to spot market volatility.

From a market perspective, no true substitutes exist; however, participants controlling supply can internally hedge through vertical integration.

Overall assessment

Based on the scoring card, it is still premature for computing power to support a robust futures market.

While the market has volatility attractive to speculators and early settlement infrastructure to support trading, it lacks supply fragmentation and standardization needed for genuine large-scale price discovery.

Most trading occurs OTC.

Brokers are building price sources, and Ornn and Silicon Data are releasing indices, with group chat trades being formalized into contract templates.

This is not meaningless, but it has yet to evolve into a mature market like WTI or PJM. Trading volumes are too small, contracts too customized, and supply too concentrated, preventing existing infrastructure from large-scale clearing.

The proper way to interpret this framework is as a diagnostic tool rather than a final conclusion. It tells us what is missing, not what is impossible.

Unsolved mysteries

Markets will develop in ways we currently cannot predict.

We have many unknowns and some preliminary hypotheses. These hypotheses are tentative and require further validation or refutation. Below, we present the strongest arguments supporting these hypotheses.

▍In the next 1-2 years, will supply become more fragmented or more concentrated?

We expect moderate fragmentation.

New cloud providers are launching new capacities faster than any other category.

As electricity becomes a core constraint, new regions are being activated, favoring operators who can establish capacity near cheap power (rather than near existing mega cloud footprints).

Fortune 2000 companies are even supporting small data centers. Expansion in this area seems inevitable.

However, the standard business model relies on large, long-term contracts with reliable counterparties (such as mega cloud providers and frontier labs).

Cloud brokers like Hyperbolic and SF Compute are doing the opposite, offering hourly billed capacity.

These companies serve AI-native startups, inference applications running on open-source weights, and long-tail research labs with no frontier-level budgets.

We believe that the adoption of open-source weights will especially lead to further fragmentation of compute capacity—because supply will “de-verticalize” from frontier labs and mega cloud providers.

▍How will standardization unfold?

Index providers are establishing standards around hourly GPU instance costs.

These data sources are rough estimates, not precise prices.

Instance prices vary due to many factors, including region, hardware configuration, and contract duration, making standardization difficult.

Differences in hardware configuration are especially prominent, reflecting how data centers tailor hardware for specific workloads and how mega cloud providers optimize for ecosystem lock-in rather than market uniformity.

When a unified market demand exists, standards will emerge.

WTI standards gained adoption because they serve a broad downstream market for gasoline, diesel, and jet fuel.

Today, demand for compute is driven by AI training and inference workloads.

Training infrastructure is customized and optimized for large, compute-intensive, long-duration tasks in centralized facilities, making underlying compute instances nearly non-duplicable.

In contrast, inference infrastructure requires simpler hardware specs and lower energy consumption; it is optimized for latency, meaning infrastructure is distributed across regions rather than co-located.

Inference is highly homogeneous, and by 2029, it is expected to account for over 65% of AI compute demand. We hypothesize that optimization at the infrastructure layer for this market will lead to more uniform requirements among vendors.

If chip-level instances still differ, hardware benchmarks could be another path to standardization.

NVIDIA has created MLPerf benchmarks to score inference and training performance across various model architectures.

Under this concept, the trading of GPU instances would be based not on hardware specs but on output quality and efficiency.

▍What might hinder the emergence of standards in the next 1-2 years?

We believe “walled gardens” and customized workloads will stifle attempts at standardization.

In the next 1-2 years, mega cloud providers and frontier labs will strive to maintain their dominance in AI infrastructure and model provision.

If they are not decoupled, they will maintain hardware tailored to their needs, which differ across companies. Adoption of new chip architectures will further fragment hardware specs, making standard-setting difficult.

▍How will open-source weights find meaningful applications?

This is the simplest path for the formation of the computing power market.

Today, the core bottlenecks are supply concentration and lack of standardization.

The widespread adoption of open-source weights democratizes inference capabilities.

This, in turn, motivates independent operators and promotes infrastructure optimization tailored to these models.

We see a similar story in Bitcoin mining: open-source software has spawned numerous miners and driven standardization around hardware configurations.

So far, open-source weights have lagged behind closed-source models in performance.

But if this trend continues, open-source weights will soon reach the performance thresholds we see in closed models today.

Companies are already embedding closed models extensively into their systems and witnessing significant productivity gains. Within three months, models that can similarly boost productivity might cost only a small fraction of current prices.

However, most companies will likely still prefer the highest-performing models.

We believe that eventually, frontier closed models will become too expensive for their tasks, and companies will optimize their AI configurations across different models.

Remember, frontier labs currently offer inference services at a loss—they will need to raise prices to sustain operations. At that point, open-source weights will have their moment.

▍What will be the unit of pricing for final transactions?

Computing power can roughly be divided into three layers: chips, chip instance hours, and tokens.

Chip layer—highly concentrated supply.

ASML monopolizes the lithography machines used by TSMC, TSMC monopolizes the chip foundries used by NVIDIA, and NVIDIA monopolizes frontier chip design.

Additionally, chips are only useful if powered and kept online for high availability. This leads us to believe that individual, deliverable chips will not be the final unit of pricing.

Chip instance hours—the actual usable time of a chip.

This is arguably the most valuable state of a chip and the core layer discussed here.

As long as there is sufficient demand for compute resources, compute as a commodity will behave similarly to electricity.

We envision compute being traded like electricity and other utilities: standardized in regional contracts (compute as a function of electricity), with overlays of spot and futures markets for hedging. Under the “chip instance hours” format, this is feasible.

Token layer—the downstream product of compute instances, which could also become the final unit of pricing.

If tokens are the main driver of compute instances, then the token market will provide demand-side hedging and lock in supply-side revenue.

Supply can hedge costs via long-term contracts or vertical integration, maintaining concentration.

However, tokens are not uniform across models. Each model has its own tokenization standards and produces different outputs, making them non-interchangeable across use cases. Nonetheless, we are closely watching developments in this area.