AI Compute Market: Why Computing Power Is the New Oil

Artificial Intelligence is transforming industries, businesses and the global economy, creating an unprecedented demand for the computing power needed to train and operate advanced AI models. Once used primarily as a verb, the word “compute” has increasingly become a noun referring to the processing capacity generated by powerful computer chips. Today, this resource is emerging as one of the most valuable inputs in the Technology industry, prompting comparisons with oil and other major global commodities.

According to Boston Consulting Group (BCG), approximately $360 billion was spent on AI compute worldwide last year. That figure was around 50% higher than spending on copper, illustrating the growing economic importance of computing power. BCG expects transactions involving AI compute to reach $2.3 trillion by 2030, a scale comparable to the global trade in crude oil today.

Despite this rapid expansion, the market for AI compute remains fragmented and relatively immature. Large technology companies can negotiate computing deals worth tens or hundreds of billions of dollars, while thousands of smaller businesses struggle to find reliable capacity at transparent prices. Buyers and sellers also lack many of the financial instruments that established commodity markets use to manage price fluctuations and reduce risk.

That imbalance is creating opportunities for new exchanges, financial technology startups and established trading firms. From marketplaces listing available graphics processing units (GPUs) to proposed futures contracts and privately negotiated derivatives, businesses are attempting to transform computing power into a standardised, tradable commodity.

The ambition is substantial: create a market that makes AI compute easier to buy, sell, price and finance. If successful, these efforts could change how companies access the infrastructure needed to develop and operate artificial intelligence.

Why AI Compute Is Becoming a Global Commodity

AI systems depend on enormous amounts of processing power. Training advanced models, running applications and responding to user prompts require access to specialised chips and the infrastructure needed to operate them. As demand for AI services grows, businesses increasingly compete for access to this computing capacity.

The resulting demand has helped make compute a valuable resource in its own right. BCG’s estimate of approximately $360 billion in global spending last year highlights the scale of the market, while its projection of $2.3 trillion in transactions by 2030 suggests that the sector could become considerably larger.

The comparison with oil is based on more than the size of the potential market. Like energy, computing power is an essential input for a growing range of economic activities. Businesses that cannot obtain enough compute may struggle to train AI models, operate applications or expand their services.

However, compute differs from a traditional commodity in important ways. A barrel of crude oil can generally be described using established measurements, even though its quality and delivery terms vary. Computing capacity is more complicated because its value depends on the type of chip, the server configuration, the location of the equipment, the duration of access and the particular workload a customer needs to run.

These differences make it harder to create a single, transparent market price. A business seeking access to one type of GPU in a particular data centre may not be able to substitute another configuration located elsewhere without affecting cost, performance or operational requirements.

As a result, the market needs mechanisms that can account for these differences while making transactions more accessible and predictable.

How the AI Compute Supply Chain Works

The supply chain begins with semiconductor manufacturers such as Nvidia and AMD, which produce advanced chips used in AI computing. Equipment manufacturers, including Dell and Hewlett Packard Enterprise, integrate these chips into servers and racks that can be installed in data centres.

Data-centre operators then provide the infrastructure needed to run the equipment and make computing capacity available to customers. These operators include large cloud-computing companies, independent infrastructure providers and numerous smaller businesses.

The largest cloud providers are commonly known as hyperscalers. Amazon, Google and Microsoft are among the major companies in this category. They operate extensive computing infrastructure and can negotiate large agreements to secure the capacity needed for their own services and customers.

Another group consists of independent providers known as neoclouds. Companies such as CoreWeave and Nebius focus on providing computing infrastructure and capacity to customers that need access to specialised hardware.

These operators typically rent computing capacity for a specified period, often under contracts lasting several years. Customers can include large cloud providers supplementing their own infrastructure, AI laboratories developing advanced models and startups running open-source AI systems.

In principle, the arrangement resembles a conventional supply chain: chipmakers produce the hardware, equipment manufacturers assemble it, data-centre operators provide access, and customers pay to use the resulting computing capacity.

In practice, however, the variety of hardware, locations, rental periods and contractual terms makes transactions difficult to standardise. This complexity is one of the main reasons market participants are looking for more formal systems to connect buyers and sellers.

Why Buying and Selling Compute Remains Complicated

The AI compute market is highly fragmented. Different chips have different capabilities, and the same type of hardware may be available in different server configurations and geographical locations. Capacity may also be offered for different periods, ranging from short-term rentals to multi-year commitments.

These variations make it difficult to compare offers directly. A buyer must consider not only the price but also the equipment being offered, where it is located, when it will be available and whether it meets the requirements of the intended workload.

Many businesses therefore rely on employees or specialised intermediaries to locate computing capacity. Informally known as a company’s “compute guy,” the person responsible for securing capacity may purchase available resources through networks of brokers and other middlemen.

Some of these transactions take place through messaging applications such as WhatsApp and Telegram. Offers can circulate through opaque networks, making it difficult for buyers to assess prices and compare competing providers in a consistent way.

This system differs considerably from the transparent trading arrangements found in established commodity markets. In mature markets, standardised contracts, public price information and formal trading venues help participants assess value and manage transactions.

The absence of comparable infrastructure in compute creates additional friction. Buyers may struggle to identify available capacity, while providers may find it difficult to reach customers willing to pay competitive prices.

Developing formal marketplaces could make the process more efficient by listing available hardware, rental terms, prices and locations. Such platforms would not eliminate every technical difference between computing resources, but they could make the market easier to navigate.

Long-Term Contracts Create Financial Risks

Because the compute market remains fragmented, providers and customers often prefer long-term contracts. For buyers, these agreements can help secure access to scarce capacity. For operators, they offer the prospect of predictable revenue that can support investments in servers, data centres and other infrastructure.

However, long-term commitments can become difficult to value when the technology and the market are changing rapidly. Equipment that is highly sought after today may face different competitive conditions in a few years, while customers and providers may experience changes in their financial positions.

Contracts can therefore create risks for both sides. A customer may commit to paying for capacity that becomes less attractive, while an operator may depend on future payments from a business that is no longer able to meet its obligations.

The financial risks are particularly relevant for smaller infrastructure providers, which may need to borrow money to acquire expensive equipment. According to the supplied report, many smaller operators buying and renting compute face double-digit interest rates.

Rising yields on CoreWeave’s bonds have also highlighted uncertainty surrounding the financing of AI Infrastructure. Borrowers and lenders must assess whether future revenues will be sufficient to support substantial investments in computing capacity.

Without reliable market prices and established risk-management tools, lenders may struggle to determine the appropriate cost of financing. The result can be expensive borrowing and uncertainty about which businesses will be able to sustain their operations as the market develops.

New Exchanges Aim to Make Compute Trading More Transparent

A number of companies are now working to develop formal marketplaces for computing capacity. One example is Compute Exchange, which lists GPUs according to factors including location, rental duration, price and availability.

Such platforms aim to make it easier for customers to identify suitable computing resources without relying entirely on private networks of intermediaries. By presenting information in a more structured format, they can help buyers compare available capacity and give providers another route to reach customers.

The longer-term ambition is to build a market in which computing capacity can be traded more efficiently, with clearer pricing and better information about supply and demand.

However, listing GPUs is only one part of the process. A mature market also requires reliable pricing benchmarks, standardised contracts and mechanisms that allow businesses to manage financial exposure when rental prices change.

This is where financial derivatives could become important. In established commodity markets, derivatives allow participants to agree on prices for future transactions or offset the financial impact of price movements.

Companies seeking to develop similar instruments for AI compute hope to provide businesses with greater certainty about future costs and revenues.

CME Group and Intercontinental Exchange Explore Compute Futures

Two major names in derivatives trading, CME Group and Intercontinental Exchange, are looking to enter the emerging compute market through proposed futures products.

Both firms have teamed up with startups developing indices linked to rental prices for popular AI chips. These indices could provide benchmarks for contracts that allow market participants to take positions on future compute prices or manage exposure to changes in rental costs.

The proposed products were awaiting regulatory approval at the time described in the supplied report.

Futures contracts could offer a more formal way for providers and customers to manage uncertainty. For example, a business that expects to need substantial computing capacity in the future may want to protect itself against a sharp increase in rental prices.

Similarly, a compute provider whose business depends on renting out servers may want to reduce the risk that market prices fall below the levels needed to cover its costs.

However, effective futures markets require benchmarks that accurately represent the underlying asset. Because compute varies by hardware type, location and rental period, creating a widely accepted index is a major challenge.

The involvement of established exchanges could help develop the infrastructure needed for standardised trading, but regulatory approval and market adoption remain important steps in the process.

Startups Develop Private Compute Derivatives

While large exchanges work towards launching regulated products, some companies are experimenting with privately negotiated derivatives known as over-the-counter instruments. These contracts are commonly used in Financial Markets, including credit and foreign exchange, to manage exposures that may not fit standardised exchange-traded products.

In March, startup Anera Labs arranged a six-figure pilot trade for June with FalconX, a cryptocurrency broker. The contract was based on the future rental price of Nvidia’s popular H100 GPUs.

The transaction illustrates how financial firms and technology companies are exploring ways to trade exposure to compute prices without waiting for a fully developed exchange-based market.

Over-the-counter contracts can be tailored to the needs of individual counterparties. That flexibility can be useful in a market where computing resources differ considerably and standardised products are still emerging.

However, privately negotiated agreements can also introduce risks. Their value depends on the terms of the contract and the ability of both parties to meet their obligations. Without strong risk-management arrangements, a transaction can expose participants to counterparty and pricing risks.

The development of these instruments suggests that some market participants believe computing capacity is becoming sufficiently valuable and price-sensitive to support financial products similar to those used for traditional commodities.

Luxor Offers Swap-Like Contracts to Compute Operators

Another startup, Luxor, offers operators swap-like contracts designed to provide a more predictable cash flow from their servers.

Under the arrangement described by the company, a contract guarantees a fixed cash flow per server per hour. If the market price for compute falls below the price agreed in the swap, the operator receives the difference. If the market price rises above the agreed level, the operator pays the difference.

Ethan Vera, one of Luxor’s co-founders, explained that payments could be structured through a revenue-sharing arrangement, depending on the contract.

The purpose is to help operators manage uncertainty in the rental market. A business investing in expensive servers may find it difficult to predict the revenue those machines will generate over time. A swap-like agreement can help protect against some price movements by establishing a financial arrangement linked to an agreed benchmark.

Luxor’s bespoke derivatives remain relatively small, with transactions measured in the tens of millions of dollars, according to the supplied report. Nevertheless, the company sees substantial potential as demand for computing capacity grows.

The effectiveness of such products depends on the quality of the pricing benchmarks and the financial strength of the parties involved. If contracts are to become more widely used, market participants will need confidence that the underlying prices are reliable and that agreements can be honoured.

Bermuda Becomes a Base for Compute Futures

While American exchanges await regulatory approval for their proposed products, other companies are developing compute-linked financial instruments elsewhere.

Architect Financial Technologies, founded in 2023, offers futures linked to the value of chips in the United States through its exchange in Bermuda.

The company has also launched a platform where AI model developers bid to provide inference rather than simply offering access to computing hardware.

Inference refers to the useful output an AI model produces in response to a prompt. When a user asks a question and an AI system generates an answer, the process of producing that response is inference.

Brett Harrison, Architect’s founder and chief executive, believes inference may be easier to standardise than compute itself.

The distinction matters because AI services are not valuable only for the processing power used to run them. Customers ultimately want useful results, whether those involve answering questions, generating content or completing other tasks.

A market focused on inference could therefore create a different way to organise transactions around AI services. Rather than trading only access to underlying hardware, buyers and sellers could negotiate around the outputs that AI systems deliver.

However, the supplied report does not provide detailed specifications for the platform or explain how inference contracts are standardised. Its significance lies in the broader attempt to create more structured markets around the growing AI economy.

Why Crypto Entrepreneurs Are Entering the Compute Market

The emerging compute market has attracted entrepreneurs and financial engineers with backgrounds in cryptocurrency trading and digital assets. Several companies and individuals involved in the sector previously worked in businesses that depended heavily on computing power.

Brett Harrison, who founded Architect Financial Technologies, was previously president of FTX US, the American arm of the cryptocurrency exchange founded by Sam Bankman-Fried. Harrison resigned two months before the parent company’s bankruptcy in November 2022, which followed revelations of fraud.

Luxor initially focused primarily on serving bitcoin miners, while Vishwa Naik, one of Anera Labs’ co-founders, previously worked as a cryptocurrency entrepreneur. CoreWeave and some other neocloud providers also began as cryptocurrency-mining businesses.

These backgrounds have created a link between the economics of cryptocurrency mining and the development of AI infrastructure markets. Both activities rely on substantial computing resources, and both have attracted businesses seeking ways to improve the economics of expensive hardware.

Vishwa Naik also previously worked as an oil trader at Reliance Industries, the Indian conglomerate. His experience in the energy market informs his comparison between today’s compute market and the earlier development of the global petroleum trade.

The involvement of entrepreneurs from these industries reflects the experimental nature of the emerging market. Participants are attempting to apply trading, financing and risk-management techniques from other sectors to a resource whose commercial importance is growing rapidly.

Wall Street Trading Firms Move Into AI Compute

Interest in the compute market is no longer limited to startups. Established financial companies are also positioning themselves to participate as demand grows and new trading products are developed.

DRW and Jump Trading, two major market-making firms, have both invested in Silicon Data, a compute-focused research company working with CME on its proposed futures products.

Market-makers help facilitate trading by providing liquidity and quoting prices at which they are prepared to buy or sell assets. Their participation can be important in the development of new markets because buyers and sellers need counterparties willing to transact.

Both DRW and Jump Trading have also assembled teams to trade the emerging compute commodity.

This month, Jump Trading acquired Ito Markets, a startup focused on compute trading. The founders of Ito Markets are expected to lead Jump’s efforts to enter the industry.

The move illustrates how established trading firms are preparing for a market that could become substantially larger if computing capacity becomes more standardised and easier to trade.

Financial firms bring experience in market-making, pricing and risk management, but the compute market presents challenges that differ from those of more established financial products. Its underlying assets are technologically complex, and their value can depend on factors that change quickly.

Why Clearing Houses and Risk Management Matter

Building a functioning market requires more than listing available GPUs or introducing futures contracts. Participants also need confidence that transactions will be completed and that counterparties will meet their financial obligations.

David Olsen, Jump Trading’s investment chief, has argued that the market will develop more rapidly with the right institutions, including clearing houses. These institutions help limit the risk of major failures by ensuring that contracts are honoured by both sides of a trade.

Clearing arrangements are a central feature of many established derivatives markets. They help manage counterparty exposure and provide a framework for processing transactions, reducing the risk that the failure of one participant disrupts other market participants.

Olsen acknowledged that the compute market remains far from having all the necessary structures in place. He also recognised that developing a mature market could take years or even decades.

The challenge is particularly important because computing infrastructure requires substantial capital investment. Operators may borrow heavily to acquire equipment, while customers may enter long-term contracts that depend on future business performance.

Reliable clearing and risk-management systems could help participants assess these commitments and trade with greater confidence. Without them, uncertainty about prices, counterparties and future revenues could continue to limit the market’s development.

Could AI Compute Follow the Path of the Global Oil Market?

Vishwa Naik of Anera Labs compares the current stage of the compute market with the period when a small group of major oil companies dominated the global petroleum trade.

Known as the “Seven Sisters,” these Western oil giants held substantial influence over the industry from the late 1920s through the 1970s. Their dominance was weakened by nationalisation in several oil-producing countries and the emergence of a more open global oil market.

Naik argues that the compute market may be approaching a comparable turning point. Today, access to large quantities of AI computing power is concentrated among major technology companies and large infrastructure providers, while smaller businesses often struggle to secure capacity on favourable terms.

The development of exchanges, pricing benchmarks and financial instruments could gradually change that structure. If more transparent markets emerge, businesses of different sizes may find it easier to buy capacity, compare prices and manage financial risks.

However, the comparison with oil should not be taken to mean that the two markets will develop in exactly the same way. Computing power is technically heterogeneous, and the underlying hardware evolves rapidly. These characteristics may make standardisation more difficult than it is for many conventional commodities.

Governments are also paying increasing attention to AI laboratories and companies developing financial markets around compute. The interaction between technology policy, infrastructure investment and market development could influence how the industry evolves.

The comparison nevertheless highlights the central question facing the sector: whether access to an increasingly essential resource will remain concentrated among a limited number of powerful companies or become available through a more open and transparent marketplace.

What the Future of the Compute Market Could Look Like

The future market for AI compute could include several layers of infrastructure. Digital marketplaces may help buyers locate available hardware, while pricing indices could provide benchmarks for rental rates. Futures and swap-like agreements could allow providers and customers to manage the financial consequences of price changes.

Over time, standardised contracts and clearing arrangements could make transactions easier to compare and potentially attract a broader range of participants.

Smaller AI startups could benefit if they gain easier access to computing capacity without needing the negotiating power of a major technology company. Infrastructure operators could also gain additional ways to finance investments and manage fluctuations in rental income.

Financial firms may find opportunities in market-making, derivatives trading and the development of pricing benchmarks. However, these opportunities will come with risks, particularly while the market remains immature and the future value of computing assets is difficult to predict.

The scale of the opportunity is one reason established exchanges, technology startups and trading firms are already investing in the sector. BCG’s projection of $2.3 trillion in compute transactions by 2030 suggests that the market could become an important part of the global economy.

That projection remains an expectation rather than a guaranteed outcome. Achieving a more mature market will depend on whether participants can establish reliable benchmarks, standardise contracts where possible and develop institutions that help manage financial risk.

Conclusion: AI Compute Needs a Market Built for Its Scale

AI compute is becoming a critical resource for businesses developing and operating artificial intelligence. With approximately $360 billion spent on it worldwide last year, according to Boston Consulting Group, and transactions projected to reach $2.3 trillion by 2030, the market is expanding rapidly.

Yet the way computing capacity is bought and sold remains fragmented. Differences in hardware, location, availability and contract duration make transactions complicated, while the lack of mature financial instruments leaves businesses exposed to uncertain prices and financing risks.

Companies such as Compute Exchange, Anera Labs, Luxor and Architect Financial Technologies are developing marketplaces and financial products intended to make compute trading more structured. Established institutions, including CME Group and Intercontinental Exchange, are exploring futures, while firms such as DRW and Jump Trading are investing in the emerging market.

The development of transparent pricing, standardised contracts and reliable clearing systems could make computing capacity easier to access and finance. However, regulatory approval, technical complexity and counterparty risks remain significant challenges, and a fully mature market may take years to develop.

As AI becomes more deeply embedded in the global economy, the demand for computing power is likely to remain a major commercial force. Whether compute eventually develops into a market comparable in scale and sophistication to oil will depend not only on technological progress, but also on the institutions and trading systems built around it.