New York, NY,
Atomic Answer: Goldman Sachs has released a new valuation framework treating compute power as a tradable financial asset similar to oil or gas. This shift forces enterprises to reclassify GPU ownership as digital oil fields, fundamentally changing how AI infrastructure is depreciated on corporate balance sheets.
A hyperscale data center can lose millions in value before any server rack actually fails. This is a tough reality for AI infrastructure investors. Top GPUs now become outdated faster than traditional accounting models expect. For example, a chip bought for AI training in early 2025 might already face price drops by 2027 as new architectures set higher standards. This volatility is why the Goldman Sachs and AI Index is getting attention in finance and cloud infrastructure circles.
The bigger challenge is not just hardware demand, but how to value it. More financing institutions now see compute infrastructure as a dynamic commodity linked to AI production, not just as fixed equipment. This shift puts compute power valuation at the heart of today’s capital allocation strategies.
Why AI Infrastructure Accounting No Longer Fits Traditional Models
For years, enterprise hardware followed predictable depreciation cycles. Servers aged slowly, and productivity improved bit by bit. AI systems have changed that pattern.
Modern GPU clusters can lose their economic value long before they actually stop working. The market penalizes slower speeds, higher energy use, and less efficient training. Because of this, GPU depreciation now looks more like falling commodity prices than traditional asset aging.
This creates challenges for the company’s balance sheet.
A cloud provider that spends 12 billion dollars on AI infrastructure cannot use old accounting rules meant for office servers or networking gear. The financial risk grows further when companies lease computing capacity through multi-year contracts that depend on changing AI demand.
This is why the proposed Goldman Sachs AI Index matters. It aims to establish a standard for measuring the economic productivity of AI compute assets across cloud providers, enterprise setups, and infrastructure markets.
Many analysts now compare this situation to energy trading.
The Rise Of Compute As A Commodity Market
Oil changed global finance once markets standardized how energy was produced, delivered, and traded through futures contracts. AI infrastructure may be reaching a similar turning point.
The idea of digital oil fields fits this analogy well. Instead of getting petroleum from the ground, large tech companies and national AI projects now get economic value from compute-heavy infrastructure.
In this framework, GPUs, network bandwidth, cooling systems, and inference throughput are all seen as measurable production assets.
The implications of the AI CapEx strategy are significant.
Companies can no longer judge infrastructure just by ownership cost. They also need to estimate future productivity, demand in secondary markets, how energy efficiency drops over time, and when to replace equipment.
Imagine two cloud AI cloud providers. One uses older, cheaper GPU clusters. The other invests in the latest technology, which costs more upfront but is much more efficient. Traditional accounting might favor the cheaper option, but a compute productivity approach could prefer the newer one since output per watt matters most for long-term value.
That distinction reshapes infrastructure investment behavior.
How Compute Power Valuation Could Restructure Capital Markets
The growing focus on valuing computing power suggests that more institutions want to turn AI infrastructure into a tradable financial asset.
Private equity funds, sovereign wealth funds, and AI infrastructure investors now want exposure to AI demand without investing directly in consumer apps. Computers give them that opportunity.
A structured index based on compute productivity could one day work like shipping indexes, semiconductor benchmarks, or energy futures markets.
This is where the idea of compute futures comes in.
A futures market for compute resources could let companies protect themselves against future shortages, price spikes, or capacity limits. Instead of rushing to secure GPUs during busy periods, they could lock in future compute access through standard contracts.
Turning compute into a financial asset might seem abstract now, but similar systems already exist throughout cloud reservations and long-term infrastructure deals.
The main difference is the scale and standardization of these new markets.
The new Goldman Sachs framework for treating compute power as a tradable asset, 2026, shows that Wall Street increasingly thinks AI infrastructure markets need more advanced financial tools.
Why AI CapEx Strategy Is Shifting Toward Asset Liquidity
In the early years of the AI boom, companies spent heavily on infrastructure and rushed to buy GPUs, no matter how efficiently they used them. Now, that approach is starting to shift.
Investors now want clearer returns from AI infrastructure spending.
A corporation operating thousands of underutilized accelerators faces the same problem as an energy producer with idle drilling equipment: capital inefficiency. The next generation of AI CapEx strategy will likely focus less on raw infrastructure accumulation and more on monetizable compute productivity.
This could lead to secondary markets where companies trade unused processing power with each other.
These changes affect more than just tech companies.
Banks that finance data centers, insurers who cover infrastructure risks, and governments funding national AI projects all need better ways to price their exposure to compute assets. The Goldman Sachs AI Index aims to meet that need.
The Risks Behind Financializing Compute Infrastructure
Commodity-style markets bring both more efficiency and more volatility.
If compute assets become widely traded, price swings could get worse during AI demand spikes or hardware shortages. Smaller companies might have trouble competing with big institutions that can lock in long-term compute contracts.
Another challenge is the rapid pace of technology change.
Unlike oil reserves, computing infrastructure is always changing. New chip designs can quickly lower the value of existing equipment. This makes GPU depreciation much faster than with traditional industrial assets.
Energy use adds more uncertainty. Data centers now face more environmental scrutiny, especially in areas with limited electricity or water for cooling.
All these factors make it harder to value computer assets over the long term.
Still, the overall trend is clear. AI infrastructure is becoming an economic asset class that shares features with energy production, logistics, and commodity trading, compute liquidity, infrastructure efficiency, and asset productivity with the sophistication of a global financial institution.
Executive Procurement Checklist
- Financing: Negotiate GPU leases based on “Compute Futures” rather than static hardware costs.
- Infrastructure Risk: High volatility in compute spot-pricing could impact startup burn rates.
- ROI Implications: Shifts AI from a “Cost Center” to an “Appreciating Asset” in certain sectors.
- Action Step: Consult with finance leads to re-evaluate the depreciation schedule of your Blackwell clusters.
Source: Analysis and perspectives on the global economy and markets from across Goldman Sachs
