The company was born from observing rising AI traffic on data platforms in 2014, leading to the realization that the capital-intensive AI industry would require the same data transparency and financial hedging instruments (indices, derivatives) as traditional financial markets.
Lenders use conservative accounting depreciation for hardware. However, real-world rental income data shows GPUs' economic value (calculated via discounted cash flow) remains significantly higher for much longer, challenging assumptions of rapid obsolescence and unlocking new financing models.
A GPU integrated into a data center (with cooling, networking, etc.) is a high-value, income-generating asset. Its "going concern" value is distinct from and typically much higher than its resale value as a standalone component on a secondary market like eBay. This is crucial for accurate valuation.
Contrary to the "iPhone model" of tech obsolescence, older GPUs like the Nvidia A100 are seeing their rental rates rise. This is driven by explosive demand for AI inference tasks, which don't always require the latest hardware, proving the chips' long-term economic viability and value.
While focus is often on the latest chips, the rental rate of the older A100 GPU serves as a crucial economic indicator. Strong rates for this hardware signal that broad, foundational demand for AI inference is healthy, independent of the demand for cutting-edge training from frontier models.
The dominant use of AI compute is moving from training massive models to running inference tasks. This shift fundamentally alters the market, enabling broader enterprise adoption via cheaper, open models and changing the demand profile for compute hardware beyond the absolute cutting edge.
The CME is creating futures contracts based on H100 and B200 GPU rental price indices. This marks the financialization of compute, allowing providers to hedge revenue and enterprises to lock in future costs, with a long-term vision for physical delivery of compute capacity.
Unlike physical commodities like oil, AI compute lacks strong regional price differences. For non-real-time tasks like model training, the physical location of the GPU and its associated latency are negligible. This allows users to source compute globally, driving prices toward a single international benchmark.
The main constraint on building new data centers is no longer GPU availability, but access to land with permitted, reliable power. The dilapidated US grid, after decades of flat energy demand, is unprepared for the AI boom, forcing developers toward ad-hoc, "behind the meter" power solutions.
A significant portion of current economic growth and employment is driven by capital-intensive AI build-outs, which also contribute to inflationary pressures. This presents a challenge for policymakers: raise rates to slow inflation at the risk of stifling a major technological revolution, or tolerate higher inflation to support it?
Criticisms of AI "hallucinations" often miss the point. The proper benchmark for AI performance is not flawlessness but the alternative: a human analyst who also makes mistakes, gets tired, or uses poor sources. AI's tireless nature and the ability to run cheap, parallel checks can ultimately lead to higher reliability.