(Bloomberg) — Silicon Information, a market intelligence agency that focuses on the price of graphics processing models, has created what it says is the primary each day index to trace the specialised chips important to powering synthetic intelligence.
The corporate’s Silicon Information H100 Rental Index tracks the hourly price of renting a GPU, in keeping with founder and Chief Government Officer Carmen Li.
“There’s zero transparency now for GPU prices,” Li mentioned in an interview. “This is step one, to assist market members acquire transparency into the convoluted world of AI prices.”
Graphics processing models are the workhorse for coaching AI fashions. The chips are designed to deal with hundreds of duties concurrently, an idea referred to as parallelism. An information heart could use a whole lot and even hundreds of those processors, which might price hundreds of {dollars}.
Based in 2024 and backed by proprietary-trading companies DRW and Leap Buying and selling, New York-based Silicon Information is constructing a platform that goals to convey worth transparency and construction to the GPU market. The objective is to show GPUs right into a benchmarked asset class that may be traded like another monetary instrument, Li mentioned.
The agency sources its pricing information throughout markets and know-how corporations, together with so-called hyperscalers and neoclouds, in keeping with Li. The index is normalized based mostly on quite a lot of components together with GPU sorts, platforms and {hardware} configuration. That’s as a result of GPU costs will be affected by a variety of things, together with their model, efficiency metrics, and {hardware} and software program specs.
Silicon Information’s index can be based mostly on historic information. It recalibrates based mostly on the significance of every issue on the time. Bloomberg, which publishes the index, is the mum or dad of Bloomberg Information.
The first customers of the index are banks that license and distribute it for inner threat administration, hedge funds that license the underlying information, and asset managers or personal fairness companies which might be financing or constructing out information facilities the place GPUs are used.
Although the index itself can’t be traded but, it supplies a benchmark for merchants and others to observe as they might a inventory index. It additionally offers perception into the price of items for these constructing AI merchandise or in want of GPU computing energy.
It’s about “making a clear market and serving to each participant perceive what their infrastructure is definitely price,” Li mentioned.
