Silicon Data, a company providing market intelligence and performance benchmarking for graphics processing units used in artificial intelligence compute, announced Thursday it raised $4.7 million in seed funding co-led by DRW and Jump Trading Group.

Woodside AI, Wintermute Ventures, Sancus Ventures and SoGal Ventures also participated in the round.

As AI adoption has grown, companies have discovered themselves in increasing need of timely insights about the cost of GPUs and compute resources. To tackle this challenge, the company has built a product suite that aggregates 3.5 million market data points across a wide selection of 50 GPU chipsets and platforms.

“The demand for compute continues to accelerate as this once-in-a-generation technology grows,” said Chief Executive Carmen Li.

The company announced the launch of several AI-powered tools that will assist enterprise businesses in understanding the market and drawing insights into GPU pricing, benchmarking and carbon footprint that will allow them to optimize for price and performance.

SiliconNavigator is a GPU pricing market intelligence platform that includes comprehensive insights into specifications, historical pricing and current market trends. Users can derive analysis from metrics in both retail and rental pricing history. That will allow them to optimize their procurement strategies and make direct technical comparisons for performance specifications such as floating point operations per second per watt.

SiliconMark provides a performance benchmarking solution that measures GPU performance metrics over time to help businesses understand how they will degrade and operate under real-world AI workloads. It also offers predictive maintenance insights based on historical data.

Silicon PriceIQ uses AI-driven forecasting of GPU market trends based on chips, adjusted by region, GPU configuration and rental terms to help users optimize purchasing decisions. SiliconCarbon helps businesses make sustainability decisions based on carbon footprint estimates by calculating emissions based on GPU type.

GPU market costs and rental prices are hard to predict and over the last year, experts have warned that price volatility should be expected. Supply depends on multiple factors, including manufacturer, type and fit-for-purpose. The most powerful GPUs are used for training and fine-tuning AI large language models such as OpenAI’s GPT and Meta Platform Inc.’s Llama, and others are used to deploy LLMs in runtime, otherwise known as inference.

With the recent release of China-based DeepSeek’s R1 step-by-step “reasoning” capable AI model in early 2025 there has been a rush by AI companies to train and release new models that follow this new paradigm. That has created a gold rush for GPUs. Late last month, Sam Altman, CEO of OpenAI, said that his company had run out of GPUs, thus delaying the launch of the company’s next-generation model series GPT-4.5.

“Silicon Data’s data aggregating capabilities would also make the development of compute market indices – which would track the price performance of a basket of GPU platforms in a similar fashion to a traditional financially settled commodity index – a very real possibility,” said Li.

The new funding will support product expansion, especially to fuel the scaling up of AI-driven performance benchmark insights from SiliconMark and market trends intelligence from PriceIQ. The company also said it intends to expand collaboration with cloud providers, exchanges and enterprises to provide better market penetration across AI, high-performance computing and finance.

“Our goal is to build the world’s most comprehensive compute intelligence platform,” said Li. “With this funding, we’re enhancing transparency, efficiency, and sustainability in compute markets, ensuring companies can make smarter, data-driven infrastructure decisions.”

In November 2024, Li spoke with John Furrier, executive analyst at theCUBE Research, during the NYSE Wired Cyber & AI Innovators Summit about Silicon Data’s capabilities and the company’s future in an episode of theCUBE Podcast. Here’s the full interview:

Image: SiliconANGLE/Microsoft Designer