Silicon Data’s Pitch to Make AI Costs Predictable

When Chinese AI startup DeepSeek claimed it had trained a GPT-4-class model for just $6 million, the AI community was quick to express skepticism. A 67-billion-parameter model developed for a fraction of what OpenAI or Google might spend? Unlikely, many, including Elon Musk, argued. Yet whether the number was accurate or inflated almost didn’t matter. The deeper problem was that no one could say with certainty what a model like that should cost to build in the first place.

AI has become one of the most resource-intensive technologies in history, but the financial infrastructure underpinning it remains opaque and unstructured. As companies pour billions into training frontier models and building out data center infrastructure, there’s a fundamental lack of transparency around core inputs, especially when it comes to GPUs, the computational engines that power it all. What is the going rate for GPU compute? How does that rate vary by region, cloud provider, or chip model? These are critical questions that many in the industry struggle to answer.

Silicon Data Aims to Make GPU Costs Transparent

This is exactly the challenge Silicon Data is setting out to solve. Founded in 2024 and based in New York, the startup recently raised $4.7 million in seed funding led by DRW and Jump Trading Group, with participation from several other strategic investors. The company is building what it hopes will become the backbone of financial intelligence in the AI infrastructure market: real-time data, benchmarks, and tools that bring transparency to GPU pricing and compute infrastructure costs.

At the center of Silicon Data’s product offering is the H100 Rental Index, the first daily benchmark that tracks the cost of renting NVIDIA’s H100 GPUs, which are among the most in-demand chips for AI workloads. The index is built using pricing data sourced across hyperscalers, neocloud platforms, and private infrastructure providers. It adjusts for a wide range of variables, including hardware configuration, platform pricing, and location. As CEO Carmen Li puts it, “This is the first step, to help market participants gain transparency into the convoluted world of AI costs.”

And it’s a step the market is eager for. Banks are already using the index for internal risk modeling. Hedge funds license the underlying data to help make investment decisions related to cloud and compute. Asset managers and private equity firms, many of which are financing or building next-generation data centers are using it to better understand the cost of goods sold in AI infrastructure projects. While the index isn’t tradable yet, it is already being treated like a commodity benchmark, something to track, price against, and eventually build financial products on top of.

This type of transparency is sorely needed. A recent report by McKinsey & Company estimates that by 2030, global capital expenditures on data centers will reach $6.7 trillion. Of that, $5.2 trillion is expected to go specifically toward AI-optimized infrastructure, while $1.5 trillion will support traditional IT. These are staggering numbers that point to an emerging capital arms race. And yet, despite the enormous sums of money at stake, many infrastructure decisions are being made in the dark, based on gut instinct, vendor relationships, or outdated assumptions.

As the cost of AI development balloons, visibility into compute markets becomes not just a competitive advantage, but a strategic necessity. For AI startups, understanding GPU rental pricing can make or break a business model. For cloud providers and data center operators, it can determine whether infrastructure investments are priced rationally or destined to become stranded assets. And for investors, the lack of clear pricing benchmarks adds risk to an already volatile sector.

Silicon Data aims to change this by turning GPUs into a benchmarked, structured asset class, much like energy, bandwidth, or even real estate. Their long-term vision includes not only tracking chip prices but building a broader market intelligence platform that covers everything from spot pricing to capacity forecasting. By doing so, they’re laying the groundwork for a more transparent, efficient AI infrastructure economy.

This approach represents a shift in how the industry thinks about innovation. Until now, most attention has gone to the models themselves, how big, how fast, how accurate. But increasingly, the bottlenecks are downstream: in supply chains, energy costs, and capital outlays. As Carmen Li notes, understanding what your infrastructure is actually worth may soon matter as much as how well your model performs.

The DeepSeek controversy may have been a flashpoint, but it exposed a deeper structural problem: the AI industry is growing faster than its financial infrastructure can keep up. If left unaddressed, this could lead to massive inefficiencies, wasted capital, and growing inequality between firms that can afford the opacity and those that can’t.