CoreDelta

Clearinghouse for the AI stack. Allowing data centers to hedge their massive hardware debt and AI labs to lock in predictable computing costs.

Comment

Inspiration

One on hand you have hyperscalers that have to protect their GPU fleets against price drops, on the other hand you have AI labs and startups that need protection from price spikes. 

The problem is that they don’t show up at the same time, for the same prices points for the same size.

 Normal markets have “market makers” that make sure there is always someone willing to buy (or sell) the commodity, in this case GPU/hr, at any price point.

Long passion for finance, my company runs litigation finance, plus I had the luck to code quant models with respected traders at large investment banks.

What it does

The agent spots mispricing of contracts on kalshi and polymarket given the current market structure. This is the basic building block of a market maker!

How we built it

Codex for the coding, basic python backbone with boto3 for AWS pricing.

In the future we will need to physically buy the rented GPUs at delivery, but for this prototype we didn't run the Akash instantiation. Although that's the whole point, when the market maker handles large enterprise volumes it WILL need to deliver the promised GPUs.

Challenges we ran into

Historic prices are very limited (AWS only), but clean price data collection is part of the work needed to build financial products (Bloomberg asks top dollars to access curated data sources!) So that's just one of the sub-products we'll need to develop in the future to make our agent more accurate or sell the data.

Accomplishments that we're proud of

First agent scanning the markets and the available contracts!

What we learned

Data pipelines are non existent, we need to record and prepare our own historical time series for almost everything, which is very exciting because it means the contracts will misplace the market more often! In fact even today after an afternoon of coding you can identify trades.

What's next for CoreDelta

Run some validation with a bank/trading desk and see if there is appetite for the product, if not try to contact hyperscalers (supply side) and large AI labs (demand side) and see if they already had hedging issues.

Built With

Share this project:

Updates