AI has a GPU pricing problem. Silicon Data wants to fix it.

25 snips

Aug 19, 2026

Steve Hou, Head of Research at Silicon Data and former Bloomberg quant, explains how GPU rental prices can become a tradable market. He discusses compute futures, why older A100 chips still command demand, and how price transparency could help neo-clouds finance data centers. Short takes on hedging compute risk and what rising rental rates reveal about AI buildout.

Ask episode

AI Snips

Chapters

Transcript

Episode notes

INSIGHT

Futures Turn Compute Timing Risk Into Tradeable Risk

Futures let participants lock a future price today to reduce timing risk when building or harvesting valuable but uncertain assets.
Steve Hou compares ancient crop contracts to modern compute needs, showing how locking future revenue stabilizes investment decisions.

INSIGHT

Compute Can Be A Tradeable Commodity Without Physical Delivery

Commodities don't need physical storability to be tradable; cash-settled indices like S&P futures are precedents.
Silicon Data argues compute's massive dollar exposure ($750B+ hyperscaler spend) creates enough liquidity to support futures.

INSIGHT

Three Natural Market Sides Will Support Compute Futures

There are three natural market sides: sellers (cloud providers), buyers (AI labs/inference platforms), and market makers who intermediate.
Steve Hou names market makers like DRW and Jane Street as expected intermediaries for compute futures.

Get the Snipd Podcast app to discover more snips from this episode

Get the app

The AI buildout shows no signs of slowing. And with hundreds of billions of dollars a year going into data centers and GPUs, compute has become the single biggest cost for anyone building AI products. But for all that spending, there still isn’t a straightforward way to put a price on compute — or for firms to hedge their exposure when the price changes.

Silicon Data just closed a $30 million Series A to change that. The startup aims to become the reference price for GPU rental and an index that a Wall Street futures contract would settle against. The company plans to launch its compute futures trading on the CME October 5th, pending regulatory approval.

On this episode of TechCrunch's Equity podcast, Rebecca Bellan is joined by Steve Hou, head of research at Silicon Data, to discuss the health of the AI buildout, and why the data is telling a different story than the doom and gloom headlines about depreciating chips and stalled data centers.

Listen to the full episode to hear more about:

What happens if OpenAI or Anthropic can’t make good on the massive compute commitments they’ve already promised, and whether “compute default” becomes a risk.

Why rental demand for older A100 chips is holding up, even as newer and more powerful GPUs hit the market.

If compute futures could make it easier for neo-clouds to finance expensive data centers by locking in future rental revenue.

What rising GPU rental rates across every contract length since March say about AI demand.

Subscribe to Equity on YouTube, Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod.

Chapters:

00:00 Intro

0:59 Futures contracts explained

4:01 Can GPU compute actually be a commodity?

7:10 Who's really trading these: hyperscalers, AI labs, and market makers

14:32 Locking in prices to make data center debt easier to swallow

18:21 Inside the pricing data

20:25 The A100 mystery: why an old chip is still in high demand

22:56 Do GPUs depreciate as fast as everyone thinks?

23:20 Making sense of the Texas and New York data center pauses

25:40 What the futures curve says about where prices are headed

28:01 Silicon Data's $30M Series A and what's next

30:36 Outro

Learn more about your ad choices. Visit megaphone.fm/adchoices