Shadeform (YC S23) - Your GPU Cloud Marketplace 🔋

CV Deep Dive

Today, we’re talking with Ed Goode, Co-Founder and CEO of Shadeform.

Shadeform is a GPU cloud marketplace that connects startups and developers with the best compute resources across 20+ cloud providers. The startup was founded by Ed and his co-founder Ronald Ding in 2023, in response to the growing demand for scalable, cost-effective AI infrastructure due to the generative AI boom. Shadeform’s goal is to enable companies to access the most competitive pricing and cutting-edge hardware without the complexity of managing multiple cloud environments.

Today, Shadeform aggregates GPU supply from more than 20 clouds like Lambda, Nebius, and Crusoe, offering a unified API and console for smooth provisioning. They serve a wide range of customers, from solo developers to enterprises, giving them instant access to thousands of on-demand GPUs without any quota restrictions. Shadeform also offers an automated compute brokerage system, enabling teams to get multiple quotes for reserved instances and interconnected clusters in as little as 24 hours.

In this conversation, Ed shares how Shadeform was built, the opportunity around managing a multi-cloud ecosystem, and his vision for making high-performance compute more accessible and efficient than ever before.

Let’s dive in ⚡️

Read time: 8 mins

Our Chat with Ed 💬

Ed, welcome to Cerebral Valley! First off, introduce yourself and give us a bit of background on you and Shadeform. What led you to co-found Shadeform?

Hey there! My name is Ed and I’m the CEO and Co-Founder of Shadeform. My career started off at Microsoft, where I worked on ML infrastructure and compute orchestration for Azure. I began obsessing about this problem in 2021 / 2022 when the centralization of the compute layer to AWS, GCP, and Azure exposed a big risk to companies and individuals from server downtime, cartel-like pricing, and censorship.

I started to see the emergence of GPU demand in late 2022 and how the market could potentially change with emerging capacity issues and new cloud providers. I always felt that the first generation of cloud was built wrong. Each cloud is essentially built on top of their virtualization layer, so we started Shadeform after working on virtual machines scale sets to build a multi-cloud, distributed VM layer that we could then build ML services on top.

How would you describe Shadeform to the uninitiated developer or AI team?

Shadeform is a GPU cloud marketplace. We partner with over 20 cloud platforms like Lambda Labs, Nebius, and Crusoe to offer their compute and services through a centralized API and console, so you can provision infrastructure across any cloud environment with a unified experience—no need to deal with countless different APIs, intricacies, account management, or UIs. We provide centralized billing and a streamlined deployment model, making it easier for teams to tap into the entire compute market instead of getting locked into a single provider.

Who are your users today? Who is finding the most value in what you're building at Shadeform? 

We break our users into a few segments. We have hobbyists and individual users, as well as fast-growing startups and enterprise teams. Our marketplace is all on-demand, so users can instantly access whatever infrastructure they need. 

We also broker compute commitments and clusters. With our automated brokerage platform, users fill out a single form and instantly receive quotes from top cloud vendors instead of sourcing everything manually or through an opaque process with a broker. This flexibility lets us serve a wide range of customers—from large, mature companies securing compute reservations in new clouds, to individual users looking for a few servers for their projects.

Which existing use-case for Shadeform has worked best? Any customer success stories you’d like to share? 

Some of the biggest success stories have been around helping inference providers secure compute during supply crunches. More recently, as a result of near SOTA open source releases, we’ve been helping inference and agent platforms get access to H200s and B200s that aren’t yet widely available on the market.

DeepSeek has changed the game—the latest models now require significantly more VRAM for hosting, meaning you can’t just run them on a single H100 node anymore. We’re at an interesting moment where new, larger models are coming online while new compute, like H200s, are only slowly becoming available, with B200s following shortly after.

This has created a weird supply chain effect, and a big part of our value is helping these teams scale when they otherwise wouldn’t be able to. We help them find and utilize the best resources across different cloud providers.

Walk us through Shadeform’s platform. Which use-case should teams experiment with first, and how easy is it for them to get started? 

It’s pretty straightforward. If you go to our site, you’ll immediately hit the signup flow, where you’ll see a list of all compute providers and their prices for different GPU types. From there, you can easily one-click deploy VMs and bare metal instances across any of those providers. You don't have to set up accounts, get quotas, and jump through hoops with the underlying provider. That's all taken care of by Shadeform.

You can use our UI and multi-cloud console, or leverage our API to integrate directly and provision compute across the market—either at the lowest cost or in the right region for your end users. If you have a workload that you want to save and create replicas for, we offer easily saveable templates for containers and startup scripts, allowing you to package up your application and deploy it across different environments without being locked into one provider.

How are you measuring the impact or results that you’re creating for your customers? 

The biggest thing for us is time to market and cost savings. By giving companies the flexibility we do, they can get the best deals while only having to manage a single integration. At the same time, they can always get access to the latest and greatest compute on-demand, so they’re not restricted to just reservations.

This gives our customers a pricing edge while also providing access to geographies and servers that aren’t available in every provider.

There are a number of fast-growing startups attacking the space of compute provisioning. What would you say sets Shadeform apart from a product or UX perspective? 

For one, it’s the added layer of support and reliability we bring to these platforms. We often have very close working relationships with our users, and offer a high-touch layer of hand-holding to help get their workloads set up. In many cases, we act as tier-one support, providing a level of service that goes beyond what the providers themselves typically offer.

The other big factor is transparency. Many compute brokers in the market completely obscure the end provider. With us, whether you're using our compute brokerage service or our on-demand platform, you see exactly who the providers are, what the prices are, and get full visibility to make an informed decision. We don’t hide the cloud vendors to maintain an edge—we open that up.

If you had to put an analogy to it, we’re basically the Kayak for compute.

Compute is notorious for supply crunches. How do you navigate these, given they're so often a critical bottleneck when it comes to AI progress more generally? 

We’ve built really strong relationships with all the leading players in this space. In some cases, we even have early access to Blackwell and other inventory as it comes online. Beyond that, we’ve also partnered with great cloud providers like Massed Compute and Boost Run—companies that have highly competitive inventory and pricing but don’t necessarily have the massive venture backing.

2025 is the “Year of the Agent”. How are you thinking about AI agents internally, and how does this factor into your product vision for Shadeform? 

For us, agent platforms are often our ideal customer—especially inference and agent platforms that need their own servers rather than relying on a third party endpoint. We want to support a thriving and growing agent ecosystem while making it easier for customers to run agents in different environments.

Overall, we see this as a major boost for compute usage, which in turn benefits Shadeform. Longer inference times and longer-running jobs mean more compute consumption overall, aligning perfectly with what we want to be, which is the single one-stop shop for all accelerated cloud compute.

What has been the hardest technical challenge around building Shadeform into the platform it is today? 

We've built software from the ground up on bare metal hardware, enabling virtualized instances all the way up to the orchestration layer. The hardest challenge, honestly, is dealing with heterogeneous supply while ensuring a reliable experience on top.

A big part of this involves debugging issues with providers, helping them improve their systems, and making sure we can consistently deliver a confident, reliable experience. Managing a workload on a single cloud is one thing. Managing workloads across 20 clouds is another. And then layering customers on top of that system adds even more complexity.

That heterogeneity means we have to bend the underlying infrastructure to meet our users' needs, which takes both manual work and automated effort.

Lastly, how do you see Shadeform evolving over the next 6-12 months? Any specific developments that your customers should be excited about? 

We're going to be rolling out more cloud services on top of our multi-cloud platform.

We're also hiring for sales, engineering, and ops right now if anyone wants to work on this problem.

Finally, we’re starting to publish a weekly newsletter where we cover provider available inventory, prices and more. Subscribe at https://shadeform.ai/newsletter

Conclusion

Stay up to date on the latest with Shadeform, learn more about them here.

Read our past few Deep Dives below:

• Build specialized RAG agents with Contextual AI 🌐

• Remyx - Your AI Production Assistant 💡

• Qdrant's GPU-accelerated vector indexing is here 🔋🌐

• Athena is your AI-powered remote hire 🎛

• Doowii - your AI-first education platform 📖

If you would like us to ‘Deep Dive’ a founder, team or product launch, please reply to this email ([email protected]) or DM us on Twitter or LinkedIn.

Join Slack | All Events | Jobs