Stoa: A Marketplace for New and Used GPUs and AI Servers

By DataTip · Published August 11, 2026

GPU and AI-server trading still runs through phone calls, forwarded spreadsheets, and long email threads. Stoa: A Marketplace for New and Used GPUs and AI Servers is an attempt to give that process a consistent workflow without pretending software can replace the relationships behind the market.

Stoa is a marketplace for new and used GPUs and AI servers that standardizes RFQs, sends them to dealers that have completed KYB checks, collects firm quotes, and tracks payment, shipping, delivery, and inspection through settlement without taking possession of the hardware. Its broader aim is to improve liquidity and create resale evidence for lenders while preserving the relationship-based trust central to GPU trading.

We are Eren, Berat, and Kaan. We started Stoa after brokering GPU deals ourselves and repeatedly encountering fragmented quotes, limited market visibility, uncertain resale values, and too much manual work between a request and a completed trade. We introduced Stoa on Hacker News as a marketplace for new and used GPUs and AI servers.

Why is GPU collateral difficult to finance?

GPU financing terms often depend on the offtaker: the company committed to using the compute. A hyperscaler’s commitment may support financing that looks investment grade. The same hardware pledged by a smaller cloud provider or startup can attract much more expensive terms.

The lender’s concern is reasonable. If the borrower defaults, what can the servers actually be sold for? There is no reliable, shared answer today. List prices and one-off appraisals do not necessarily show what a specific configuration, in a specific condition and location, can command in a real sale or liquidation.
That uncertainty makes GPU collateral difficult to assess. Lenders need evidence from actual resale activity, not only an equipment description. That evidence is hard to assemble when transactions remain scattered across private conversations.

How does fragmented GPU trading create inconsistent quotes?

We began brokering GPU deals to understand the friction firsthand. The workflow was still built around calls, spreadsheets passed between companies, and long email chains. A buyer looking for the best available offer might contact several brokers and dealers separately, repeat the same request, and reconcile quotes based on different assumptions.

We saw the problem in a straightforward example. One seller quoted us for a server node, while another quoted what appeared to be the same thing. Neither seller was necessarily wrong. They had different information and could see only their own corner of the market.

Different quotes therefore do not automatically indicate bad faith. They may reflect differences in inventory visibility, demand, configuration, condition, or delivery. Two offers that look similar may not be comparable because the dealers are working from different information.

The process is inefficient for sellers as well. They have to search for demand one buyer at a time, often through separate intermediaries and relationship networks. Buyers want a competitive process; sellers want qualified demand. A market of this size should not require every participant to rebuild the same outreach effort from scratch.

AI GENERATED

How does Stoa: A Marketplace for New and Used GPUs and AI Servers handle an RFQ?

Stoa puts each request into a common format and sends it to dealers that have completed know-your-business checks. Before the request goes out, the buyer confirms the details that define what is being offered and evaluated:

• Exact configuration and quantity
• Hardware condition and warranty
• Location and delivery terms
• Inspection criteria and what will be checked
  The KYB process verifies the dealer’s company, who owns it, and who is authorized to trade for it. Dealers then return firm quotes against the same request without seeing one another’s bids.

That structure gives every dealer a defined specification to answer instead of a sequence of informal conversations with changing assumptions. The buyer states the requirements once and can compare responses against the same baseline.

After a quote is accepted, Stoa tracks payment, shipping, delivery, and inspection through settlement. It does not take possession of the hardware. Stoa organizes the transaction and its records; it is not the owner or custodian of the equipment.

This workflow does not remove the need for inspection or counterparty checks. It gives those responsibilities a clearer place in the transaction and creates a more consistent record from the initial RFQ through settlement.

Can a marketplace improve GPU liquidity and resale evidence?

Stoa reported receiving more than a million in requests for quotes during its first month. The immediate goal is practical: make buying and selling GPUs and AI servers less painful.

The longer-term argument concerns liquidity and financing. As completed trades accumulate, they may give lenders actual resale evidence instead of leaving them to rely mainly on list prices and isolated appraisals. A record of completed transactions can show how hardware moved under particular conditions, although it does not guarantee a valuation or financing result.

The same process may help connect demand and supply more efficiently. Buyers can express demand through one structured request, while dealers can respond to defined opportunities rather than searching for prospects one at a time. But a cleaner RFQ process is not the same as certainty. Outcomes still depend on configuration, condition, warranty, location, delivery terms, inspection, and the counterparties involved.

Why can’t GPU trading be software-only?

We knew from the beginning that GPU trading could not be reduced to software. Inventory is not shown to everyone. Dealers need to trust the people bringing them clients, and clients need to trust that quotes will turn into real trades rather than another dead end.

AI GENERATED

Those relationships develop through repeated dealings and a history of follow-through. We built them by brokering deals ourselves. Stoa provides a cleaner route from the first RFQ through settlement, but it does not replace the trust underneath that route.

That limitation matters. Stoa should not be treated as a universal answer for every hardware transaction. Readers should not infer guarantees about pricing, custody, counterparty performance, or hardware condition. A standardized workflow can clarify the request and preserve a better record; it cannot make every dealer, quote, or inspection outcome equivalent.

Who built Stoa, and why did it start?

We have known each other for more than ten years. Between us, we have founded companies, traded interest rate derivatives, and built trading and pricing systems for oil and gas.

We learned GPU trading by doing the deals ourselves. Stoa grew from the recurring problems we encountered: fragmented outreach, uneven market visibility, uncertain resale evidence, and the relationship-based trust required to turn a quote into a completed transaction.

The product is an attempt to put structure around an existing market, not a claim that the market can be replaced by an interface. The RFQ and settlement workflow are the visible system. Dealer relationships, client relationships, and the history of who follows through remain part of the process.

What does Stoa charge?

Stoa charges a tiered fee on completed trades, with lower fees at higher volumes. Signup is free through Stoa’s signup page. Commercial terms should be checked directly with Stoa because they can change.

When should buyers be cautious about using a marketplace?

Be cautious when a transaction depends on requirements that are not fully captured in the RFQ, or when you need guarantees about pricing, custody, counterparty performance, or hardware condition. Stoa describes a way to organize trades; it does not remove the need for verification, inspection, and judgment about the relationships involved.

The founders are asking people who have bought, sold, financed, or liquidated GPUs to share their experience, particularly where the process breaks down between quote, inspection, settlement, and the eventual resale question.

Key takeaways

• GPU financing can change materially with the offtaker, even when the same hardware is pledged as collateral.
• Standardizing configuration, condition, warranty, location, delivery, and inspection terms makes dealer quotes more comparable.
• A marketplace can create resale evidence and improve liquidity, but it cannot guarantee valuation, hardware condition, or counterparty performance.
• Stoa’s workflow handles RFQs and settlement tracking; relationships and follow-through still support the trade underneath.

Practical tips

• Write the RFQ as a complete specification before requesting bids, including inspection criteria and delivery terms.
• Treat two similar server quotes as potentially information-different until configuration, condition, warranty, and location are reconciled.
• Separate the workflow question from the trust question: a cleaner transaction record does not replace dealer and client verification.
• For financing or liquidation decisions, distinguish list prices from evidence of completed resale transactions.

Share your GPU trading experience

If you have bought, sold, financed, or liquidated GPUs, Stoa is asking for practical feedback on where the current process breaks down.