Stoa Markets is a Y Combinator S26 startup that runs an RFQ (request-for-quote) marketplace where institutions buy and sell physical GPUs and AI servers — not rented compute time, but ownership of the hardware itself. Vetted dealers submit blind, firm quotes within 48 hours, and Stoa handles the KYB checks, contracts, and settlement without ever taking possession of the gear.
Short answer: Stoa Markets is a YC S26 startup where institutions post what GPU hardware they want to buy or sell — from a single card to a 1,024-server cluster — and vetted, KYB-verified dealers return firm quotes within 48 hours. Stoa tracks payment, shipping, and inspection through settlement. It's built for data centers, AI labs, and brokers, not individual buyers.
Most Show HN launches that cross my desk are consumer apps or dev tools. Stoa is neither. It's market infrastructure for the hardware behind every AI buildout story I cover. When I tested the pitch's claims against Stoa’s own site, the Launch HN thread, and the founders' replies in the comments, most of it held up. Here's what Stoa does, how the RFQ process works, and where the fraud concerns raised in the comments land.
What you'll need to know before you look at Stoa
Stoa isn't a consumer product. That's the first thing to get straight. Founders Berat Celik, Eren Berke Saglam, and Kaan Yigit built it for institutions: data center operators, AI labs, cloud providers, OEMs and resellers, brokers, and lenders financing hardware purchases. Between them, the three have backgrounds trading interest rate derivatives and building pricing infrastructure for oil and gas. In a Launch HN reply, co-founder Eren said the platform "focuses on institutions" partly because of export-control requirements, though it can technically handle anything from "a single card to 1,024 servers." If you're picturing a Craigslist for GPUs where anyone can list a spare card, that's not this. Every dealer on the platform passes a know-your-business (KYB) check before they can quote.
Step-by-step: how a trade actually moves through Stoa
1. Describe the trade
A buyer or seller specifies the hardware — GPU model (A100, H100 SXM5, H200, B200, B300, GB200 NVL, and similar are all listed on Stoa's site), condition grade (New, Refurbished, or Used), and quantity, from a bare card up to a full cluster. Stoa's interface lets you type this in plain language or upload an existing quote PDF, and the platform uses AI to help draft a standardized RFQ from it.
2. Review and send
Before the request goes out, Stoa makes the terms explicit: delivery window, deadline, and condition requirements. This matters because, per the founders' own pitch, identical server configurations were getting wildly different quotes from different dealers — one example they cited was the same config priced at $200,000 by one dealer and $240,000 by another, with no way for the buyer to know which number was closer to fair.
3. Vetted dealers quote blind
The RFQ goes out to dealers who've already cleared Stoa's KYB verification. They submit firm quotes without seeing each other's numbers — a blind-bid structure meant to produce real price discovery instead of one dealer anchoring to another's number.
4. Execute at a firm price
Accepting a quote creates a binding commitment, with the agreed price, delivery terms, and conditions recorded on the platform. Founders say quotes typically come back within 48 hours, versus what they describe as weeks of cold-calling dealer networks the old way.
5. Settle step by step, without Stoa touching the hardware
Settlement runs through a recorded sequence — confirmed, paid, shipped, delivered, inspected, settled — with evidence required at each step. Stoa never takes possession of the GPUs; it's tracking the transaction and holding payment until delivery is confirmed, closer to an escrow-plus-workflow layer than a warehouse.
Example prompts you can copy
If you're evaluating a GPU purchase, whether through Stoa or a dealer directly, these prompts are useful for pressure-testing a quote with an AI chat tool before you commit:
“ I have a quote for [N] units of [GPU model, e.g. H100 SXM5], condition grade [New/Refurbished/Used], at $[price] per node. Based on publicly reported GPU market pricing, does this look reasonable? What questions should I ask the seller about warranty, export compliance, and delivery lead time? “
“ Summarize the difference between buying GPU hardware outright versus renting GPU compute by the hour. For a team planning to run [workload], which model makes more financial sense over 18 months, and what's the breakeven point? “
“ What KYB and export-control requirements typically apply to institutional GPU hardware sales in the US? What documentation should a buyer expect a verified dealer to provide before wiring payment? “
These work because they ask for a checklist and a reasoning process instead of a single number — a model can't verify a live quote for you, but it can flag the questions a buyer should be asking before money moves.
Common mistakes to avoid
The most common mix-up in the Hacker News comments themselves was confusing Stoa with a compute-rental service. As one commenter put it bluntly: "vast.ai rents. this one sells." Stoa is a hardware ownership marketplace; if you want compute by the hour instead of a server you own, that's a different category of product entirely. Second, don't assume "48-hour quote" means "48-hour delivery" — the quote turnaround and the physical shipping timeline are separate, and dealers on the thread noted lead times of 16 to 24 weeks are common for new hardware regardless of which marketplace you use. Third, several commenters pressed the founders specifically on fraud risk, asking about protections "beyond KYC/escrow" and warning about "long firm" schemes where an initially trustworthy counterparty absconds after building a track record. Stoa's KYB checks and payment-held-until-delivery structure address some of this, but they're not a licensed escrow service, and a marketplace this new hasn't been tested by a bad actor at scale yet — worth weighing before you run a large trade through any young platform. Fourth, don't take the platform's own comparison pricing as independently audited; it's Stoa's own illustrative example, not a number I or anyone outside the company verified against a real closed trade.
Stoa Markets vs. GPU cloud rental marketplaces
| Stoa Markets | Vast.ai / RunPod (GPU cloud rental) | |
|---|---|---|
| What changes hands | Ownership of physical GPUs or servers | Hourly access to compute, no ownership |
| Who it's for | Institutions: data centers, AI labs, brokers, lenders (KYB required) | Individual developers up to enterprise, self-serve signup |
| Price discovery | Blind RFQ; multiple vetted dealers quote in ~48 hours | Posted hourly/spot rates, updated continuously |
| Settlement | Payment held until delivery, shipping, and inspection are confirmed | Billed per hour or minute of usage |
| Example pricing (Stoa's own site) | 8-node H100 SXM5, Grade A, listed at roughly $243,400/node on-platform vs. $262,900/node off-market | Not applicable — no hardware sale involved |
The pattern in that table is the real distinction: Stoa is trying to fix price discovery for people buying servers outright, while Vast.ai and RunPod solve a different problem — access to compute without a capital purchase at all. They're not really competitors so much as adjacent answers to "I need GPUs" depending on whether you want to own them.
Tools that make this easier
If you're trying to place Stoa in the bigger AI-infrastructure buildout picture, my piece on xAI, SpaceX, and the race for AI buildout covers what a company actually building at Stoa's scale — hundreds of thousands of GPUs — looks like from the demand side. AI data centers are driving up power bills covers the other constraint on hardware buildouts that a marketplace like Stoa doesn't touch: power. Nvidia’s $750 billion in AI deals is useful context for why GPU supply is fragmented enough that a company like Stoa sees an opening in the first place, and why the AI trade increasingly runs on borrowed money explains why lenders specifically are one of the buyer categories Stoa lists. If you want the finance-side skepticism on all this AI capex, Morgan Stanley’s debate over how much of SpaceX’s valuation is AI is a good companion read. I've covered other recent Show HN launches the same way I approached this one — my hands-on test of Needle2, a 14MB on-device tool-calling model, is the closer-to-consumer counterpart to this piece. And my AI tool ratings hub is where I track the software side of this same "which AI product is actually worth using" question.
My take
Stoa is solving a real, unglamorous problem: GPU pricing is opaque, and identical hardware configs get wildly different quotes depending on who you happen to call. Standardizing the RFQ and verifying dealers with KYB checks is a sensible fix, and $300M+ in RFQs in the first month suggests institutional buyers agree there's demand for it. The open question, which several Hacker News commenters raised directly and the founders didn't fully close, is how much protection KYB-plus-escrow-style settlement actually provides against a bad actor who passes verification and defaults later. For a large institutional buyer already doing this kind of diligence manually, Stoa looks like a genuine time-saver. I wouldn't treat it as a substitute for your own contract review on a seven-figure hardware purchase.
Frequently Asked Questions
Is Stoa Markets free to use?
Stoa doesn't publish a public fee schedule on its marketing site. The platform is positioned as a marketplace connecting buyers, sellers, and vetted dealers rather than a subscription product, but you'd need to request access to see actual transaction terms.
How is Stoa Markets different from Vast.ai or RunPod?
Stoa sells hardware outright — you own the GPUs afterward. Vast.ai and RunPod rent compute by the hour with no ownership transfer. A Hacker News commenter summarized it as "vast.ai rents. this one sells," which is the cleanest way to remember the split.
Who can actually use Stoa Markets?
Institutions: data center operators, AI labs, cloud providers, OEMs and resellers, brokers and dealers, and lenders or funds financing hardware. Co-founder Eren said on the Launch HN thread that Stoa focuses on institutional buyers partly due to export-control requirements.
How fast do you get a quote on Stoa?
Firm quotes from vetted dealers typically come back within 48 hours of submitting an RFQ, according to Stoa's own site and launch materials, compared to what the founders describe as weeks of manually calling dealer networks.
Does Stoa Markets protect against fraud?
Stoa requires KYB verification for dealers and holds payment until delivery and inspection are confirmed. Hacker News commenters pressed on whether that goes far enough — comparing it to licensed escrow and flagging "long firm" schemes where a verified counterparty defaults after building trust — and it's a fair question for any buyer moving a large trade through a young platform.