Stoa Brings Trading-Desk Mechanics to the GPU Marketplace
Stoa, a marketplace for new and used GPUs and AI servers, has launched out of Y Combinator's Summer 2026 batch with a design borrowed from institutional trading. Buyers describe the hardware they need in plain language or attach a quote PDF, and an AI system drafts a request for quote (RFQ) — a structured document specifying what the buyer wants. The buyer can edit that draft before it goes out to verified dealers. Those dealers respond with firm price quotes, and accepting a quote creates a binding trade. Stoa
Founded by Eren, Berat, and Kaan, Stoa grew out of the founders' own GPU brokering business. They entered brokering to understand firsthand why GPU hardware trading stayed manual and inefficient, and the marketplace is the productized result. The three founders have known each other for over ten years and bring backgrounds in founding companies, trading interest rate derivatives, and building pricing systems for oil and gas. Hacker News
The marketplace covers datacenter GPUs ranging from individual cards to full clusters, in new or used condition. Before dealers can receive RFQs, they must pass know-your-business (KYB) checks — the corporate equivalent of identity verification. On the buyer side, the process requires confirming exact configuration, quantity, condition, warranty, location, delivery terms, and inspection criteria before dealers return firm quotes against the same request. Stoa does not take possession of the hardware it brokers. Hacker News
Settlement follows a recorded sequence of steps: confirmed, payment, shipped, delivered, inspected, and settled. Each step requires evidence, creating an auditable trail from order acceptance to final settlement. Stoa charges a tiered fee on completed trades, with lower fees at higher volumes. Stoa
In its first month, Stoa received more than $300 million in requests for quotes. The company is a member of Y Combinator's S26 batch, confirmed by YC's own LinkedIn announcement in early August. Hacker News Y Combinator
The GPU secondary market has long suffered from opaque pricing, trust gaps between buyers and sellers, and settlement risk. What Stoa is attempting will be structurally familiar to anyone who has worked in electronic bond or commodities trading: standardize the RFQ workflow, enforce counterparty verification, and create a binding settlement chain with evidence at each step. The founders' derivatives and oil-and-gas pricing backgrounds show in the design choices, particularly the tiered fee structure tied to completed trades rather than listings or inquiries, and the evidence-gated settlement timeline.
The AI-drafted RFQ is worth noting as a practical application of large language models in business procurement rather than a gimmick. A buyer attaching a vendor quote PDF and receiving a structured, editable RFQ back compresses what is typically a manual specification step, one that has historically been a source of errors and disputes in hardware procurement. Whether the AI-generated drafts prove reliable enough at scale to reduce, rather than introduce, friction will depend on how well the system handles the long tail of GPU configurations, firmware revisions, and cluster-level requirements that buyers care about.
The $300 million in first-month RFQ volume signals demand for a structured marketplace in this category, though volume of requests is not the same as completed and settled trades. The gap between RFQ volume and actual transaction throughput will be the metric to watch as the platform matures.
Stoa's decision not to take possession of hardware keeps it asset-light but places the burden of logistics, insurance, and physical verification on the counterparties and the settlement workflow. The evidence-gated timeline is the mechanism meant to hold this together: if each step requires proof before advancing, the system creates accountability without Stoa needing to operate as a warehouse or logistics provider.
The broader question for platforms like Stoa is whether marketplace liquidity — meaning enough buyers and sellers actively trading in one place — concentrates fast enough to make the RFQ model the default venue for GPU hardware, or whether the secondary market fragments across informal channels, direct broker relationships, and competing platforms. The tiered fee structure is designed to reward volume and concentrate trading, but network effects in hardware marketplaces have historically been slower to compound than in purely digital goods markets, where settlement is instantaneous and physical logistics are absent. GPU trading sits in an awkward middle: the assets are high-value, physically delivered, and configuration-sensitive, which raises the stakes on every step of the settlement chain.
For now, Stoa has built the rails. The market will determine how much traffic runs on them.
