In the chaotic world of GPU trading, deals still get done the old-fashioned way: phone calls, personal connections, and handshake agreements on six-figure hardware purchases. A dealer in Texas might have a rack of NVIDIA H100s. A lab in San Francisco needs them urgently. Between the two sits a fragmented market where prices shift like sand and nobody quite agrees on what anything should cost.
Stoa, a San Francisco startup that emerged from Y Combinator, is attempting to inject something Wall Street traders would recognize—the request-for-quote system—into what has essentially been a high-stakes equipment swap meet.
The company's pitch is straightforward enough: institutional buyers post specs for the GPUs they need. Verified dealers submit binding quotes within 48 hours. The buyer picks one, the platform holds payment in escrow until hardware arrives, and both sides are contractually locked in. No more ghosting. No more price renegotiation at the loading dock.
Since launching in late July, the platform claims to have processed more than $300 million in RFQs during its first month—a figure the company disclosed on LinkedIn. How many of those requests turned into actual, settled transactions? That's the question Stoa hasn't answered yet, and it matters. Early RFQ volume often reflects curiosity and price shopping more than genuine deal flow.
Still, for a platform that's been live for only weeks, the initial activity suggests something: either genuine demand for transparent pricing infrastructure, or at least widespread frustration with how GPU deals currently work. Possibly both.
The Mechanics Behind the Marketplace
Stoa's workflow borrows heavily from institutional finance. The founding team includes a CTO, Kaan Yigit, who previously built systems as a quantitative developer on Uniper's commodities trading desk—experience that shows in the platform's architecture. This isn't a consumer-facing app. It's plumbing designed for participants who understand repo markets and credit RFQs.
The process begins when a buyer submits hardware specifications through the platform: configuration, condition (new, refurbished, used), and delivery requirements. Verified dealers—entities that have passed Know Your Business checks—respond with firm quotes. Once accepted, both parties are bound. Payment sits with Stoa until delivery confirms, then releases. Every step generates an audit trail.
Right now, the platform handles NVIDIA's H100, H200, B200, and A100 GPUs, along with associated servers and cluster configurations. Access is restricted to U.S.-based institutional participants. No retail buyers, no international flow—at least not yet.
Berat Celik, the CEO, completed a master's in engineering at Cornell Tech before joining Y Combinator. His co-founder and CFO, Eren Berke Saglam, is a Dartmouth graduate who holds Series 7 and 63 licenses. The regulatory credentials matter in a market where participants are used to operating in regulatory gray zones.
The company's Terms of Service—updated in late July—position Stoa as a marketplace intermediary rather than a transaction principal. It doesn't take custody, doesn't warehouse inventory, doesn't make markets itself. The platform prohibits market manipulation, deceptive content, and scraping, the standard guardrails for any venue trying to formalize what's been an informal market.
The Real Play: Data, Not Just Transactions
Here's where it gets more interesting. Stoa charges a platform fee on completed trades, but participants who join its "data partner" program can trade at zero percent on eligible transactions. Data partners are entities with active GPU flow willing to share pricing information.
Translation: Stoa is building a price index.
The company is developing something called the STOA-H100 Composite, which aggregates and normalizes pricing data across verified sources to generate daily reference prices. As of mid-summer, the index page existed on the company's website but displayed "value unavailable"—whether that means it's gated behind institutional access or simply not operational yet remains unclear.
If the index achieves credibility, though, the potential applications extend far beyond facilitating individual trades. Institutional participants talk about using reliable index data for collateral marking on GPU-backed lending, inventory valuation on balance sheets, depreciation tracking, and general pricing benchmarks in what has been an opaque asset class.
Building a trusted benchmark is no small feat. It requires consistent data contributions, transparent methodology, and widespread adoption. Stoa is betting that institutional players frustrated by price opacity will trade fee-free in exchange for contributing to that data pool. Whether that bet pays off depends entirely on participation rates—and trust.
A Crowded Field, Different Approaches
Stoa isn't operating in a vacuum. Several companies recognized roughly the same opportunity at roughly the same time: AI infrastructure has matured into a distinct asset class, but the financial infrastructure hasn't kept pace.
Architect Financial Technologies announced earlier this year that it would launch regulated futures and options on compute costs after securing a designated contract market license. Synova Global launched NativX, an exchange-as-a-service platform positioning compute and connectivity as tradable economic inputs. Oblivus entered the GPU rental marketplace in July.
Each takes a different structural approach. Architect is building derivatives. Synova is constructing exchange infrastructure. Oblivus focuses on rental markets. Stoa sits somewhere between a broker network and an exchange—providing standardized RFQ mechanics without taking principal risk.
The fragmentation might eventually consolidate. Or different models might serve different corners of the market. For now, the sheer number of entrants suggests market participants believe the current state of affairs—opaque pricing, relationship-dependent deals, limited audit trails—isn't sustainable as GPU transactions scale.
What the First Month Actually Tells Us
The $300 million RFQ figure is eye-catching, but context matters. RFQ volume and settled transaction volume are different animals. A buyer might submit an RFQ to five platforms simultaneously, generating volume on each without executing anywhere. Or dealers might quote aggressively to test the platform without serious intent to transact.
Early marketplace traction often reflects pent-up demand for price discovery rather than actual transaction velocity. Buyers want to know what GPUs should cost. Dealers want to gauge demand. Both can generate RFQ activity without money changing hands.
Stoa hasn't disclosed conversion rates—how many RFQs turned into binding transactions, how many of those settled successfully, what the average transaction size looks like. Those metrics will matter more than headline RFQ volume as the platform matures.
The company targets a broad participant base: GPU brokers, data center operators, AI labs, cloud providers, OEMs, resellers, lenders, and investment funds. That's ambitious. The challenge for any two-sided marketplace is achieving enough liquidity on both sides to make the platform faster and more reliable than just calling someone you know.
The Test Ahead
Stoa's website describes the platform as "the market behind AI hardware." That's a bold claim for infrastructure launched weeks ago. Whether it becomes the market, a market, or just one option among many depends on execution and adoption.
The company has live infrastructure, verified participants, and what appears to be genuine early interest. The founding team brings relevant financial market experience to a hardware trading problem. The data partner program suggests strategic thinking beyond pure transaction fees.
But marketplace businesses live or die on network effects. Stoa needs enough buyers to attract dealers, and enough dealers to attract buyers—and it needs both sides to trust the platform enough to lock into binding transactions instead of keeping backup options open.
For an industry that has operated on relationships and phone calls, that represents a meaningful behavioral shift. Whether institutional participants are ready to make that shift, or whether they'll simply use Stoa for price discovery while closing deals the old way, remains the central question.
The company can be reached at [email protected] for general inquiries and [email protected] for media questions. The hard part—turning interest into sustained liquidity—is just beginning.
