AI inference is becoming a commodity, but it's still bought like a luxury — fixed list prices, premium models for jobs that don't need them, real-time SLAs for work that could wait. Keld builds the neutral marketplace underneath it.
AI inference is becoming a commodity, but it's still bought like a luxury. Enterprises pay fixed list prices, route everything to premium models — even the jobs that don't need them — and pay for real-time speed on work that could happily wait minutes or hours. At the same time, the providers who can deliver near-frontier quality at a fraction of the cost are nearly impossible to discover. The result is an enormous, quiet overspend running through almost every company that has started building with AI.
That isn't a model problem; it's a market problem. Inference has no neutral place to clear — no shared view of what a given job should cost, no way to put price- and deadline-sensitive demand in front of the supply that can serve it, and no mechanism that lets the cheapest qualified provider win the work. Keld exists to build that missing market.
Keld is the world's first marketplace for AI inference. It sits underneath the AI you already run and changes the unit of work: instead of sending a prompt to one fixed provider, you send a job with a deadline (how long you can wait) and a ceiling (the most you'll pay), naming a model or just a use case. Keld routes that job through the marketplace to the best-value provider that meets your bounds, and settles at or below your ceiling. Your stack doesn't change; the economics underneath it do.
A market only works when it's neutral, transparent and open to discovery. Keld favours neither buyers nor sellers — every job is matched on price, deadline and performance, never on who pays us. We surface real clearing prices and fulfilment volumes instead of opaque rate cards. And we make it possible for lower-cost providers delivering near-frontier quality to actually be found and used, rather than overlooked because no one had heard of them. Neutrality, transparency and discoverability aren't slogans for us — they're the three properties without which a market for inference can't be trusted.
Most organisations can't say what their AI actually costs, let alone optimise it — so we start there. Keld Atlas maps your spend across teams, models, projects and use cases, for free, with no routing changes. Once you can see where the money goes, optimising the workloads that don't need a premium model or a real-time SLA is a switch you flip, not a migration. We'd rather prove the savings on your own data than ask you to take a number on faith — so we earn the right to help you optimise by first making your spend legible.
Over time, a transparent market for inference does more than cut invoices. It tells buyers what a unit of intelligence actually costs, rewards the providers who deliver quality efficiently rather than the ones with the biggest marketing budgets, and lets capacity that would otherwise sit idle find the work that needs it. We think that's how a healthy AI economy ends up priced — not by a handful of list prices set from the top, but by supply and demand meeting in the open. Keld is the infrastructure for that to happen, and the marketplace is only the beginning of what an open inference economy needs.
Keld is built by Dave Otten, Doug Shore and Federico Enni — three technology-industry veterans who previously scaled JW Player into the video platform behind thousands of enterprise customers, serving many of the world's largest publishers and broadcasters at the intersection of video streaming and advertising. Running infrastructure at that scale taught us what it takes to make high-volume, real-money systems dependable: distributed systems, data, security, billing and developer experience that hold up under enterprise load. We started Keld to bring that hard-won discipline to the fragmented, fast-moving world of AI inference.
The three of us are only the start. Keld is a growing, AI-first team of industry experts who have spent their careers building platforms at high scale and enterprise quality. We're hiring across engineering, product and go-to-market.
A small team at the intersection of large-scale platform engineering and AI systems.
Dave sets Keld's direction and leads the company. He previously co-founded and led JW Player, scaling it to thousands of enterprise customers, and now focuses on making neutral, transparent inference pricing a category enterprises can rely on.
LinkedInFederico shapes Keld's product and go-to-market, translating how enterprises actually buy and run AI into the roadmap and pricing model. He draws on years building enterprise platforms at scale, including at JW Player.
LinkedInDoug leads engineering, building the exchange, gateway and matching systems that route every job to the best-value provider within its bounds — applying the distributed-systems discipline he honed scaling JW Player's infrastructure.
LinkedInWe're a small team and we're growing. We're hiring →