Providers have idle compute. Buyers pay list price. ComputeSDK Market connects the two: providers sell their own spare capacity at their own price, and your jobs fill at the cheapest listed price under a cap you set.
$ compute actions dispatch acme/api \
--workflow test --max-bid 0.00005
dispatched run_01K7ZP9M
max price (your cap) $0.0000500/s ≈ $0.18/hr
─────────────────────────────────────────────────
northscale iad $0.0000333 filled
voltmere us-east $0.0000417 available
halden sfo1 $0.0000556 over cap
klyne - $0.0000708 over cap
placed on market (northscale) at $0.0000333/s ≈ $0.12/hrTop up prepaid credits once and spend them on whichever provider runs the job. No per-provider accounts, no new billing relationships.
Metering runs per second under everything — quote the same price however you think about it.
Nothing listed at your price? The job walks on to your own providers, exactly like it does today.
Launch video
A listing, a cap, a fill — and the walk-on to your own providers when nothing's listed at your price.
Launch video
A market for compute — coming with early access.
Sellers are the providers themselves. Connect your own credential, list spare capacity at your own per-second rate, and see the unfilled demand your capacity could take. Pause or withdraw anytime — during early access we onboard providers individually.
You run the infrastructure; you set the price. Reach out in Slack, on X, or email support@computesdk.com.
`compute market` — the sell side
Connect your credential once, then list capacity at your own per-second price. Every fill runs on your infrastructure — the market never holds your compute.
$ compute market credential connect northscale
connected northscale ••••7c1b
$ compute market sell --price 0.0000333 --per second
live standard $0.0000333/sec ≈ $0.12/hr
auto-renews every hour · up to 8 jobs at once$ compute market listings
standard $0.0000333/s live 4 jobs running
$ compute market settlements --month 2026-09
gross $812.40 · platform fee $162.48 · net $649.92
exported settlements-2026-09.csvListings renew in one-hour windows, capped at a concurrency you choose. Pause, resume, or withdraw from the CLI at any time.
Add `--max-bid` to any run — a per-vCPU-second cap — or set one org-wide. A job never pays more than its cap, so price movement can't burn you.
Providers list capacity at their own per-second rates. The run lands on the cheapest listing at or under your cap, and live prices are readable over the CLI and API before anyone commits.
You pay the provider's listed rate, never your ceiling. Every fill is recorded, so the price of every job is auditable.
The market isn't a separate product — it's an entry alongside your own providers. Put it first to buy at the best price, or last as a fallback behind capacity you already own.
Buy
Cap the price once or per dispatch. The cheapest listing at or under your cap runs the job at its listed price — anything above the cap walks on to your own providers. Worst case is status quo.
# the market is one entry in your placement order
compute actions settings set-order market northscale
# cap the price once, or per dispatch — per vCPU-second
compute actions dispatch acme/api --workflow test \
--max-bid 0.00005Fallback
Nothing listed at or under your cap? The market entry is skipped and the job walks on to your own providers — same placement records, same bill as today.
$ compute actions dispatch acme/api \
--workflow build --max-bid 0.00002
dispatched run_01K8M2XQ
market: cheapest listing $0.0000333/s is over your cap
→ walking to the next entry in your placement order
placed on northscale:us-east-1 (your key)
billed on your own provider account, as usualGetting started
Add `--max-bid` to a run, or set an org-wide cap in settings. Put `market` anywhere in your placement order — before, between, or after your own providers.
The cheapest listing at or under your cap runs the job. Nothing at your price? The next entry in your policy takes it.
Top up prepaid credits and spend them on any provider in the market. We settle with providers monthly — you never open another compute account.
Pricing
Metering runs per second under everything — the same price, expressed however you think about it.
Top up prepaid credits and spend them on whichever provider fills. No per-provider accounts, no commitments — your cap is the only price guarantee you need to think about.
Your infrastructure, your price — list spare capacity on your own credential and fill the demand on the market. We meter every job and settle monthly — during early access, onboarding is a conversation, not a contract.
Not launched yet — we're onboarding both sides individually. Tell us what you'd sell, or what you'd run.
FAQ
Both sides. Your `--max-bid` (or org cap) is the most you'll pay per vCPU-second; providers set the least they'll take for their own capacity. Jobs fill only where the two meet — cheapest first.
The market entry is skipped and your placement policy keeps walking — to your own provider keys, or to a standing cap that fills later when a listing drops under it.
No — that's the point. You pay ComputeSDK in prepaid credits and we settle with providers monthly. Your own provider accounts still work alongside the market through BYOC.
Only the providers themselves — a provider lists its own spare capacity on its own credential and its own price. Nobody resells anyone else's compute.
Not yet — it's early access on both sides. Buyers and providers both request access inside the platform; we flip the flag on approval and onboard each provider individually.
Per second under everything. The same price can be quoted by the second, minute, hour, or day — the meter doesn't care how you think about it.