Feature Request: Peer-to-Peer API (Compute) Credit Transfers & Market

ChatGPT Feature requests
1

Problem

Teams and indie devs have bursty workloads and uneven cashflow. Idle credits sit unused while others hit rate/credit ceilings during launches or spikes.

Proposal

Enable credit transfers within/between verified accounts, plus an opt-in internal marketplace for buying/selling credits (or “capacity blocks”) under OpenAI-defined guardrails.

Goals

  • Smooth capacity utilization; reduce failed calls/throttling during spikes.

  • Improve developer UX (budget caps without hard stalls).

  • Create price signals that help OpenAI forecast demand and allocate GPUs.

Core Design (high-level)

  • Credits Ledger Service (append-only, double-entry): accounts, balances, holds, transfers, fills, refunds.

  • Matching Engine (optional market): maker/taker orders for credit-hours (or normalized token-units) scoped by model family and validity window.

  • Escrow/Holds: reserves on order placement; release on cancel/expiry/fill.

  • Org Policy Layer: per-org toggles: allow intra-org transfers only, allow inter-org, market opt-in, KYC tier required.

  • Quota Governor: merged limits = purchased capacity + transferred credits; preserves per-minute/second rate limit semantics to prevent abuse.

  • Billing Reconciliation: nightly settlement → existing invoices; fees shown as line items.

  • APIs

    • POST /credits/transfers (source, dest, amount, scope, expires_at)

    • GET /credits/balance (by model family & window)

    • Market (if enabled):

      • POST /market/orders (side, qty, limit_price, model_scope, window)

      • GET /market/book | GET /market/trades

  • UX

    • Console > Billing > “Credits & Capacity”: Transfer tab + Market tab.

    • Usage graph overlays: purchased vs transferred vs consumed.

Abuse & Risk Mitigation

  • Compliance: treat as store-of-value / money-transmission-adjacent. Require KYC/AML for inter-org transfers; restrict to approved regions; no consumer P2P.

  • Scope Constraints: credits are non-cashable, non-withdrawable, and expire; model-scoped to prevent “generic IOUs”.

  • Rate-Limit Invariants: no bypass of safety throttles; anti-wash trading.

  • Fraud detection: device/IP risk, velocity checks, linked-entity graphs.

  • Regulatory: clarify TOU; issuer is OpenAI, not e-money; consider credits-as-discount-rights to avoid deposit characterization (consult counsel).

Economics

  • Fees: maker/taker (bps), listing fee for large blocks, or simple spread.

  • Utilization Upside: reduces unspent credits; more stable GPU planning.

  • User Value: teams buy bursts instead of upgrading permanently.

MVP Slice (60–90 days, high level)

  1. Intra-org transfers only, no marketplace.

  2. Model-family scoping + 30-day expiry.

  3. Ledger, holds, reconciliation, audit trails.

  4. Console + API; audit exports (CSV).

  5. Compliance review for phase-2 inter-org pilot.

Future

  • Reservations (scheduled capacity windows).

  • Spot credits (discounted, preemptible).

  • Cross-product (Assistants, Batch, Fine-tuning capacity blocks).

    I’m happy to discuss edge cases (e.g., credit clawback on policy violations, org merges/splits, or disaster-recovery semantics). —Mike A.