GPU markets / inference pricing / agent compute

The compute market is becoming tradable.

A neutral map of how GPUs are priced, where compute is rented, how inference turns hardware into tokens, and how x402-style payments could let agents buy small units of compute directly.

Walk The FloorUse the toolsMap the providers

The Floor

checked 2026-07-12

LaneUnitRisk
Training cluster$/GPU-hrquota
Managed inference$/tokenlatency
Peer marketplacelive bidhost
Agent tool$/callpolicy

Open the four pits →

Signature

The Floor is the map you walk, not a price you copy.

Drop a training job, a latency-bounded chat path, a checkpointed batch, or an agent call onto four pits. The matching market lights up. The mismatched markets dim because they sell the wrong unit. Dated catalog examples stay dated. This is not procurement advice.

Walk The Floor

What the walk teaches

  • GPU-hour, token, live bid, and per-call settlement are different products.
  • Spot is a risk discount, not a cheaper version of reserved capacity.
  • A token price hides the GPU. A bid hides the SLA. A call hides the cluster.
  • Worksheets live under Tools. Quotes live on the provider page you open next.

Key facts

Verdict

Compute is now bought in four overlapping markets.

The buyer can rent raw accelerators, buy a managed model API, accept live marketplace risk, or pay for a narrow compute result. Those markets share hardware, but they optimize for different things. Training clusters optimize for synchronized capacity. Inference APIs optimize for latency, batching, and quality. GPU marketplaces optimize for price discovery. Agent-native services optimize for machine-readable, per-call settlement.

This site does not rank vendors. It explains the units and tradeoffs so a buyer can ask the right question: not "what is the cheapest GPU?" but "what is the cheapest reliable way to finish this workload under these constraints?"

The Floor

Walk four pits and route a ticket to the market that actually sells that unit.

Free calculators

Estimate GPU-hour cost, compare dated GPU prices, and model rough inference cost.

Hyperscalers

Enterprise controls, quota, regions, IAM, data gravity, and calculators.

GPU-first clouds

Clearer accelerator catalogs and cluster products for training and inference teams.

Peer and decentralized

Live supply, host choice, interruptible economics, bids, and benchmark-hour markets.

The map

Start on The Floor if you need the market structure before a spreadsheet. Use How Compute Is Priced to normalize GPU-hour, token, request, storage, and egress meters. Use The Marketplaces to understand the provider categories without vendor shilling. Use Inference vs Training Markets to separate long-running accelerator demand from low-latency serving economics.

Use the GPU Cost Estimator, GPU Price Compare, and Inference Throughput Cost Calculator when you need a working worksheet rather than another static explainer.

The agent page, Agent-Native Compute, intentionally stays in the compute lane. It links out to x402-specific references for payment details and focuses on what machine payments change for compute procurement.

What to verify before buying

  • The exact GPU model, VRAM, interconnect, region, and trust tier.
  • The billing unit and whether it changes by batch, priority, cached input, output, or interruption risk.
  • Storage, data transfer out, idle workers, cold starts, and support terms.
  • The measured cost per useful unit on your workload, not a generic provider benchmark.

Sitemap

How Compute Is PricedA buyer-focused guide to GPU-hour, token, spot, reserved, storage, egress, batching, and utilization pricing in compute markets.The GPU and Compute MarketplacesA vendor-neutral map of centralized GPU clouds, neoclouds, peer marketplaces, and decentralized compute networks.GPU Cloud Price Comparison: How to Read the TableHow to compare GPU cloud prices without mixing up per-GPU rates, node prices, marketplace risk, storage, egress, and inference throughput.Inference vs Training MarketsWhy model training, fine-tuning, batch inference, and real-time inference produce different compute markets and pricing models.Agent-Native ComputeHow x402, per-call inference, and machine-readable payment flows could let software agents buy compute autonomously.Buyer Guide: Choosing GPU or Inference ComputeA practical checklist for choosing GPU cloud, marketplace, decentralized compute, or managed inference for a workload.Compute Market GlossaryDefinitions for GPU marketplace, inference pricing, decentralized compute, x402 payments, batching, and GPU cloud terms.Sources and Pricing BibliographyAnnotated sources for ComputeMarket.io, including provider pricing pages, inference API pricing, decentralized compute docs, and x402 references.Compute Market FAQAnswers to common questions about GPU marketplaces, decentralized compute, renting GPUs cheaply, inference pricing, and agent payments.The FloorInteractive explainer of the four compute-market pits. Not a live quote.Free GPU and Inference Cost ToolsClient-side calculators and a dated provider price comparison table.GPU Cost EstimatorEstimate GPU rental cost from dollars per GPU-hour, hours, token volume, throughput, GPU count, and utilization.GPU Price CompareCompare dated, first-party GPU-hour examples for H100, A100, L40S, and RTX 4090 across providers.Inference Throughput Cost CalculatorEstimate rough self-hosted LLM inference cost per request from model size, context length, batch size, output tokens, and GPU hourly price.