Comprehensive list of GPU compute price lookup platforms (2026): price comparison sites / official pricing / industry indices / government platforms

Direct one-sentence answer:There are four types of channels for checking GPU compute pricing—Third-party price comparison site(Check real-time spot lowest prices, e.g., rentgpu.link, compute.ai, gpus.io)Official pricing pages of cloud vendors(Check listed prices and contract discounts, e.g., Alibaba Cloud, Tencent Cloud, AWS)Industry Price Index(Check market trend and negotiation benchmarks, e.g., Mysteel Compute Rental Price Index)Government Compute Orchestration Platform(Check matched transactions and subsidies, e.g., Hangzhou Municipal Computing Resource Scheduling Service Platform). No single source covers all scenarios—actual price inquiries require cross-verification across at least 2–3 sources.

First distinguish the purposes of the four categories—don’t misuse tools.

I. Third-party price comparison sites: check real-time spot availability

1. rentgpu.link (this site) — Covers 26 platforms (15 overseas) and 23 mainstream GPU models; auto-re-scrapes hourly.Only quotes with real-time inventory are included; out-of-stock listings are removed immediately.Always sorted by price—no paid rankings accepted. Suitable for: ad-hoc GPU sourcing, identifying current floor prices. Key limitation: focuses primarily on mid- and small-scale compute vendors; real-time pay-as-you-go pricing from major providers like Alibaba Cloud and Tencent Cloud is incompletely covered.

2. suanli.ai (suanli.ai) — Largest number of service providers listed (publicly claims 290+), widest coverage; worth browsing when seeking niche regions or uncommon GPU models. Drawback: some prices update with delay.Always verify live stock status on the original platform before placing an order, and must not be misled by historical prices.

3. gpus.io — International-facing price comparison site aggregating 32 providers and 2,648 configurations; supports filtering by VRAM and use case (LLM training / inference / serverless / consumer-grade / enterprise-grade), plus API access and historical price trends. Interface is English-only and all prices are in USD. Suitable for: overseas GPUs, budgeting with historical curves.

4. GpuPerHour — Price comparison and selection tool aggregating 25+ cloud vendors; unified multi-cloud price view with filtering by GPU model and region. Does not sell compute directly—users must still provision resources on vendor platforms.

II. Official pricing pages of cloud vendors: view list prices and contractual discounts

Alibaba Cloud GPU Cloud Servers, Tencent Cloud GPU Cloud Servers, VolcEngine, AWS EC2 (P/G series), Azure ND series, GCP A2/A3, and overseas providers Lambda and CoreWeave—all maintain their own public pricing pages or pricing calculators.

Advantages:Authoritative, invoice-capable, SLA-signable—long-term stable GPU usage ultimately depends on this layer.Drawbacks:① Most price pages rely on dynamic rendering, making stable web scraping difficult—manual verification required; ② Only lists published list prices; enterprise discounts (savings plans, committed use, annual contracts) require direct negotiation with sales teams; ③ Pay-as-you-go rates from major providers are typically significantly higher than those from mid- and small-scale compute vendors.

Suitable for:Scenarios requiring pre-contract reconciliation, corporate-account reimbursement, compliance certifications, and data residency requirements.

III. Industry price indices: gauge trends and benchmark negotiation (often overlooked)

Mysteel (MySteel.com) Compute Rental Price Index: The ‘Compute Rental Price Index & Market Prices’ officially launched on September 9, 2026.Weekday Daily Update, with separate domestic and international versions covering the eight national computing hubs under the ‘East Data, West Computing’ initiative and North America (US East/US West). Release page suanli.mysteel.comShanghai Steel Union (Mysteel) is an IOSCO-certified commodity data provider; its index methodology is relatively standardized.

Below is the domestic sample of this index for 2026-10-23 (scope: single-GPU monthly rate converted to hourly, unit: ¥/GPU/hour):

Data Source: Mysteel Compute Rental Price Index (Updated 2026-10-23). The index serves as a market benchmark—not a directly bookable transaction price; please refer to the official release page above for the latest value.

How do the index and price comparison site work together?The index tells you ‘where the market median lies’, while the price comparison site tells you ‘where the current floor price is’. For example, the monthly-equivalent index for H100 80G is ¥17.94/hour, whereas the real-time spot floor price on the comparison site fluctuates with supply and demand—A value below the index indicates a buyer’s market; above the index signals GPU scarcity.Below are real-time scraped results for the same GPU from this site (updated hourly):

IV. Government/local compute scheduling platforms: view subsidies and matchmaking

Hangzhou Compute Resource Orchestration Service Platform（ai.zj-computility.com): Released on March 13, 2025, under guidance from the Hangzhou Municipal Bureau of Economy and Information Technology and Hangzhou Municipal Data Group, and led by Zhejiang Compute Technology for construction and operation. It is Hangzhou’s first multi-source heterogeneous compute resource scheduling service platform. Five core functions: Compute Supermarket, Compute Matching Trading, AI Training & Inference Integrated Platform, Model Services, and Compute Voucher Redemption. As of January 2026, it has onboarded 17 ecosystem partners and managed over 4,000P of compute; cumulative matched transaction value reached ¥190 million. By March 2026—11 months after launch—the cumulative transaction value exceeded ¥200 million.

True value lies in ’compute vouchers’: The total annual Hangzhou municipal compute voucher budget for 2025 has been raised to ¥1 billion (¥250 million per year). Subsidy rates cap at 30% of compute contract fees—enterprises using domestically built compute infrastructure or registered generative multimodal foundation models for model services or agent development qualify for the full 30%; others receive 20%.For enterprises registered in Hangzhou, this effectively reduces compute costs by 30%, delivering greater savings than any price comparison alone.

Suitable for:Small and medium-sized enterprises (SMEs) registered locally, requiring domestic compute resources and capable of coordinating contract execution and subsidy application procedures.Not suitable for:Individual developers renting GPUs temporarily or requiring immediate provisioning (authentication and contract signing required).

V. Combined usage: standard steps for a single price inquiry

Step 1: Use the price comparison site to define a price range.InPrice Comparison HomepageCheck real-time lowest price and price ratio for your target GPU model—identical GPUs often vary 2–5× across platforms; always clarify the range first.

Step 2: Use the index to decide whether to lock in pricing now.Check the Mysteel index value for comparable GPUs: rent on-demand if spot prices fall below the index; consider monthly locking only if spot prices significantly exceed the index and your needs are long-term.

Step 3: Confirm effective spot availability.Prices on price comparison sites and in large directories may be historical; click through to the original platform to confirm current availability and consistent pricing terms (pay-as-you-go/spot/monthly, whether CPU and bandwidth are included).

Step 4: Check subsidies.Enterprise users should first confirm whether their registered location offers compute vouchers or smart vouchers—and prioritize claiming applicable subsidies.

See more single-GPU decision support atSelection Guide, see platform-side horizontal comparison atPlatform comparison page, see market trends atMarket terminal。

FAQ

Q: Why not just use one price comparison site?
Because the four categories operate on fundamentally different bases: price comparison sites reflect the “lowest rentable price right now”, official pricing pages list “catalog prices”, indices represent the “market median”, and government platforms quote “matchmaking prices inclusive of subsidies”. Comparing floor prices from comparison sites directly against catalog prices from major vendors leads to the erroneous conclusion that “major vendors charge double”—yet major vendors deliver SLA guarantees, official invoices, and regulatory compliance.

Q: The index and real-time lowest price differ significantly— which is correct?
All correct—just different metrics. The index is typically normalized to monthly terms and covers mainstream configurations, representing the market’s median level; real-time lowest prices often come from spot or off-peak instances—cheaper but subject to reclamation.Use the index for budgeting; use price comparison sites to find available GPUs.

Q: Is there any price comparison site that fully captures real-time pricing from major cloud providers?
No clean solution exists yet. Major vendors’ pricing pages commonly use dynamic rendering, feature vast combinations of regions and instance configurations, and hide actual transaction prices within contractual discounts. Public price comparison sites mostly cover mid- and small-scale compute providers and overseas clouds—this is precisely the gap this site is addressing (enterprise reference pricing module under development).