GPU Cloud Rate Monitor – Runpod & Lambda Price Changes

GPU Cloud Rate Monitor

Turn official Runpod and Lambda rate cards into comparable, source-linked GPU pricing records. Keep the price unit, GPU count, memory allocation and effective date beside each number, then compare subsequent complete observations.

This Actor helps AI agencies prepare compute estimates, infrastructure buyers compare advertised offers, and research teams monitor rate changes. It reads public pricing pages without an account, API key, browser or paid proxy.

Start with the prefilled input. It returns the advertised Runpod H100 PCIe Community and Secure rates, usually two records. Each saved rate costs $0.002, plus Apify's $0.00005 Actor Start event at the default memory setting. Two saved rates therefore produce $0.004 in result events plus the start event; platform usage is separate. Unmatched filters, unchanged observations skipped by onlyChanges, and failed source responses do not produce result charges.

The distinction that matters

An hourly GPU price is not necessarily the hourly cost of an instance. Lambda's 8-GPU group publishes a price per GPU-hour; usdPerInstanceHour multiplies that price by eight. GPU count, VRAM per GPU, CPU/RAM allocation and storage text remain visible. RAM values keep their source unit—GB and GiB are not silently mixed.

Effective dates matter too. At the initial November 8, 2026 verification, Lambda's instance page advertised rates taking effect November 16, 2026, at 00:00 UTC, with current prices available only after login. Those public rows are labeled announced_future; the Actor does not retrieve or invent the logged-in price. The actual observed footer and parsed effective time travel with every Lambda row. Undated advertised rates are labeled advertised_undated, never “guaranteed current.”

Three useful inputs

Runpod H100 PCIe tiers

{"providers":["runpod"],"gpuModels":["H100 PCIe"],"runpodTiers":["community","secure"],"maxResults":5}

Lambda H100 instance sizes

{"providers":["lambda"],"gpuModels":["H100"],"lambdaGpuCounts":[1,2,4,8],"maxResults":5}

A40 compute estimate and saved changes

{"providers":["runpod"],"gpuModels":["A40"],"runpodTiers":["secure"],"computeHours":100,"comparePrevious":true,"snapshotName":"a40-budget-watch","maxResults":5}

gpuModels contains case-insensitive literal name phrases. It is a filter over the selected public rate cards, not a compatibility or performance search. An empty list selects every parsed model; use a sufficiently large maxResults for complete history. Provider values are runpod and lambda. Runpod tiers and Lambda GPU-count filters apply only to their respective provider. There are at most two pricing-page requests plus robots checks, with a shared 30–120 second deadline.

Output you can use

Field	Meaning
recordId, title	Stable rate identity: provider, exact GPU name, VRAM, tier and GPU count.
provider, gpuName, tier	Source provider, literal advertised model and commercial tier.
gpuCount, acceleratorAllocation	Whole-GPU count priced in this instance; fractional MIG offers are excluded.
usdPerGpuHour	Advertised price per GPU-hour, normalized to USD.
usdPerInstanceHour	Per-GPU rate multiplied by the explicit instance GPU count.
vramGbPerGpu	Published GPU memory in GB.
vcpusPerInstance, ramAmount, ramUnit, storageText	Published resource configuration, preserving GB/GiB and source storage wording.
resourceBasis	Runpod configuration card or Lambda instance table; not a capacity promise.
rateTiming, rateEffectiveAt, rateNote	Undated, future-announced or effective listed rate, with original source note.
computeHours, estimatedComputeUsd	Optional user hours multiplied by per-instance rate; no taxes, storage, networking or performance adjustment.
comparisonStatus, changedFields, previous	New, changed, unchanged or comparison not requested, with prior complete observation.
usdPerGpuHourChange, usdPerInstanceHourChange	Numeric deltas against that previous observation, or null on the first check.
sourceUrl, observedAt	Official source and fetch time.
availability, availabilityChecked	Always null and false. No capacity check is performed.

Download JSON or CSV from the dataset. SUMMARY reports parsed and filtered rate counts, source errors, unprocessed selections, output caps and history status. RUN_REPORT is a readable HTML version.

Repeatable comparisons

Enable comparePrevious to save a baseline only after all selected source requests and matching output succeed. Use onlyChanges on later runs to save new or changed rates. An unchanged repeat can correctly return zero rows.

History is isolated by provider selection, filters, GPU counts, maxResults, compute-hour assumption and snapshotName. Source errors, a deadline, a spending cap or an output cap cutting through matching results leave the previous baseline intact. A concurrent writer is detected before committing. Run the same snapshot sequentially. Disappearing offers are not reported as deleted, unavailable or sold out.

Source coverage and limits

• Runpod's official pricing page: structured on-demand Pod offers with matching configuration cards. Community and Secure tiers remain separate. The Actor cross-checks structured prices against their displayed card attributes. Serverless, clusters, spot prices, fractional MIG partitions and console capacity are excluded. Models without a structured offer are outside this feed.
• Lambda's official pricing page: the 1/2/4/8-GPU on-demand instance tables. Reserved cluster tables are excluded. Published future rates remain explicitly future; current rates behind login are outside scope. Lambda uses dollar-denominated public cloud rates; the provider also describes its published pricing as US dollars per hour.

Neither source promises that the shown configuration can be booked now. Rates can change after collection. The output is not a quote, profit forecast, resale recommendation or claim that different GPUs finish a workload in the same time. Compute estimates exclude sales tax, storage, network charges, negotiated discounts and other provider charges. Runpod configuration resources describe the published card, not every available host.

The Actor honors robots.txt, rejects arbitrary source URLs and stops further source requests on HTTP 401, 403 or 429. It does not use account cookies, private endpoints, paid proxies or access workarounds. A changed table or contradictory price fails visibly rather than emitting a fabricated rate.

Automations and AI tools

Save this input as a private Apify task and schedule it at a cadence appropriate for your purchasing workflow. Send the dataset to a spreadsheet, warehouse or workflow tool through your own Apify integration. For API or MCP use, supply the same small input and inspect SUMMARY.incomplete, rateTiming and the source URLs before acting on the results. This Actor reads rate cards; it never rents hardware or sends notifications itself.

For troubleshooting, open an issue on this Actor with the run ID, selected provider and filters. Do not include credentials.