Where the data comes from

1. Newegg catalog through Rakuten Advertising: Pending for current offers. Rakuten technical access and Newegg advertiser/feed approval are separate pending steps, and neither has been supplied for this property. This review found no general cache limit in the agreement. That absence is not permission to retain price history: the licence is limited and revocable, and the agreement does not provide the affirmative historical-use right this product needs. No Newegg price is collected or displayed while those permissions remain pending. Terms: Rakuten Publisher Membership Agreement and Rakuten data-feed guidance.

2. Best Buy Products API: Pending for this product. The public terms describe attributed display and a maximum 72-hour cache, while also limiting third-party-benefit price analysis. Best Buy remains pending for this cross-retailer use until written approval covers the intended presentation. Its data cannot enter retained history or drive historical analysis. Terms: Best Buy API legal terms.

3. eBay Browse API: Reject. The licence requires eBay content to be visually isolated from third-party listings. That rule forbids the mixed retailer comparison table this product needs. The licence also limits listing cache age to six hours, requires deletion when content disappears, and prohibits using eBay content to suggest or model prices. Terms: eBay API License Agreement.

4. eBay Marketplace Insights API: Reject as unavailable and unsuitable. Sold-item history is a limited-release restricted API that requires business-unit approval. Its market-trend and pricing uses require express written consent, and the available history is only 90 days. Terms and access status: Marketplace Insights overview and the eBay API License Agreement.

5. Amazon Creators API: Pending. No working access has been supplied for this property. Amazon's current documentation gives Offers a one-hour cache duration and most other product fields a one-day duration. Amazon access and the intended current-card use remain pending; documentation of cache duration is not approval for this product. Amazon data may not enter retained history, price comparisons, valuation, derived trends, or buy/wait logic. Amazon Creators API best programming practices, reviewed 2026-07-29.

6. Walmart affiliate catalog and Walmart.io Product Lookup API: Reject for history; pending for current display. API access depends on an approved Impact publisher relationship. Product information must be refreshed within 24 hours, and Walmart bars redistribution of its data feed to third parties. No retained series or derived verdict may depend on it. Terms: Walmart affiliate terms.

7. B&H Photo Video affiliate program: Reject as a data source. B&H has useful component inventory, but the program exposes static links and banners rather than a product feed or API. It may later supply outbound links, but it cannot support a measured comparison table. Terms: B&H affiliate agreement.

8. Adorama affiliate program: Reject. Its agreement prohibits including price information in product descriptions and also bans caching. A price product cannot use a source that withholds the price. Terms: Adorama affiliate agreement.

9. Newegg Business through Rakuten: Pending. This is a separate merchant program with strong category fit. It may become another current-offer source after separate approval. The reviewed page did not establish durable-retention rights, so no history use is granted. Terms and program page: Newegg Business affiliate program and the Rakuten agreement.

10. Target through Impact: Pending, not cleared. The partner pages returned an empty client-rendered shell, and neither useful component depth nor operative retention terms could be verified. Absence of a visible restriction is not permission. Terms URL could not be read; candidate page: Target Partners.

11. Micro Center: Reject until a primary publisher path exists. The inventory would fit, but no official self-service affiliate or data program was confirmed. The site and community pages blocked the review. Terms URL for a publisher data program was not found; candidate site: Micro Center.

12. ShopBLT: Reject. The domain blocked every automated request, including robots.txt, and the repeated claim that it offers a public price-list file could not be verified. We will not build on a rumour or bypass the block. Terms URL was not reachable; candidate site: ShopBLT.

13. Provantage: Reject. No publisher-facing affiliate or data program was found, and primary pages blocked the review. Terms URL for this use was not found; candidate site: Provantage.

14. Insight: Reject. Only a business-reseller relationship was found, not a publisher price feed with display and retention rights. Terms URL for this use was not found; candidate site: Insight.

15. CDW: Reject. The available path is a B2B reseller or customer integration, not a publisher-facing comparison feed. The API portal blocked the review. Terms URL for this use was not established; candidate portal: CDW API Connect.

16. Connection: Reject. No publisher-facing affiliate or data-feed program was found. Terms URL for this use was not found; candidate site: Connection.

17. Rakuten Product Search API: Pending behind merchant approval. Rakuten says the API exists to gather product information and make comparisons, supports up to 5,000 results, and permits 100 calls per minute. The response schema and price fields were not readable in the client-rendered developer portal, and access still depends on approved advertiser relationships. Terms and product documentation: Product Search API guidance and the Rakuten agreement.

18. Impact.com catalogs: Pending, not cleared for retention. Catalog access depends on brand approval and can be restricted per partner. Publisher-side cache and retention terms were not located, so these catalogs are not a history source. Terms information: Impact catalog documentation.

19. CJ Affiliate Product Search API: Pending, not cleared for retention. The API is available to CJ publishers, but useful feeds depend on advertiser relationships. The reviewed Publisher Service Agreement did not yield a product-feed retention clause. Terms and product information: CJ Product Search API announcement.

20. Awin Create-a-Feed and Partner Feed API: Pending, not cleared for retention. The account path exists, but US component-merchant coverage and publisher-side retention rights remain unverified. Terms and feed information: Awin product-feed documentation.

21. Skimlinks Merchant API: Reject as a price source unless product data is proven. The existing account is domain-bound and cannot be reused across properties without permission. More importantly, the developer site did not establish that the API returns product-level prices at all. Terms URL and price capability could not be verified; candidate documentation: Skimlinks developers.

22. Geizhals: Reject. The available integration appears to let merchants submit their own offers, not let a publisher retrieve an aggregated US price set. No official public data-distribution licence was found, and the source does not fit a US launch. Terms URL for the proposed use was not found; candidate site: Geizhals.

23. idealo: Reject. Like Geizhals, the identified API direction is merchant ingestion rather than third-party comparison-data distribution. It is also not a US-launch source. No official public developer licence for our use was found; candidate site: idealo.

24. PCPartPicker: Reject. Its robots.txt disallows our crawler, states ai-input=no, and reserves Article 4 DSM rights. We did not fetch its product pages and will not use it as a source or inspection target. Rights signal: PCPartPicker robots.txt.

25. US Bureau of Labor Statistics Public Data API: Use for retained monthly index history. BLS data is US federal public-domain material, and the API is keyless at the base tier. The eleven series below were retrieved 25 July 2026 and carry reference months through June 2026. This is our derived category analysis, not a retailer feed and not per-SKU history. BLS.gov cannot vouch for the data or analyses derived from these data after the data have been retrieved from BLS.gov. Terms: BLS API terms and BLS copyright information.

26. FRED: Reject in favor of the primary BLS source. FRED mirrors the BLS series but adds a separate API key and terms layer. Its legal pages could not be primary-verified, and search-indexed language appeared more restrictive. There is no product benefit in accepting that uncertainty when BLS is the original public-domain source. Terms pages attempted: FRED legal and FRED API terms.

27. Stanford DAM memory-prices dataset: Reject. Of 705 downloaded rows, 214 are marked source=keepa. Those rows cover the still-updating DRAM portion from August 2024 and NAND portion from October 2016 through July 2026, so the live layer is Amazon-derived. The older McCallum portion stops in July 2024 and has no stated licence. Rights and provenance page: Stanford memory prices.

28. John C. McCallum memory prices: Reject as a live source. The original history survives through archives, but the former domain is now a Swedish casino affiliate site and must not be linked as an authority. The archived dataset also lacks a usable current licence. Archive index: Internet Archive snapshots of the former dataset.

29. Epoch AI machine-learning hardware data: Limited reference use with attribution. The dataset is CC BY and clean for credited reuse, but only 58 of 176 rows have a release price, only 24 are consumer-class GPUs, and the newest consumer GeForce entries are the RTX 4090 and RTX 4080 from September 2022. It can support a thin launch-price reference or backtest, not current GPU pricing. Licence and data page: Epoch AI machine-learning hardware.

30. Our World in Data historical memory and storage cost: Limited long-run backdrop use with attribution. The CC BY series is an aggregate dollars-per-terabyte history derived from the McCallum series. It is useful as historical context but is not a per-model record and cannot support a current product verdict. Licence and chart: historical cost of memory and storage.

31. TrendForce and DRAMeXchange: Reject. The terms grant personal, non-commercial use and prohibit reproduction, derivative works, unapproved data use, and crawling. That excludes its DRAM, NAND, and SSD price series from a commercial comparison and verdict product. Terms: TrendForce terms of use.

32. InSpectrum: Reject. It sells subscription databooks and does not publish a free licensed price series for reuse. “All rights reserved” is not a grant. Terms or product page: InSpectrum.

33. Backblaze Drive Stats: Reject as irrelevant. The dataset is reusable with attribution, but it contains drive-reliability records and no price field. A clean licence does not make it a price source. Terms and data page: Backblaze Drive Stats.

34. United-Compute/gpu-price-tracker: Reject. The repository has no licence file and its product links are overwhelmingly Amazon short links. It fails both the licence requirement and the no-Amazon-lineage rule. Rights and provenance: GitHub repository.

35. HardwareDealsCo/gpu-deals: Reject. The repository code is MIT-licensed, but the retailer source behind its price data is undisclosed. A code licence does not clear the collected data, and undisclosed lineage cannot be shown to be non-Amazon. Rights and provenance: GitHub repository.

36. gradedSystem/GPU_Price_Specs: Reject. The repository is unlicensed and has been stale since September 2024. It supplies neither current coverage nor reusable rights. Rights and provenance: GitHub repository.

37. Hugging Face exalsius/gpu-prices: Reject. The candidate dataset is empty, so it cannot provide usable retail-price evidence. Dataset page: Hugging Face dataset.

38. Hugging Face afhubbard/gpu-prices: Reject as irrelevant. It covers cloud GPU rental prices, not retail PC components. Dataset page: Hugging Face dataset.

39. Kaggle component-price datasets reviewed as a group: Reject until each licence is primary-verified. Kaggle rendered the relevant licence badges client-side, so the review could not establish reusable rights for the candidate datasets. No Kaggle row enters the product on a guessed licence. Platform terms: Kaggle terms.

40. Intel RCP: Limited, hand-curated CPU anchor. Intel publishes Recommended Customer Price as guidance, typically for 1,000-unit direct-customer quantities, not a consumer offer. Intel blocked every automated request during diligence, including robots.txt, so we will not build a scraper. A dated, manually sourced RCP can be shown as a reference, clearly separated from street price. Official definition: Intel RCP explanation.

41. AMD direct-store MSRP: Limited, hand-curated CPU and partial GPU anchor. AMD's store distinguishes MSRP from its current direct-sale price, but every direct fetch during diligence reset or timed out. Coverage is limited to products AMD sells itself, and automated collection is not cleared. Terms could not be verified; candidate store: AMD US store.

42. NVIDIA launch pricing: Limited to Founders Edition references. Official launch materials can anchor a subset of GPUs, but partner cards from ASUS, MSI, Gigabyte, Zotac, and others have no NVIDIA manufacturer MSRP. The current marketplace timed out and cannot support an automated feed. Terms for that marketplace were not verified; candidate site: NVIDIA marketplace.

43. Corsair: Reject as an MSRP source. No distinct MSRP field was found, and Corsair's reviewed terms prohibit scraping or extraction. Store price is not MSRP. Terms URL could not be preserved in the diligence output; candidate site: Corsair.

44. ASUS: Reject as an MSRP source. The US store exposes a current store price, not a distinct launch MSRP, and static HTML did not expose a price suitable for a cleared feed. Terms for automated collection were not verified; candidate store: ASUS US.

45. MSI: Reject as an MSRP source. The store price is a current merchant price, not a separate manufacturer launch anchor. No collection terms were primary-verified. Candidate store: MSI US store.

46. Gigabyte: Reject as an MSRP source. The vendor provides reseller location rather than a direct store with a distinct MSRP field. No reusable automated source or operative terms were established. Candidate site: Gigabyte.

47. ASRock: Reject as an MSRP source. The directly fetched site led to resellers and did not publish a manufacturer price anchor. No data-reuse terms for this use were established. Candidate site: ASRock.

48. Seasonic: Reject as an MSRP source. No direct store or distinct MSRP source was found, and the vendor's own domain was not fetched during diligence. Terms remain unverified. Candidate site: Seasonic.

49. Kingston: Reject as an MSRP source. No direct-store MSRP anchor was found, and primary collection terms were not checked. A search summary is not enough to clear a source. Candidate site: Kingston.

50. G.Skill: Reject as an MSRP source. No direct store or distinct MSRP field was found, and the vendor domain was not primary-verified. Terms remain unverified. Candidate site: G.Skill.

51. SK Hynix: Reject as an MSRP source. No direct consumer store or distinct launch-price anchor was found, and primary terms were not checked. Candidate site: SK Hynix.

52. Samsung: Reject as an MSRP source. The vendor store exposes current prices, not a distinct reusable MSRP series. Automated-collection terms were not established. Candidate site: Samsung US.

53. Crucial: Reject as an MSRP source. The site returned an explicit web-application-firewall rejection during diligence, and no distinct MSRP field or collection permission was established. Candidate site: Crucial.

54. Western Digital: Reject as an MSRP source. The direct store exposed unrendered bindings rather than a verified distinct MSRP field. It is a current store, not a launch-price dataset, and collection terms were not checked. Candidate site: Western Digital.

55. Seagate: Reject as an MSRP source. The direct store likewise returned client-side bindings without a verified distinct MSRP field. No automated collection permission was established. Candidate site: Seagate.

Each link below is the official BLS time-series endpoint. Values are monthly index points. They are not dollars, do not describe a specific SKU, and cannot be compared as if an index level of 180 meant a product cost $180.

The weak spots are part of the result. Neither PSU series measures retail PC power supplies closely enough to stand in for a PSU price history. They mix products with different demand and cost structures, so they are context, not a verdict input.

The two motherboard proxies also disagree sharply in the June 2026 snapshot. Printed-circuit-assembly manufacturing is 50.4% above its 12-month mean, while bare printed-circuit-board manufacturing is 8.6% above its mean. We show both figures with their series names. We do not average them into a smoother number.

The semiconductor-manufacturing PPI covers a broad upstream industry. It is not “the RAM price,” even when it happens to move in the same direction as a retail RAM series. We label it with its official BLS title and never substitute it for a retailer or per-SKU observation.

BLS supplies decades of lawful monthly category-index history. It supplies no product listing, retailer offer, or per-SKU price. Our per-SKU history starts with our first eligible retained observation. Until enough time passes, the product state says not enough data to judge and names the real window. We do not present Stanford's Keepa-contaminated rows, an Amazon tracker, a marketplace API cache, or someone else's scraped repository as our own product history.