GPU price history: what we have so far

Retail GPU history: not enough data to judge. Public upstream history: available from 2017-01 through 2026-06. Evidence cutoff: 2026-07-26. BLS WPU117839, fetched 25 July 2026 and source registry, reviewed 2026-07-25.

Daily first-party price collection is active, but the current trailing record is still too short and does not safely assign every retained identity to a GPU category. The monthly BLS indexes for integrated-circuit packages and semiconductor manufacturing remain useful long-run context, but they cannot be relabelled as GPU prices. Source registry, updated 2026-08-04.

The two retained upstream lines

The first row is calculated from BLS WPU117839 and the second from BLS PCU334413334413. Both files were fetched 25 July 2026, and their March through June 2026 values are provisional.

The integrated-circuit-package file reached its retained low of 29.538 in August 2025 before reaching 31.201 in June 2026. The broader semiconductor file also reached its retained low in August 2025, at 29.392, before ending at 29.695 in June 2026. Similar low dates do not make the two indexes independent confirmation of a retail GPU cycle. BLS WPU117839 and BLS PCU334413334413, fetched 25 July 2026.

Why this is not a graphics-card price chart

A retail GPU observation needs the chip, board partner, partner variant, memory capacity, condition, merchant, and time. The BLS indexes contain none of those fields. They also do not isolate cooler design, board power, retailer stock, launch price, or a temporary promotion. Product-identity and source methodology, published 2026-07-26.

The public index window is long enough to describe those indexes retrospectively. The retail record is collecting but has not passed the time, reading, and coverage floors shown above. Its status is therefore not enough data to judge, with no promised completion date. Source registry, updated 2026-08-04.

Why there is no verdict on the history

The no-lookahead recommendation backtest covered 1,252 decision points across 11 BLS series. The three-month wait candidate was contradicted in 29 of 48 testable signals, 60.4%, above the declared 40% limit. The buy candidate was contradicted in 62 of 118 three-month signals, 52.5%. Those error rates are why this page describes the recorded indexes but gives no buy-or-wait advice. Verdict methodology, run 2026-07-26.

A future retail state would require at least 14 days, 10 usable price readings, and 60% coverage. Crossing those thresholds would not change the failed recommendation test. Price-state methodology, published 2026-07-26 and verdict test, run 2026-07-26.