Why are graphics card prices so high?

There is a real answer to part of this question and an honest limit to the rest. The measured part is what the US government's factory-level and consumer indexes have done. The limit is that no source we hold can tell you how much of the difference sits in chip supply, product mix, distribution or shop margin.

closely matched indexes2

current toJune 2026

evidence current to26 July 2026

Retail pricenot publishedWhy

A shop's price is the shop's data, and showing it needs that shop's permission. No shop has approved a price feed for this site: Rakuten Product Catalog technical approval, and the separate Newegg advertiser-feed approval, neither of which has been supplied for this property. Rather than copy prices we have no right to publish, we left them out and built the parts of the job that do not need them.

Every source we use, limit or refuse, each with the reason.

The measured part

What the factory-level indexes are doing

These are the series that track graphics cards more closely than a general shopping basket does. Each figure is the index level against its own previous twelve reported months.

2 closely matched index series for graphics cards
What it measures	How closely it matches	Latest month	Index level	Against its own 12-month average
Semiconductor and related device manufacturing (PPI)	good	June 2026	29.695	-0.4%
Integrated circuit packages, including microprocessors (PPI commodity)	good	June 2026	31.201	+3.2%

Source: US Bureau of Labor Statistics, public domain, read 2 August 2026. Where two series for one category disagree, both are shown. Averaging a disagreement away would delete the finding, and the disagreement is often the most interesting thing on the page.

The honest limit

What we cannot tell you, and why

No breakdown of the gap is published here

Our sources can show a gap between a factory-level index and what people report paying. They cannot say how much of that gap sits in chip supply, in which products are being made, in module assembly, in distributor stock or in shop margin, because no source we hold reports those layers separately.

Picking one of them and naming it would be a story rather than a measurement. Plenty of articles will give you that story. We would rather show you the two numbers and say plainly that the distance between them is unexplained by anything we can cite.

In full

The evidence in detail

One GPU market, very different shelf prices

The current retail picture is split by model. A US tracker updated July 24, 2026 listed an RTX 5090 at $3,999 against a $1,999 launch MSRP, an RTX 5080 at $1,289 against $999, and an RTX 5060 Ti 8GB at $359 against $379. The same snapshot therefore contained a flagship at twice MSRP and a lower-tier card below MSRP. It is an affiliate-supported deal tracker, and each offer can change, but the cross-model spread is more informative than a single “GPU prices” average. Tom’s Hardware, updated July 24, 2026.

The BLS data is much calmer than that flagship example. Integrated circuit packages, including microprocessors, the best upstream proxy retained for GPUs, was 31.201 in June 2026. That is 3.2% above its trailing 12-month mean and 4.4% above June 2025. The broader Semiconductor and related device manufacturing index was 0.4% below its trailing mean and 0.6% below June 2025. BLS series WPU117839 and PCU334413334413, June 2026 observations fetched July 25, 2026.

Those indexes measure packages and semiconductor manufacturing prices, not graphics cards. They cover products far beyond gaming GPUs and omit the board, video memory, cooler, power delivery, distribution and retail layers. Their value is negative evidence: there is no broad upstream increase remotely like the 100% MSRP premium in the cited RTX 5090 offer. The flagship premium therefore cannot be described as a simple reflection of semiconductor manufacturing prices as a whole. BLS series WPU117839 and PCU334413334413 metadata, fetched July 25, 2026.

Availability is part of the price

The July 24 retail snapshot reported 8GB RTX 5060 and 5060 Ti variants near MSRP while describing higher-tier RTX 5070 Ti and RTX 5080 supply at reasonable prices as poor. That evidence supports a model-specific availability explanation. It does not prove why the inventory differs, and it does not establish a premium for every board partner or country. Tom’s Hardware, updated July 24, 2026.

This is why a GPU average can mislead. The buyer cannot substitute an entry card for a flagship while keeping performance, memory capacity and workload constant. A market can contain normally priced lower-tier cards and scarce high-tier cards at the same time. The current tracker demonstrates that exact split through its MSRP comparisons; the BLS package index shows only modest upstream elevation. Read together, they locate the dramatic part of the story closer to particular finished products and their channels than to a uniform increase in integrated-circuit packages. Tom’s Hardware, July 24, 2026; BLS series WPU117839, June 2026.

The whole-PC backdrops are quiet too. Electronic computer manufacturing was 0.1% above its trailing 12-month mean in June 2026. The consumer index for Computers, peripherals, and smart home assistants was 1.0% above its mean and 0.8% below June 2025. These broad categories do not isolate add-in graphics cards, but neither shows a system-wide move resembling the cited flagship markup. BLS series PCU334111334111 and CUUR0000SEEE01, June 2026 observations fetched July 25, 2026.

A test that can actually change the conclusion

For a high-tier GPU, the first condition is repeated availability of the same model near its launch MSRP across multiple sellers. One temporary listing is not enough, and a lower-tier card near MSRP does not resolve a flagship shortage. A model-level retail record should record the exact GPU, memory configuration, board variant, seller, stock state and timestamp. Only then can a premium be separated from a short promotion or a single outlier listing.

The upstream condition is different. If WPU117839 moved materially away from its own recent mean across several observations, that would strengthen a package-cost explanation for movement across many GPU models. If the index stayed close to normal while only selected cards carried large premiums, product-specific supply and channel conditions would remain the better-supported explanation. This is a conditional test, not a claim about the next observation. BLS series WPU117839, monthly observations through June 2026 fetched July 25, 2026.

Memory should be tested separately rather than folded into the package proxy. SK hynix’s July 8, 2026 product description places GDDR7 in graphics and high-performance compute markets while distinguishing it from HBM, server DDR5 and other DRAM products. That supports treating video memory as its own supply line. It does not provide GDDR7 contract prices or quantify its share of a finished card’s retail price. SK hynix Newsroom, July 8, 2026.

What remains unknowable from these indexes

We cannot use the BLS data to declare a particular GPU cheap or expensive. There is no model mix, performance measure, VRAM capacity, MSRP, inventory count, seller margin or regional tax in either upstream series. The external price snapshot is current and specific, but it is not retained first-party history and cannot establish how long an offer persisted.

We also cannot reduce a flagship premium to one cause. The evidence rules out a similarly large movement in the broad package and semiconductor indexes, but it does not distinguish allocation, board supply, memory supply, retailer inventory or buyer demand. Today’s honest answer is segmented: some lower-tier GPUs are near MSRP, selected high-tier cards remain far above it, and the broad upstream indexes are only modestly elevated or flat. Tom’s Hardware, July 24, 2026; BLS series WPU117839 and PCU334413334413, June 2026.

Where next