Video cards are expensive because of a perfect storm: manufacturing costs have climbed with each new chip generation, AI demand has created unprecedented competition for the same production capacity, and Nvidia’s business has shifted so dramatically toward data centers that gaming is no longer the priority it once was. A top-tier consumer GPU now launches at $2,000, and street prices can double that.
Advanced Chips Cost More to Make
Every new generation of GPU uses smaller, more complex transistors that cost more to produce. The chips inside today’s high-end cards are manufactured by TSMC, the Taiwanese company that fabricates silicon for both Nvidia and AMD. A single wafer produced on TSMC’s 4nm process (the node used for current flagship GPUs) costs around $18,000 to $20,000. TSMC is expected to raise those prices by roughly 10% for 2025, pushing 3nm wafers above $20,000 each.
Each wafer yields a limited number of usable GPU dies. For Nvidia’s RTX 5090, one estimate puts the raw die manufacturing cost at around $290. That sounds modest until you add everything else: GDDR7 memory, the voltage regulation hardware that manages power delivery, the circuit board itself, the cooling system, assembly, and testing. Those components push the total bill of materials into the $500 to $600 range before Nvidia or its board partners add any profit margin, marketing costs, or retail markup.
This is a meaningful increase from previous generations, when simpler manufacturing processes and smaller dies kept production costs lower. The trend only moves in one direction: each new node is more expensive to develop and operate than the last.
AI Demand Is Eating the Supply Chain
The single biggest factor inflating GPU prices right now is artificial intelligence. Nvidia’s data center accelerators, like the H100 and its successors, are built on the same TSMC manufacturing lines and use the same advanced packaging technology as consumer GPUs. Demand for these AI chips is so intense that enterprises face lead times stretching close to a year just to secure an allocation.
That scarcity creates what analysts call “demand spillover.” Companies, AI startups, and researchers who can’t get their hands on dedicated data center hardware are buying the next best thing: high-end consumer cards. A workstation loaded with four RTX 5090s offers serious performance for training and running AI models, so these buyers are competing directly with gamers for the same limited supply. The RTX 5090 launched at a $2,000 MSRP but quickly hit over $4,000 on the secondary market, roughly doubling in price.
The bottleneck isn’t just raw chip production. Each high-performance GPU requires advanced packaging from TSMC, a complex assembly process that the company is struggling to scale fast enough. Every H100 that gets packaged is capacity that isn’t available for consumer chips, and right now, the customers paying for AI hardware are willing to pay far more per unit than gamers are.
Gaming Is No Longer Nvidia’s Priority
Nvidia’s business has undergone a dramatic transformation that directly affects how it prices and allocates consumer GPUs. In mid-2020, gaming was the company’s biggest revenue driver, accounting for 51% of total sales. Data centers made up just 25%. Today those numbers have essentially flipped: data centers generate about 90% of Nvidia’s revenue, while gaming has shrunk to roughly 5.5%.
To put concrete numbers on it, Nvidia’s most recent quarter brought in $68 billion in total revenue. Gaming contributed $3.8 billion of that. Everything from GeForce desktop GPUs to gaming laptops to the GeForce Now streaming service fits inside that slice. Data centers, meanwhile, pulled in over $62 billion.
When gaming represented half your business, you had every incentive to price competitively and keep shelves stocked. When it represents less than 6%, the calculus changes. Nvidia can price consumer cards at a premium, allocate more manufacturing capacity to data center products, and still post record revenue. There’s simply less economic pressure to make affordable GPUs a priority. Gaming revenue as a share of Nvidia’s total has fallen from about 17% in fiscal year 2024 to under 8% in fiscal year 2026, and nothing suggests that trend is reversing.
Prices Have Been Climbing for Years
The sticker shock didn’t start with AI. Flagship GPU prices have been creeping upward for over a decade. Nvidia’s GTX 780 launched at $650 in 2013. The GTX 1080 debuted at $600 in 2016 (actually a slight dip). Then the RTX 2080 Ti pushed to $1,200 in 2018, and the RTX 3090 hit $1,500 in 2020. The RTX 4090 launched at $1,600 in 2022, and the RTX 5090 arrived at $2,000 in 2025.
Each generation added complexity: ray tracing hardware, AI-powered upscaling engines, larger dies with more transistors. Those features cost real money to develop and manufacture. But the price increases also reflect Nvidia’s market position. AMD competes in the mid-range and upper-mid-range, but hasn’t released a true flagship competitor in the most recent generation, giving Nvidia essentially unchallenged pricing power at the top of the stack. Intel’s Arc GPUs have entered the market but only at budget and mid-range price points. Without serious competition at the high end, there’s no market force pushing prices down.
What This Means for Buyers
If you’re shopping for a GPU right now, the forces driving prices up are structural, not temporary. Manufacturing costs will keep rising as chipmakers move to smaller, more expensive nodes. AI demand shows no sign of slowing. And Nvidia’s business model is now built around data center revenue, with gaming as a side business.
The best value in the current market sits in the mid-range. Cards priced between $300 and $500 at MSRP still offer strong gaming performance and aren’t as affected by AI-driven demand spillover, since they lack the raw compute power that makes high-end cards attractive for machine learning workloads. Buying at or near launch, before secondary market inflation kicks in, also helps. The $2,000 to $4,000 gap on the RTX 5090 is largely a supply and demand problem in the aftermarket, not the result of the card costing twice as much to build.
Used GPUs from the previous generation can also be a practical option, especially when data centers and AI labs cycle out older hardware. The performance gap between generations is real but often smaller than the price gap suggests, particularly at resolutions below 4K.