Editor
Dinesh Raj chevron_right
Price a new graphics card in Malaysia this week and the number will not make much sense. Nvidia's flagship GeForce RTX 5090 launched here at RM10,390, yet the cheapest boards on local shelves now start closer to RM11,799, and liquid-cooled versions sit near RM18,000. This is not simply a Malaysian markup. It traces straight back to the same artificial intelligence data-centre boom the country is spending billions to attract.
The gap between what a card should cost and what people actually pay has become the defining fact of the 2026 PC market. According to TechSpot's mid-2026 pricing tracker, the RTX 5090 has been trading at roughly 65 percent above its US$1,999 list price, with custom models topping US$3,000 and some listings well beyond that. The step-down RTX 5080, meant to be the sensible pick at US$999, carries a premium of its own of about 25 percent. When retailers do get stock near list price, as Tom's Guide has documented, it tends to sell out within minutes.
The reason is not gamers or scalpers this time. It is memory. Modern graphics cards need fast GDDR7 memory, and that is the exact component the AI industry is buying in bulk. Astute Group reports that hyperscalers are on track to spend close to US$700 billion on AI infrastructure in 2026 and are placing open-ended orders for every chip of memory the fabs can produce. That leaves consumer cards at the back of the queue. AMD raised the price of its graphics-and-memory kits to board partners by around 10 percent from July 2026, and Nvidia followed, with the heaviest increases landing on cards that carry 16GB or more of memory.
Here is where it gets uncomfortable for Malaysia. The country is racing to become a regional data-centre hub, with Cyberjaya and Johor drawing tens of billions of ringgit in AI infrastructure investment. That build-out is real economic growth. It is also, at the component level, part of the very demand that is pricing Malaysian video editors, 3D artists and gamers out of a decent GPU. The same appetite for AI memory that fills our industrial parks helps empty the shelves at our computer stores.
For anyone here planning a build, the practical advice is unglamorous. There is no quick fix coming. Intel's chief executive has said there will be no real relief until 2028, because new memory plants from Micron, Samsung, SK Hynix and TSMC will not add meaningful supply before late 2027. Buying on speculation, hoping prices fall soon, is a losing bet. If you need the machine now, an RTX 5080 close to list price is a far better deal than an inflated 5090, and a well-priced last-generation card can still edit 4K footage without complaint.
None of this is unique to graphics cards. We have watched the same squeeze reach the chips inside our phones, and it runs through the whole race to build faster silicon. The AI boom is not only reshaping what our computers can do. It is quietly deciding what the rest of us can afford to buy.
Images courtesy of Rafael Pol and Vadim Bogulov on Unsplash.
