Anyone who has been around in PC gaming for a long time will know that every time a new high-end graphics card gets launched, supply is never enough to meet demand. In the case of the GeForce RTX 5090, board partner MSI says that its cards will have limited availability and it's down to an insufficient number of GB202 GPUs from Nvidia.
While this news probably isn't a shock to any graphics card enthusiast, it's worth noting that this isn't some rumour or leak—it's an official statement by MSI, one of Nvidia's key GPU partners, as reported by IT Home (via Wccftech).
As with any manufactured product, the overall availability can only be as good as the weakest link in the supply chain. In this instance, according to MSI, it's the number of GPUs being provided by Nvidia.
MSI, Asus, Gigabyte, and all other AIB vendors purchase graphics processors from Nvidia, which then distributes them mostly from its centres in Hong Kong. However, one can't simply rock up and ask for 100,000 chips—orders need to be placed well in advance and then Nvidia will allocate processors based on a number of factors, such as the size of the order, relationship with the partner, what deals are in progress, and so on.
It's not just MSI that is struggling to meet demand. Zotac Korea says that there's no chance of any 5090s being available (via Videocardz) until early February and that, as things currently stand, there is no confirmed date for the release of its RTX 5080 models.
This all tallies with the remark from UK retailer Overclockers saying that it will only have a 'single digit' number of 5090 cards available when the GPU is released for purchase.
So what to make of this? Is Nvidia deliberately constraining the supply of its GB202 chips, to charge board vendors more for each tray of processors they order? Or is it down to the fact that at 750 mm2 in size, each silicon wafer isn't going to produce many fully working dies, even if the yields are very good, so TMSC just can't make enough of them?
Even if the former isn't remotely true, the latter will be to a certain degree, and I can't imagine that Nvidia is sitting on a huge pile of Blackwell chips. We picked up through the grapevine various mutterings that Nvidia commenced manufacturing of RTX 50-series GPUs quite late in 2024, though we couldn't pin down such rumours nor determine any possible explanation for such a decision.
However, one factor that may have played a role is the consumer-grade Blackwell chips are made on the same process node as datacentre Blackwell, i.e. TSMC N4 (Ada Lovelace GPUs are made on a custom version of that node).
Given just how much money Nvidia makes from selling its massive AI processors, I should imagine TSMC order books were chock full of GB100s, leaving little spare capacity to start RTX 50-series chips in earnest.
And it's worth noting that TMSC's N4 process node is in hot demand. As well as Nvidia's entire Blackwell range, the Taiwanese fabrication giant also produces AMD's Zen 5 chiplets, as well as its Strix Halo and the latest Hawk Point APUs on the same node. They're all much smaller than the GB202, so fewer wafers need to be allocated for AMD's orders, but it can only churn out so many each month.
Not that this is comfort if you were hoping to snag an RTX 5090 at its MSRP when they hit retailers' shelves (if they even reach them) at the end of this month. The 575 W monster is the most powerful gaming graphics card money can buy but it would seem that for the next month or so, no amount of money may be able to get you one.
