Tuesday, February 17th 2026
AMD Instinct MI455X Reportedly Facing Delays, Only Low-Volume Production This Year
According to SemiAnalysis, which recently released its InferenceX benchmark comparing the inference performance of the latest AI accelerators, AMD's next-generation Instinct MI455X accelerators, designed to compete with NVIDIA's "Vera Rubin," are facing serious manufacturing difficulties, causing delays in AMD's roadmap execution. Only low-volume production will occur this year, while customers will be generating AI tokens in Q2 2027. This means that NVIDIA's "Vera Rubin" VR200 rack-scale system will be the most powerful system shipping this summer, without competition. As AMD's first rack-scale system, the "Helios" AI rack with the MI455X UALoE72 configuration packs 72 GPUs and 6th Gen AMD EPYC "Venice" CPUs in a heterogeneous system designed to run as a single homogeneous GPU, delivering massive AI performance with multi-ExaFLOP compute output.
Given that it is AMD's first complete rack-scale solution, some manufacturing issues were to be expected, so the company is ironing them out before starting high-volume production. For the Instinct MI455X, AMD claims it will deliver up to 40 FP4 and 20 FP8 PFLOPs, roughly twice the compute performance of the current MI350. The GPUs also transition to HBM4 memory with a 432 GB capacity and get a total bandwidth of 19.6 TB/s. Each GPU provides 300 GB/s of scale-out bandwidth and adds new AI data formats like FP4 and others.
Source:
SemiAnalysis
SemiAnalysisEngineering samples and low volume production of AMD's first rack scale MI455X UALoE72 system will be in H2 2026 while due to manufacturing delays, the mass production ramp and first production tokens will only be generated on an MI455X UALoE72 by Q2 2027.
17 Comments on AMD Instinct MI455X Reportedly Facing Delays, Only Low-Volume Production This Year
'We only do one GPU'
'We barely make em'
'Oh gosh, lowest market share ever, dafuq'
This is great. We need a reality check
Reality checks is the last thing on Nvidia's mind.
EDIT: Samsung and Intel will likely be benefiting as some TSMC customers look elsewhere for capacity.
Same company cant consistently release GPUs in every tier, dropps support for all their previous architectures, ends driver support early, ignores consumer feedback, expects companies to just use whatever tech they lay on the floor, and cant figure out why consumers buy more RTX 5060s then the entireity of AMDs lineup.
RDNA4 is comprised on two separate dies.
And 6 separate SKU's.
And please dont tell me you quote JPR numbers again?
AMD's market share in dGPU's its at around 19%. Not 6% or whatever the latest JPR numbers were.
It'll be particularly bad for stacked die products, like 3D Vcache, and interposer products, since the complex packaging lines are even more constrained than the fabrication lines.
This would only be great if it applied across the board. A smaller competitor not being able to get enough wafers to produce their product because it's being eaten by the big guys will only further cement market leaders.
I understand that people are upset at AI right now but you don't want the GPU or other markets to further consolidate (or for token competitors to be so far behind they are kept around merely for appearances).
We just gained a third player, consolidation among AIBs is completely irrelevant, as Nvidia and AMD have shown us each on their own merit.
There is also China.
There is a LOT of movement and I think nobody buying GPUs wants to buy them all from Nvidia. It leads to tunnel vision, and kills innovation, in a segment and industries where innovation is the only driver for progress. That won't mix, it won't last and history has also underlined this time and time again: design wins keep coming, from more companies than just Nvidia.
Increased demand does not lead to consolidation, it leads to a more vibrant marketplace. Chip development is like a fruit tree that keeps on giving because we want more and more of it, and as a result, it will also diversify and spawn new opportunities.
Zoom out beyond the current transitional period and you can see we've only gained more of everything.
MI455X is due H2 2026 as expected, no delays. Fake news.