AMD readies AI workstation to challenge Nvidia’s GB10, preorder

AMD is set to open pre‑orders for its latest AI‑focused workstation, aimed at rivaling Nvidia’s GB10 platform, within the next few days [1]. The move comes as AMD pushes its Instinct GPU line, which it claims can deliver higher memory capacity and lower inference costs than Nvidia’s current offerings [1].

Key takeaways

• AMD’s Instinct MI325X GPU offers up to 432 GB of HBM4 memory, nearly double Nvidia’s Rubin‑based memory capacity [1].
• Data‑center revenue rose 39 % YoY to a record $5.4 billion in the last quarter, underscoring growing demand for AMD’s AI chips [1].
• Analysts note that hyperscalers are still evaluating custom silicon alternatives, which could limit the immediate impact of AMD’s new workstation [1].
• Export‑control limits on sales to China reduce the Instinct MI308’s contribution to AMD’s upcoming quarter, highlighting supply‑chain constraints [2].
• The market’s view of AMD’s AI prospects is mixed; while some see a structural shift, others warn Nvidia’s CUDA ecosystem remains dominant [1][3].

AMD’s hardware push and market backdrop

AMD’s latest AI workstation leverages the Instinct MI325X GPU, which the company says provides 432 GB of HBM4 memory—about twice the capacity of Nvidia’s newest Rubin architecture [1]. The larger memory pool is intended to let massive models such as Llama‑4 run on a single chip, reducing latency and simplifying system design. AMD also argues that its GPUs lower the cost per token for inference, a claim that has attracted bulk purchases from firms like Meta and Microsoft [1].

The hardware announcement arrives amid a broader shift in the AI hardware market. AMD’s data‑center segment posted a record $5.38 billion in revenue, up 39 % year over year, and the company projects a 60 % YoY increase in data‑center revenue over the next several years [1]. Analysts point to these figures as evidence that hyperscalers are seeking alternatives to Nvidia’s single‑vendor dominance [2]. However, they also caution that major cloud providers are increasingly moving inference workloads onto in‑house chips such as Amazon’s Trainium 3 and Google’s Maia 200, which can be up to 60 % cheaper than third‑party GPUs [1].

Timing, supply constraints and outlook

Pre‑order availability for AMD’s workstation is slated for early May, just ahead of the company’s Q1 2026 earnings call on May 5 [2]. The earnings release will be a key barometer for whether the new AI hardware can translate into revenue growth. Analysts note that the guidance includes only $100 million of Instinct MI308 sales to China, reflecting ongoing export‑control restrictions that have already trimmed prior quarters’ top lines [2].

Supply‑chain considerations also loom large. AMD relies on TSMC’s 3 nm process, competing with Nvidia and for limited capacity, which could affect the ability to meet demand for the new workstation [1]. If AMD can secure sufficient fab allocation, the product could help the company capture a larger share of the AI accelerator market; if not, Nvidia’s entrenched CUDA ecosystem and its own custom silicon may keep AMD in a secondary role [3].