AMD launches Instinct MI325X, a new AI chip to take on Nvidia’s GPUs

AMD has just launched its latest AI chip, the MI325X, with its sights set on competing directly with Nvidia’s dominance in the data center GPU market. Unveiled at an event in San Francisco, the chip is set to enter mass production by the end of this year, marking AMD’s push to strengthen its presence in the AI chip space to challenge Nvidia’s GPUs.
The MI325X builds on the architecture of AMD’s earlier MI300X, introducing an upgraded memory system designed to speed up AI computations. Looking ahead, AMD is also gearing up for its next-generation MI350 chips, set to arrive in the latter half of 2025.
These new chips will come with more memory and a redesigned architecture aimed at delivering even better performance than their predecessors. The AMD Instinct MI325X is the first AI GPU to feature 256 GB of HBM3e memory, with the MI355X expected to launch next year offering an 8x performance increase.

Instinct MI325X
In addition to the MI325X, AMD announced an update to its server CPU lineup, with a new chip designed to ensure GPUs are constantly fed with data, speeding up AI tasks. The flagship CPU features almost 200 processing cores, carries a hefty price tag of $14,813, and promises up to a 37% speed boost for handling advanced AI workloads.
“AI demand has actually continued to take off and actually exceed expectations. It’s clear that the rate of investment is continuing to grow everywhere,” AMD CEO Lisa Su said at the event.
The launch comes at a time when demand for AI processors has surged, driven by major tech players like Microsoft and Meta. If AMD’s chips prove to be a viable alternative to Nvidia’s offerings, it could pressure Nvidia’s pricing, which has remained high due to overwhelming demand. Currently, Nvidia holds about 90% of the market, but AMD is hoping to carve out a significant share, eyeing the AI chip market projected to hit $500 billion by 2028, CNBC reported.
AMD didn’t announce any new big cloud customers for its AI chips, though the company has previously stated that Meta, Microsoft, and OpenAI use its GPUs. Pricing for the MI325X hasn’t been revealed yet, as it’s typically sold as part of larger server systems.
By speeding up its product release cycle, AMD is positioning itself to compete more aggressively with Nvidia. The MI325X will go head-to-head with Nvidia’s upcoming Blackwell chips, expected to ship next year.
While AMD’s stock has seen only a modest 20% rise this year, Nvidia’s has skyrocketed by over 175%. A successful launch of the MI325X could draw more investor attention, particularly from those looking to diversify their AI bets beyond Nvidia.
One of the challenges AMD faces is Nvidia’s CUDA software, which has become a go-to tool for AI developers, effectively locking them into Nvidia’s ecosystem. In response, AMD has been ramping up efforts on its own software, ROCm, to make it easier for developers to transition to AMD’s hardware.
AMD is emphasizing that its accelerators are particularly suited for tasks where AI models generate content or make predictions, citing the advanced memory capabilities of its chips. The company claims that the MI325 platform offers up to 40% more performance than Nvidia’s H200 on Meta’s Llama AI model.
As the AI race heats up, AMD is determined to challenge Nvidia’s dominance, but it remains to be seen how quickly developers and cloud providers will embrace its alternatives.
You can watch the live event below.