This year, AMD anticipates introducing an updated MI300 AI accelerator featuring HBM3E memory, which will be succeeded by the Instinct MI400 in 2025.
Anticipated AMD Next-Generation MI400 AI GPUs by 2025; MI300 Refresh Also in Store
AMD is rumored to be preparing to release new AI-focused products in the near future, as a "refresh" variant of their original Instinct MI300 AI GPU is rumored to be available, featuring an upgrade to the most recent HBM3e standard for onboard memory.
This indicates that AMD has made the transition to the more recent standard with the release of the "revamped" MI300, as the company's prior products were built upon the comparatively older HBM3. Based on the information provided by @Kepler_L2, the forthcoming AI GPU may prove to be an exceptionally priced-to-performing product.
Although specific details regarding the forthcoming SKU are not available, incorporating HBM3e appears to be a wise decision, given that competitors in the industry are progressively adopting this standard. Hopper GH200, an AI GPU from NVIDIA, has already been introduced; it is the only GPU on the market to include HBM3e memory.
AMD Targets AI Market with MI300 GPU, Aiming to Outpace NVIDIA's B100
In summary, the HBM3e memory standard provides a 50% increase in performance compared to the current HBM3 standard. It can deliver bandwidths of up to 10 TB/s per system and 5 TB/s per chip, in addition to memory capacities of up to 141 GB.
Kepler does mention that AMD's redesigned Instinct MI300 GPU will be targeted at NVIDIA's forthcoming Blackwell B100. Although the new MI300 represents a generational upgrade, it is said to be the superior option because it will offer clients a better value. Although the B100 will utilize an HBM3e memory solution, it will be a more expensive model; thus, AMD may gain traction in this mid-to-high segment of the AI industry.
In addition to the forthcoming MI300, AMD is reportedly preparing to unveil its next-generation Instinct MI400 AI accelerator in 2025. AMD CEO Lisa Su herself previously confirmed that this accelerator was in development.
Nevertheless, the potential performance enhancements that the accelerator might impart remain wholly unknown to us. However, it is rumored to debut with a "range of specifications," which means that, similar to its predecessor, there may be multiple SKUs in the lineup.
Although AMD, like its archrival NVIDIA, has yet to succeed in the AI markets, its future looks promising.
Photo: Timothy Dykes/Unsplash


Trump Pushes Tech Giants to Build Power Plants to Offset AI Data Center Energy Costs
OpenAI Pentagon AI Contract Adds Safeguards Amid Anthropic Dispute
Nvidia Earnings Preview: AI Growth Outlook Remains Strong Beyond 2026
OpenAI Targets $600B Compute Spend as IPO Valuation Could Reach $1 Trillion
Samsung Stock Hits Record High on Nvidia HBM4 Supply Deal, Boosting AI Chip Rally
Nvidia Earnings Preview: AI Chip Demand, Data Center Growth and Blackwell Shipments in Focus
Pentagon Weighs Supply Chain Risk Designation for Anthropic Over Claude AI Use
Apple to Begin Mac Mini Production in Texas Amid $600 Billion U.S. Investment Plan
xAI’s Grok Secures Pentagon Deal for Classified Military AI Systems Amid Anthropic Dispute
Hyundai Motor Plans Multibillion-Dollar Investment in Robotics, AI and Hydrogen in South Korea
Anthropic Resists Pentagon Pressure Over Military AI Restrictions
Meta Signs Multi-Billion Dollar AI Chip Deal With Google to Power Next-Gen AI Models
Amazon’s $50B OpenAI Investment Tied to AGI Milestone and IPO Plans
DeepSeek AI Model Trained on Nvidia Blackwell Chip Sparks U.S. Export Control Concerns
Nintendo Share Sale: MUFG and Bank of Kyoto to Sell Stakes in Strategic Unwinding
Coupang Reports Q4 Loss After Data Breach, Revenue Misses Estimates 



