AMD is making significant strides in developing artificial intelligence (AI) chips, but is still struggling to compete with NVIDIA’s overwhelming dominance. While powerful GPUs like the newly introduced Instinct MI350 offer up to a 35x increase in AI inference performance compared to previous models, according to IDC and Jon Peddie Research, NVIDIA still holds 85% to 92% of the AI chip market.
AMD has outlined an ambitious roadmap.
The Instinct MI400, set to be released in 2026, will offer 432 GB HBM4 memory and 40 PFLOPS FP4 processing power. These GPUs will be paired with AMD’s EPYC Venice CPUs, which boast up to 256 cores and promise 70% higher performance compared to the current generation. All these products will be manufactured using TSMC’s 2 nm production process.AMD also introduced its Helios infrastructure. This system is positioned as a competitor to NVIDIA’s GB200 NVL72 and the upcoming “Oberon” AI servers.
AMD claims it will surpass NVIDIA, particularly in memory capacity and bandwidth—two factors quite critical for AI workloads. But NVIDIA isn’t standing still either: with its Rubin Ultra architecture, planned for 2027, it aims to deliver up to 5 exaFLOPS of performance in FP8 format.Looking ahead, AMD will release EPYC Verano CPUs and Instinct MI500X GPUs; These are expected to be manufactured using TSMC’s 1.6 nm technology.
Although AMD has made significant progress in terms of hardware, the biggest obstacle is on the software side. NVIDIA’s CUDA platform remains the undisputed standard in the industry. While AMD’s alternative, ROCm 7, has shown improvement, analysts still note that it lags behind in ease of use and reliability.
AMD was encouraged by OpenAI CEO Sam Altman’s confirmation that AMD chips will be used in future data centers.
It’s also competitive in terms of price/performance: AMD’s MI300X chip sells for around $15,000, while NVIDIA’s H100 is over $40,000.In short, AMD is closing the gap in hardware, but unless software maturity and widespread adoption are achieved, NVIDIA maintains its undisputed leadership in AI — at least for now.
