In today’s AI chip market, NVIDIA has long maintained a dominant position, with its products widely used in various AI computing scenarios globally. However, market dynamics are never static. According to Reuters on June 14, AMD is actively seeking to break NVIDIA’s monopoly through a series of strategic initiatives, aiming to make a strong mark in the AI sector.
Strategic Collaborations and Comprehensive Planning
To achieve major breakthroughs in AI, AMD has implemented a number of proactive and effective strategies. One key move is the acquisition of server manufacturer ZT Systems. This acquisition not only enhances AMD’s hardware manufacturing capabilities but also strengthens the integration between its chips and server systems, helping to create more competitive hardware products.
At the same time, AMD has also been busy in the software space. In recent weeks, it has acquired several small software companies. Although these companies are modest in size, they specialize in algorithm development and software optimization. These acquisitions allow AMD to consolidate various technical strengths, accelerate improvements to its ROCm software platform, and expand its technical talent pool.
Vamsi Boppana, AMD’s Senior Vice President of AI, stated that these acquisitions are not coincidental but part of a carefully planned, long-term, cross-generational strategy. He believes that the AI industry is a prolonged technical contest, and only through consistent technological accumulation and strategic planning can one secure a competitive edge in the future.
AMD’s software optimizations have already delivered significant benefits to AI startups. Cohere, a company focused on building customized AI models for large enterprises, is a prime example. Its CEO, Aidan Gomez, noted that AMD’s software improvements have greatly boosted efficiency. Previously, migrating software to AMD chips could take weeks, but now it only takes a few days.
Although Gomez didn’t reveal the exact proportion of Cohere’s software running on AMD chips, he emphasized that this segment represents a substantial share of the global computing infrastructure. This highlights not only AMD’s software progress but also its growing appeal to AI startups.
Additionally, AMD Executive Vice President Forrest Norrod revealed that OpenAI had a significant impact on the design of AMD’s upcoming MI450 AI chips. As a leading AI player, OpenAI demands extremely high standards in chip performance and architecture. Its input deeply influenced the MI450’s memory design and scalability to meet high-performance AI application needs.
According to Norrod, OpenAI’s feedback was instrumental in shaping the chip’s final design. In essence, the MI450 is the result of a deep, collaborative exploration between AMD and OpenAI.
New Product Launch Targets the Competition
At the “Advancing AI 2025” event held on June 12, AMD officially launched its next-generation Instinct MI350 GPU accelerator series, drawing wide industry attention. The series is built on the latest CDNA4 architecture and utilizes TSMC’s second-generation 3nm process, achieving major technological advancements. It integrates 185 billion transistors and 288GB of HBM3E memory, with bandwidth reaching 8TB/s — highly impressive specs.
In terms of compute performance, the MI355X, for example, delivers peak AI performance of 2.5 PFLOPS (FP16), nearly double that of the previous MI325X. Most notably, the MI350 series offers 35x better inference performance compared to the MI300 series, making it much faster and more efficient at handling complex AI inference tasks.
Comparing publicly available data, the MI350 series’ peak AI performance of 2.3 PFLOPS slightly surpasses NVIDIA’s B200 (2.25 PFLOPS). This performance edge indicates that AMD now possesses the capability to compete head-to-head with NVIDIA in AI chip performance.
The MI350 series is expected to launch in Q3 of this year, with companies like Oracle and Dell among the many adopters. Support from numerous manufacturers could help expand AMD’s share in the AI chip market, bolstering its challenge to NVIDIA.
Reportedly, AMD’s MI400 series chips will serve as the core architecture of the new Helios servers launching in 2024. In AI computing, servers are critical infrastructure for chip operation. NVIDIA has also introduced similar server products, making server performance a key battleground in the broader fight for AI market leadership.
In addition to the MI350 series, AMD has for the first time revealed details about its next-generation MI400 series, set for release in 2026. The MI400 will use TSMC’s 2nm process, integrate 432GB of HBM4 memory, and deliver a bandwidth of 19.6TB/s — more than doubling the AI performance of the MI350 series. It directly targets NVIDIA’s upcoming Rubin GPU, showcasing AMD’s determination to push the envelope in technical advancement.
Ambitious Vision for the Future
AMD CEO Lisa Su remains highly confident about the company’s future. She stated that the MI400 series combines AMD’s expertise across silicon, software, and systems, and will be delivered as a fully integrated AI platform. In her view, the MI400 is not just a hardware product, but a testament to AMD’s innovation and technological strength.
Regarding the newly launched MI350, Su expressed high expectations during the earnings call, describing it as a “catalyst” for AMD’s data center GPU business. She projected double-digit revenue growth in 2025 for the data center segment, and set a long-term target of achieving AI chip revenues in the “tens of billions of dollars.” These ambitious goals reflect AMD’s strong determination to capture greater market share in the AI sector.
Through strategic collaborations, acquisitions, software enhancements, and new product launches, AMD is steadily building its strength and mounting a strong challenge to NVIDIA. In the future, as more products roll out and technologies evolve, the AI chip market may undergo a significant transformation.
