AMD Declares War on NVIDIA! Helios Mass Production to Start Shipping by Q3 End, OpenAI to Deploy Massively by Year-End

AMD (AMD-US) CEO Dr. Lisa Su announced on Thursday (23rd) that the second-generation AI server Helios, targeting NVIDIA's (NVDA-US) flagship products, has entered mass production and is expected to begin shipping by the end of the third quarter. OpenAI also announced that it will deploy Helios server racks on a large scale by the end of this year and accelerate expansion by 2027.

Helios is equipped with AMD's latest MI455X AI accelerator and the new-generation central processing unit (CPU) codenamed Venice, both chips manufactured by TSMC (TSM-US) (2330-TW). During a keynote speech in San Francisco, Dr. Su stated that customer demand for Helios is 'extremely strong'.

AMD is attempting to expand its market share in the rapidly growing data center chip market, challenging NVIDIA's long-standing dominance, particularly targeting the AI inference computing market. Inference is the process where models compute and generate answers after users pose questions to chatbots like ChatGPT.

OpenAI to Deploy Helios by Year-End

OpenAI's Vice President of Compute Strategy, Sachin Katti, stated during the event that the company plans to begin large-scale deployment of Helios by the end of this year and continue to accelerate expansion in 2027. OpenAI also plans to adopt AMD's next-generation MI500 AI chip, further deepening the collaboration between the two companies.

AMD announced last October that it had signed a multi-year agreement with OpenAI, expected to bring the company hundreds of billions of dollars in annual revenue, while giving OpenAI the option to acquire up to approximately 10% of AMD's shares.

AMD also announced this Wednesday that it will supply up to 2GW of Instinct MI450 chips to AI startup Anthropic starting from the first half of 2027, with the agreement also including up to $5 billion in investment in the company behind Claude.

Anthropic co-founder Tom Brown stated that Claude can independently complete the setup of AMD AI servers over a single weekend, demonstrating that both humans and AI systems can build truly usable models on the AMD platform.

AMD also announced a collaboration with AI chip and server company Cerebras Systems (CBRS-US) to integrate both companies' AI servers, initially to be deployed in Cerebras data centers. The news drove Cerebras' stock up nearly 10%; AMD, meanwhile, fell 4%.

Emarketer analyst Jacob Bourne believes that AMD's decline primarily reflects weakness in the broader U.S. market and selling pressure on semiconductor stocks, rather than investors dismissing the prospects for Helios.

2030 Computing Market Expected to Reach $2 Trillion

Dr. Su estimates that the global computing market will expand from $365 billion in 2025 to $2 trillion in 2030, with the market for chips used to accelerate AI computing reaching $1.4 trillion, and the CPU market around $220 billion.

She stated that given the enormous size of the market, no single company can meet all the demand alone, and industry players must collaborate through a complete ecosystem. AMD also showcased Helios data center racks at the event, with cloud service providers Vultr and TensorWave, among others adopting AMD hardware, attending.

NVIDIA, meanwhile, revealed extensive technical details about its Vera CPU this week, emphasizing that when paired with the Rubin GPU, it can maximize the amount of work AI agents can complete under a fixed power consumption. NVIDIA is actively showcasing its energy efficiency advantages, highlighting that competition in the next-generation AI data center hardware market is rapidly heating up.

Compared to NVIDIA's mature software-hardware integrated ecosystem, AMD is accelerating the establishment of an alternative platform through Helios complete racks, collaborations with model developers such as OpenAI and Anthropic, and alliances with Cerebras and cloud service providers. As Helios enters mass production and begins shipping in the coming months, AMD's challenge to NVIDIA will officially move from product launch to large-scale deployment phase.