$AMD(AMD)$ 【AMD Strikes at NVLink: Can the Open Helios Cabinet Disrupt the AI Landscape?】
Nvidia has dominated the AI chip market for a decade, and the real turning point may explode at the "full cabinet system" level. Last night, AMD unveiled the MI355X/MI400 and launched the Helios cabinet: 72 GPUs + UALink open interconnection, the entire cabinet is treated as "a supercomputer." Higher HBM capacity and power consumption ratio lead to a 29% reduction in inference cost per token (compared to Nvidia's B200), with Crusoe purchasing 13,000 units at once as evidence of demand. OpenAI, Meta, and xAI have already locked in orders, and Oracle plans to deploy 131,000 units in clusters.
AI business revenue for 2024 has exceeded $5 billion, and management has raised the TAM to over $500 billion by 2028; if AMD's AI-related revenue grows to $8 billion in 2025, non-GAAP EPS will rise to $6, which, even at a 30 × PE ratio, points to $180, while the current price is still below $120. The market underestimates not only the "second supplier" but also the accelerator effect of AMD's transition to a "CPU + GPU + Network + Cabinet" full-stack platform.
💡Question: Can an open ecosystem really leverage NVLink's moat? Will Helios become a cost shockwave for cloud vendors?
Not an investment suggestion.