AMD sees 2nd-quarter revenue above estimates on AI spending

Advanced Micro Devices forecast second-quarter revenue above Wall Street estimates on strong demand for AI infrastructure, sending its shares up 12%.

The US chipmaker projected revenue of about US$11.2 billion, plus or minus US$300 million, ahead of analysts’ expectations of US$10.52 billion.

First-quarter revenue reached US$10.25 billion, beating estimates of US$9.89 billion, while adjusted earnings came in at US$1.37 per share. Data center revenue rose 57% to US$5.8 billion.

The company also forecast adjusted gross margin of about 56% and said server CPU revenue would rise more than 70% year on year in the second quarter.

AMD said higher memory and component costs could weaken consumer demand, with second-half PC shipments set to fall and gaming revenue expected to drop by more than 20% from the first half.

🔗 Source: Reuters

🧠 Food for thought

Implications, context, and why it matters.

AI-related computing work is helping drive AMD’s server CPU growth

GPUs lead most AI training work, yet AMD expects server central processing unit (CPU) revenue to rise more than 70% as AI server jobs lift demand.
Reinforcement learning is a way to train AI systems through feedback. Retrieval-augmented generation (RAG) adds outside information to AI models. Both can strain general-purpose CPUs with data processing, simulation, and tool use 1.
One OpenAI buildout inside Microsoft’s Fairwater data centers pairs a 48-megawatt CPU and storage building with a 295-megawatt GPU cluster. That works out to about a 1-to-6 CPU-to-GPU power ratio 1.
Some leading AI labs are short on CPUs, and AMD sees a path to more than 50% of server processor revenue share from 40%, said Lisa Su, CEO of AMD 2, 1.

AMD’s outlook points to wider AI supply constraints

AMD’s revenue forecast looks strong in operations, though the available sources do not confirm a steady supply of High Bandwidth Memory (HBM) or CoWoS, a packaging method that links chips and memory more tightly for high-performance AI systems.
HBM and CoWoS remain industry chokepoints, with forecasts putting HBM above US$100 billion by 2030. Shortages are expected to last through at least 2026 and may cap GPU shipments 3, 4.
That leaves chipmakers competing on supply access as much as product design.
Firms that miss out on these scarce parts may struggle to capture AI demand as the data center GPU market heads toward US$228.04 billion by 2030 5.