Arm Q2 results a record as chip shipments top 300 billion
For its fiscal second quarter and three months ended September 30, Arm has posted revenue of $844 million, which is up five per cent year-on-year and due to record levels of royalty hauls and continued strength in licence revenue.
CEO Rene Haas said: “Demand for our high-performance Armv9 and CSS compute platforms continues to exceed expectations and to accelerate our licensing and royalty revenue growth. AI everywhere is generating new opportunities for the Arm compute platform from the cloud to the edge.”
More than 300 billion Arm-based chips have now shipped cumulatively thanks to Arm’s world-leading compute ecosystem of technology partners, which also features more than 20 million software developers.
Notably, products based on Arm Compute Subsystems (CSS) are now becoming available, including Microsoft Azure Cobalt and Google Axion in the data centre, and MediaTek’s new Dimensity 9400 for smartphone.
Royalty revenue of $514 million, which is up 23 per cent year-over-year. This is driven primarily by smartphone market recovery and the continued adoption of Armv9, which generated around 25 per cent of royalty revenue in the quarter, up from around 10 per cent a year ago.
Licence and other revenue of $330m was a 15 per cent decline as expected due to the normal fluctuation in timing and size of multiple high-value licence agreements.
However, the Nasdaq-quoted company reports stronger partner demand for Arm CSS and has more than doubled Arm CSS licensees signed in this fiscal year.
Non-GAAP operating income of $326m resulted in a 38.6 per cent non-GAAP operating margin and non-GAAP earnings per share of $0.30.
Haas said demand for AI everywhere was leading partners to make long-term commitments to more powerful and power-efficient Arm technology.
During the quarter, Arm and Meta announced the optimisation of Meta’s new Llama 3.2 large language models (LLMs) for Arm CPUs with KleidiAI libraries, driving 5x improvements in prompt processing and 3x improvements in token generation compared to Llama 3.2 without KleidiAI.
