Advanced Micro Devices saw its shares fall more than 8% Tuesday, despite record quarterly sales, after SpaceX chief Elon Musk said that his company would no longer buy AMD's chips.

In dueling earnings calls Tuesday afternoon, AMD highlighted surging sales of data-center chips, while Musk praised its biggest rival, market leader Nvidia.

"Going forward, we have decided to build exclusively on Nvidia, because we think the Blackwell architecture is the best architecture," Musk told investors, explaining his strategy for investing in AI computing infrastructure.

As recently as May, Musk had said that his two main companies, Tesla and SpaceX, would likely continue to buy both Nvidia and AMD chips.

AMD reported $11.5 billion in revenue for the quarter ended June 27, narrowly beating analysts' expectations and the company's guidance. Analysts polled by FactSet had expected quarterly revenue of $11.3 billion.

The chip maker reported its fifth consecutive quarter of record sales of its bread-and-butter processors as the rise of artificial-intelligence agents continues to drive an unprecedented wave of demand.

Data-center revenue doubled over the last year to a record $6.7 billion and now represents 58% of AMD's total revenue, up from 42% a year ago, AMD said.

The company, which is helmed by Chief Executive Lisa Su, reported net income of $2.3 billion for the quarter, exceeding the $1.7 billion Wall Street had expected, while operating income came in at $2 billion, slightly below analysts' expectations.

AMD has pivoted sharply over the last year to serve customers running inference workloads, or the type of computing required to query AI models. Even more pronounced has been the rise in demand for so-called agentic AI, which involves creating tools known as agents to help with tasks such as coding. Agents rely more heavily on CPUs, or central processing units, rather than the more-advanced GPUs, or graphics processing units, required for training models.

In July, shares in AMD and other makers of computing hardware declined sharply over fears that AI companies had overextended themselves and wouldn't be able to meet financing commitments. By late July, the company's stock price had slipped below $430, but over the last week, it recovered most of those losses.

As the AI market evolves, companies are demanding more firepower to run AI models and other tools more quickly. In response, semiconductor companies such as AMD and market leader Nvidia are increasingly shifting their focus from designing the most powerful and efficient chips to offering end-to-end solutions for AI computing, including GPUs, CPUs, server racks, connectivity hardware, programming software and other products.

Last month, AMD announced a partnership with Cerebras, a maker of picture book-size chips known as wafer-scale engines, which are customized for fast inference computing, to incorporate its processors into AMD's latest AI system, known as Helios.

The first shipments of Helios, a server rack designed according to specifications drawn up by Meta Platforms and meant to compete with Nvidia's NVL72 systems, are expected in the coming months, AMD has said.

Su has been on a dealmaking spree over the last year, most notably striking multiyear partnerships to supply chips to OpenAI, Meta and most recently, Anthropic, which agreed to deploy up to 2 gigawatts' worth of computing using AMD's latest GPUs, known as the MI450, as well as the Helios server racks.