AMD reports Instinct AI chip sales exceeded expectations, driven by large Microsoft purchasing commitments.
AMD's Instinct MI300 GPU line achieved a significant milestone in the second quarter, exceeding $1 billion in quarterly revenue for the first time. According to CEO Lisa Su, a "steep ramp" of the MI300 chips accounted for more than a third of AMD's reported $2.8 billion in data center revenue for the quarter. The strong performance of Instinct GPUs, alongside robust sales of fourth-generation EPYC CPUs and Ryzen client CPUs, propelled AMD's total second-quarter revenue to $5.8 billion, a 9 percent increase year-over-year and a 7 percent sequential gain. Data center revenue specifically surged more than 115 percent year-over-year, though other segments showed mixed results: client computing was up 49 percent to $1.5 billion, while gaming declined 59 percent to $648 million and the embedded segment fell 31 percent from a year ago.
The stronger-than-expected demand for Instinct chips prompted the Santa Clara, California-based company to upgrade its full-year data center GPU revenue forecast to $4.5 billion, $500 million higher than the guidance issued in April. For the third quarter, AMD expects total revenue of roughly $6.7 billion, plus or minus $300 million, representing a 16 percent year-over-year increase and a 5 percent sequential gain. The company's stock price rose more than 7.5 percent in after-hours trading following the announcement.
Microsoft has been a key driver of the MI300 ramp, using the GPU-only MI300X to power OpenAI's GPT-4 Turbo model and multiple Copilot services, including Microsoft 365 chat, Word and Teams. The Windows giant also became the first large hyperscaler to launch public MI300X-based instances through Azure in the second quarter, with model repository Hugging Face among the first customers to deploy MI300X-based Azure instances for one-click AI model deployment. Additionally, Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro all have Instinct server platforms in production, while multiple hyperscalers and tier-two cloud providers are on track to launch MI300 instances in the third quarter.
AMD has also advanced software support for its Instinct platform, including industry-first support for Stability AI's Stable Diffusion 3.0 image-generation model and support for Meta's Llama 3.1 models. Su said AMD remains on track to launch its next-generation Instinct GPU, the MI325X, later this year, to be followed by the MI350 series in 2025 and the MI400 series in 2026. "Looking across our AI business, customer response to our multi-year Instinct and ROCm roadmaps is overwhelmingly positive, and we are very pleased with the momentum we are building," Su stated.
On the CPU front, AMD saw a "strong double-digit percentage increase" in EPYC sales in the second quarter, driven by growing adoption among cloud service providers and enterprise customers. More than 900 EPYC-powered instances are now available across all cloud service providers, 34 percent higher than a year ago, with large cloud providers increasingly using EPYC to power internal workloads. Major cloud customers Netflix and Uber are using the fourth-generation EPYC "as one of the key solutions to power their mission-critical, customer-facing workloads." Enterprise customers also drove momentum, with sell-through activity increasing by a "strong double-digit percentage sequentially," and AMD closed multiple large wins with financial services, technology, healthcare, retail, manufacturing and transportation customers, including Adobe, Boeing, Industrial Light and Magic, Optiver and Siemens. Su noted that more than one-third of AMD's enterprise server wins in the first half of the year were with businesses deploying EPYC in their data centers for the first time. The next-generation EPYC processor, code-named Turin, began shipping to lead cloud customers in the second quarter, with broad OEM and cloud availability expected later in the year.