AMD CEO Lisa Su says the $500 million upgrade in the company’s Instinct GPU 2024 sales forecast was based on the completion of ‘some important customer milestones’ such as meeting reliability requirements in data centers and optimizing the chips on certain AI workloads.

AMD said it now expects to make more than $5 billion from sales of its Instinct data center GPUs this year due to high demand from hyperscalers like Meta and Microsoft as well as other customers using the chips to run AI workloads.

This marks an upgrade from the $4.5 billion AMD forecast for Instinct GPU sales in July and the $4 billion guidance it issued in April as the Santa Clara, Calif.-based chip designer seeks to challenge Nvidia, which saw revenue more than double year-over-year to $30 billion in the last quarter alone, largely due to its continuing dominance of the AI computing market.

[Related: AMD Says Instinct MI325X Bests Nvidia H200, Vows Huge Uplift With MI350]

AMD disclosed the sales momentum for its Instinct GPUs during the Tuesday call for its third-quarter earnings, where the company’s CEO, Lisa Su, said “significantly higher data center and client processor sales more than offset declined in gaming and embedded product sales.”

As a result, AMD’s third-quarter revenue came in at $6.8 billion, which was 18 percent higher than the same period last year and 17 percent higher sequentially. The company’s gross margin grew 3 points year-over-year to 50 percent and its gross profit increased 24 percent year-over-year to $3.4 million while its earnings per share was up 161 percent to 47 cents.

It expects to make roughly $7.5 billion in revenue for the fourth quarter, which would amount to a roughly 22 percent year-over-year increase and a 10 percent sequential increase.

“This is an incredibly exciting time for AMD as the breadth of our technology and product portfolios, combined with our deep customer relationships and diversity of markets we address, provide us with a unique opportunity as we execute our next arc and make AMD the end-to-end AI leader,” Su said on the earnings call.

AMD’s third-quarter revenue beat Wall Street’s expectations by roughly $100 million while its adjusted earnings per share were slightly in line with analyst estimates. Its fourth-quarter revenue forecast fell just short of Wall Street’s expectations, however.

The company’s stock price was down more than 7.5 percent in after-hours trading Tuesday.

Business Unit Breakdown

AMD said third-quarter data center revenue more than doubled from the same period a year ago, growing 122 percent year-over-year to a record $3.5 billion, thanks to a “strong ramp” of Instinct GPU shipments combined with growth of EPYC CPU sales. This was 25 percent higher revenue than the previous quarter, according to the company.

Client revenue, on the other hand, grew 29 percent year-over-year and 26 percent sequentially to $1.9 billion, primarily thanks to “strong demand” for its Zen 5-based Ryzen processors.

The chip designer suffered, however, when it came to its embedded revenue, which declined 25 percent year-over-year to $927 million due to customers working through existing inventory but increased 8 percent sequentially due to improving demand. It also saw gaming revenue sink 69 percent year-over-year and 29 percent sequentially, mainly due to lower semi-custom sales.

Why AMD Boosted Instinct GPU Forecast

Su said the $500 million upgrade in AMD’s Instinct GPU 2024 sales forecast was based on the completion of “some important customer milestones” such as meeting reliability requirements in data centers and optimizing the chips on certain AI workloads. This allowed the company to ramp up Instinct shipments faster than previously expected.

Two major customers that helped fuel sales in the third quarter were Microsoft and Meta, which both expanded their use of AMD’s MI300X GPUs for internal workloads, according to Su.

While Microsoft is using the MI300X to run multiple Copilot services powered by OpenAI’s GPT-4 AI models, Meta has “broadly deployed MI300X to power their inferencing infrastructure at scale, including using MI300X exclusively to serve all live traffic for their most demanding Llama [405-billion-parameter] frontier model,” she said.

The company also saw the availability of MI300X public cloud instances expand among Microsoft, Oracle Cloud and smaller cloud providers while “multiple startups and industry leaders” such as Databricks and Essential AI drove “strong” adoption of such instances.

Su said AMD has also received positive feedback from customers regarding its pending $4.9 billion acquisition of cloud computing specialist ZT Systems.

The acquisition, which is set to close in the first quarter of next year, is expected to enable hyperscaler customers to “rapidly deploy AMD AI infrastructure at scale while giving server vendors “optimized board and module designs for a wide range of differentiated enterprise solutions,” according to the CEO.

At the same time, Su said, AMD’s plan to divest ZT Systems’ data center infrastructure manufacturing business has “received significant interest from a number of parties to-date.”

Looking out to the next two quarters, Su said AMD customer engagements around Instinct GPUs are “actually broadening quite well.”

“Our largest cloud customers are broadening the set of workloads that they’re running on AMD instinct, and we’re also very engaged with a number of large cloud and enterprise customers that are actively working with us and optimizing their workload,” she said.