By Anhata Rooprai and Max A. Cherney
Aug 4 (Reuters) - AMD forecast quarterly revenue above Wall Street estimates on Tuesday, banking on strong demand for its chips from massive data-center capacity expansions to power AI technologies.
Yet, its shares fell nearly 9% in extended trading, suggesting that investors were looking for an even stronger outlook after the stock more than doubled this year spurred on by AI optimism.
"AMD is now in a similar position to Nvidia and the hyperscalers, where investors
are looking for evidence that AI infrastructure investments will continue translating into accelerating returns," said Emarketer analyst Jacob Bourne.
The Santa Clara, California-based company is regarded as chip giant Nvidia's closest rival in the market for graphics processing units, as major technology companies and governments worldwide ramp up spending on AI infrastructure.
AMD expects third-quarter revenue of about $13 billion, plus or minus $300 million, while analysts estimate $12.52 billion, according to data compiled by LSEG. Its expects adjusted gross margin to be about 56%, largely in line with estimates.
The company expects its data-center sales to more than double in 2027, and forecast total revenue growth above its more than 35% target and annual earnings above its $20-per-share goal set at its 2025 analyst day, CEO Lisa Su said on a post-earnings conference call.
The forecast suggests that AMD's investments to challenge Nvidia's dominance in the market for AI chips are beginning to pay off, with sales of its data-center processors accelerating in the second quarter.
AMD's quarterly data-center revenue more than doubled to $6.72 billion, beating expectations of $6.48 billion. It grew 16.3% sequentially from $5.78 billion in the first quarter.
Second-quarter revenue jumped 50% to $11.54 billion, beating the estimate of $11.28 billion. Adjusted profit of $1.66 per share also surpassed the estimate of $1.62.
A BROADER AI HARDWARE PLAY
AMD has stepped up its AI product launches and moved beyond selling individual chips to offering AI systems that combine processors, networking gear and related hardware, giving customers an integrated option.
It also allows the company to better rival rack-scale products by Nvidia, which in March unveiled its next-generation "Vera" CPU to pair with its "Rubin" line of graphics processors.
Su said AMD's 2027 rack-scale AI platform would bring together MI500 GPUs, Verano CPUs and Pensando networking in a system designed to support both copper and optical connectivity.
AMD is also benefiting from growing demand for central processing units, which are used alongside pricey graphics processors in servers for inference — data crunching that occurs when a user queries a chatbot.
This has helped it capture market share from Intel, which raised its spending plans for the next two years to capitalize on the demand boom.
At AMD's AI event in July, Su had said the company's second-generation Helios AI servers, featuring the MI455X AI accelerator and "Venice" processor made by TSMC, were in full production and would begin shipping in the coming months.
The company has also secured major customers and infrastructure agreements in recent months.
In July, AMD agreed to sell Anthropic tens of billions of dollars worth of AI servers powered by up to 2 gigawatts of MI450 chips from early 2027, and invest up to $5 billion in the IPO-bound Claude maker, contingent on deployment milestones.
It locked in up to 2.5 GW of data-center capacity through a deal with Core Scientific a week ago, while gaining warrants to purchase the company's stock.
(Reporting by Anhata Rooprai in Bengaluru and Max A. Cherney in San Francisco; Editing by Shilpi Majumdar)
