AMD CEO Lisa Su’s AI Chips Business Is Thriving. Can Supply Keep Up?
Observer ·
Lisa Su controls AMD’s push into A.I. infrastructure. The company’s data-center business now supplies nearly two-thirds of its sales, with major customers receiving conditional rights to buy AMD shares.
Lisa Su is steering AMD toward data-center and A.I. computing , and the chipmaker’s business is thriving. On Nov. 3, the CEO announced its third-quarter financial results . Revenue reached $13.4 billion, up 45 percent year over year.
Its standout performer was its data-center business, which generated record revenue of $8.5 billion, up 97 percent year over year. AMD attributed the growth to strong demand for its EPYC processors and the launch of its Instinct MI450 Series graphics processing units, or GPUs, used for A.I. workloads. Data centers now account for nearly two-thirds of AMD’s revenue.
“We delivered a record quarter with revenue exceeding the high end of our guidance as Data Center segment sales nearly doubled year-over-year,” Su said in the Nov. 3 earnings report, highlighting demand for AMD’s Instinct GPUs and EPYC central processing units, or CPUs.
The data-center segment’s operating margin rose to 34 percent from 25 percent a year earlier. Across AMD, the gross margin was 54 percent under generally accepted accounting principles, the same measure used in the company’s headline results.
More broadly, the client and gaming segment reached $3.7 billion in revenue, down 8 percent year over year. Gaming revenue fell 40 percent to $773 million due to lower semi-custom and graphics revenue. At the same time, client revenue rose 7 percent to $2.9 billion, driven by sales of AMD Ryzen mobile processors.
In a Nov. 5 Yahoo Finance interview with Executive Editor Brian Sozzi , Su went as far as describing AMD as a “data center first company,” adding that “people need more computing to power all of these applications.” Su also acknowledged on the Nov. 3 earnings call that the company had more demand than it could supply.
Su’s answer to compute supply-chain challenges is forward planning, with the company investing in research and development, supply-chain capacity and infrastructure alongside its commercial partners. “We are selectively investing with our lead partners where we see it is helpful to have, for example, the data centre infrastructure identified way in advance, and that is multiple years in advance. I think that is the right thing to do to ensure that we are jointly ensuring that there is the right infrastructure as our new chips become available,” Su said on the call.
Alongside its data-center ambitions, AMD is looking to become a key supplier for the largest A.I. developers. Anthropic , OpenAI , Meta , Microsoft and Oracle Cloud Infrastructure were running models and workloads on the first AMD Helios rack-scale systems, built around Instinct MI450 Series GPUs and EPYC CPUs, according to AMD’s Nov. 3 report. These initial deployments do not mean the customers’ full multiyear buildouts are complete.
“As part of AMD’s partnership expansions, CEO Lisa Su is continuing to nurture deals with major tech companies,” Yaz Palanichamy , an analyst at Info-Tech Research Group , tells Observer via email.
Su’s multiyear agreement with Meta covers up to six gigawatts of GPUs. Both Meta and OpenAI also received warrants—rights to buy AMD stock—for up to 160 million shares each, subject to purchase, stock-price and other conditions. Those incentives give two of AMD’s largest customers a potential ownership stake as their purchases grow.
The financial ties run both ways. AMD invested $1 billion in Meta’s October equity offering, at its Oct. 27 Meta partnership update and Nov. 4 quarterly filing show. The chipmaker is putting capital into a customer as it tries to sell that customer more computing power.
Anthropic’s Nov. 10 IPO filing also lists a $2.5 billion purchase of Anthropic stock by AMD in an accompanying private placement, contingent on the offering closing and other conditions. The purchase would count toward AMD’s existing commitment to invest up to $5 billion in the customer, reducing the remaining commitment dollar for dollar.
AMD’s results arrived days before its key competitor Nvidia opened an independent board investigation into reports that CEO Jensen Huang had been targeted by a foreign intelligence effort. Nvidia’s board said on Nov. 8 that it had no reason to believe national security or company secrets had been compromised and expressed confidence in Huang and his leadership team. Its independent directors appointed outside counsel Nov. 10 to conduct the inquiry.
For its longer-term outlook, AMD plans to increase supply in 2027 across its CPU and GPU offerings, working with supply-chain partners to secure memory and other components. On Nov. 10, AMD announced that sixth-generation EPYC server platforms were generally available and shipping from server vendors. Obstacles remain on the road ahead.
“For organizations that are looking to deploy AMD’s hardware, there are certainly hurdles with both infrastructure costs and setup time that would need to be considered,” Palanichamy says. Even AMD’s older Instinct MI325X accelerator , introduced in 2024, has a specified peak board power of 1,000 watts. Palanichamy points to power upgrades and advanced cooling as costs buyers need to consider.
Software adds another expense. Buyers switching code from Nvidia’s software to AMD’s systems may need dedicated engineering hours to make the transition, Palanichamy tells Observer. He adds that recent software updates have helped lessen that burden.
On Oct. 8, Su received the National Medal of Science from President Donald Trump at the Donald J. Trump Institute of Peace in Washington.
Su joins the ranks of ex-Oracle CEO Safra Catz , ex-OpenAI CTO and cofounder and CEO of Thinking Machines Lab Mira Murati , and Fei-Fei Li , cofounder of World Labs, as one of the leading voices in tech.