Lisa Su

Overview

Lisa Su is an electrical engineer and semiconductor executive. She has served as president and chief executive officer of AMD since the company's board appointed her to the role in 2014, and as chair of AMD's board since 2022.^[1][2] Her career before AMD ran through research and engineering management at IBM and a senior technology post at Freescale.

Education and early career

According to AMD's announcement of her appointment, Su holds bachelor's, master's and doctorate degrees in electrical engineering from the Massachusetts Institute of Technology.^[1] The three degrees trace the usual path of an engineering researcher: undergraduate training, a master's, then a doctorate built around original research.

AMD's release states that Su spent 13 years at IBM before joining Freescale in 2007 as chief technology officer.^[1] A chief technology officer sets a company's technical direction — which technologies to bet on, how research turns into products — rather than running day-to-day operations. Thirteen years at one semiconductor firm is long enough to cover several generations of chip design, the kind of continuity that a technology officer's job depends on.

Joining AMD and becoming chief executive

Su joined AMD in 2012 and was most recently its chief operating officer before her appointment, according to the same release.^[1] A chief operating officer is the executive responsible for making a company's operations run: schedules, manufacturing relationships, product delivery. The step from that job to chief executive is a common one in large hardware companies, because the operational view it gives covers most of what a chipmaker does.

AMD's board appointed Su president and chief executive officer effective immediately, in a release dated October 8, 2014, succeeding Rory Read.^[1] The appointment took effect at once rather than after a transition period, which means the board handed her the company's direction without an interval in which someone else held it.

Chair of the board

On February 14, 2022, AMD's board elected Su chair of the board, according to the company's release.^[2] The chair leads the board of directors, the body that hires and oversees the chief executive and answers to shareholders. Holding the chair and the chief executive job at once concentrates both the management role and the board's leadership in one person, a structure some companies adopt and others deliberately separate.

Technical direction: AI accelerators

While Su was chief executive, AMD unveiled the Instinct MI300X accelerator, a GPU-only design with 192 GB of HBM3 memory.^[3][4] Two pieces of vocabulary make the specification meaningful. A GPU, or graphics processing unit, is a processor built to run many small calculations in parallel; that design turns out to suit the arithmetic behind machine learning, which is why GPUs became the standard hardware for training and running large models. HBM, or high bandwidth memory, is memory stacked in layers and placed close to the processor so data can move in and out faster than with conventional memory modules. Capacity and bandwidth both matter for AI work: a model's parameters and the intermediate data it produces have to fit in memory, and the faster that memory feeds the processor, the less time the chip spends waiting.

A "GPU-only" design means the accelerator is built around the GPU rather than combining it with general-purpose CPU cores on the same package. The MI300X's 192 GB of HBM3 is a capacity figure, and capacity is what determines whether a large model can be held on a single accelerator or must be split across several.

Place in the industry

Su's tenure spans a period in which the demand for AI computing reshaped what chipmakers were asked to build. Her public record combines an engineer's training — a doctorate in electrical engineering — with the operating and technology roles that precede a chief executive appointment: chief technology officer at Freescale, chief operating officer at AMD.^[1] The MI300X belongs to the product line AMD put forward for AI workloads during that period.^[3][4]