Lisa Su's Zen bet brought AMD its first profit since 2011 in 2017

By the early 2010s AMD was the perennial runner-up to Intel in CPUs and Nvidia in GPUs, and its Bulldozer architecture had been an unmitigated performance failure. From 2012 through 2016 it accumulated US$2.8bn of losses and had essentially vanished from the server CPU market.

Under CEO Lisa Su, AMD made one all-in bet on a brand new x86 design called Zen, followed by the Ryzen desktop CPUs and the EPYC server CPUs launched in 2017. The bet was that a fresh architecture, not a cheaper version of an old one, could win back customers.

It worked. AMD reported 2017 revenue of US$5.33bn, up 25%, a swing to a US$43m net profit after a half-a-billion-dollar improvement from 2016, and a higher gross margin. Su called 2017 a key inflection point in the company's history, with full-year profitability restored for the first time since 2011.

The company still faced a large single-thread performance gap to Intel and a weak position in high-end graphics, but the product had brought it back into both markets it had lost.

AMD swung to a US$43m net profit in 2017 after 2012-16 losses of US$2.8bn, with revenue up 25% to US$5.33bn and gross margin expanding, on the Ryzen and EPYC launches.

Replacing the failed Bulldozer design with a brand-new Zen architecture fixed the core product problem.

Ryzen re-entered the desktop PC market and EPYC re-entered the server market, both lost for years.

A richer product mix and better design lifted gross margin, which had been chronically low.

Cost discipline and new platforms let revenue growth fall straight to the bottom line.

A company that has lost product leadership cannot market its way out; it has to rebuild the architecture from first principles and accept the long, capital-hungry bet on a single generation of chips.

AMD kept improving on Zen through the following generations and eventually rode the demand for AI accelerator chips, turning a perennial runner-up into a genuine rival to Intel and Nvidia. The turnaround showed an architecture bet can rescue a chipmaker, but the company remains vulnerable to the same execution risk that nearly killed it.