In 2017, there was a time when mentioning AMD in a hyperscale data center seemed almost absurd. The rooms were run by Intel. Every server, every rack, and every procurement spreadsheet all pointed to the same Santa Clara behemoth whose name had come to represent the silicon that powers the global internet. At the time, AMD’s server CPU share was almost zero. Not close to zero symbolically. Actually, it’s almost zero. Then something changed that few outside the chip industry fully understood, somewhere between Naples and Genoa, between Lisa Su’s quiet engineering bets and Intel’s increasingly visible manufacturing struggles.
AMD’s server CPU revenue share increased to about 41% by the beginning of 2026. It was about 25% three years ago. The simplicity of the math is brutal. More than four out of ten dollars spent on the most lucrative category of computer hardware worldwide are now captured by a company that, less than ten years ago, virtually did not sell server chips. After the company’s printing press was forced open, Intel’s data center business was established. When you walk through the procurement story of any major cloud customer, it seems like the conversation no longer automatically begins with Xeon.
Zen was the architectural turn, but letting someone else construct the chips was the strategic turn. At the time, AMD’s 2009 spin-off of its factories into GlobalFoundries appeared to be akin to surrender. Jerry Sanders famously said, “Real men have fabs,” and AMD withdrew. What appeared to be a retreat turned into leverage. By partnering with TSMC, AMD was able to benefit from the Taiwanese foundry’s unrelenting push to 7nm and then 5nm, while Intel was mired in a multi-year maze attempting to operate its own 10nm node. Cloud architects took note. They are always observing. When operating a hyperscale farm in Oregon and receiving monthly electricity bills, power-per-watt is not an abstract metric.
Rome was the first to crack. Milan was part of the second wave. By the time Genoa and Bergamo arrived, discussing EPYC in cloud meetings was the norm rather than the exception. They were rolled in by Amazon Web Services. Microsoft Azure came next. They were added across instance families by Google Cloud. They were distributed throughout Meta’s own specially designed server fleets. AMD’s “land and expand” strategy seems to have succeeded because it never demanded everything at once. With more cores, improved I/O, and prices that made finance teams nod silently, it simply kept coming up.
Meanwhile, Intel continued to falter at precisely the wrong times. Sapphire Rapids slipped, then slipped once more. Customers were forced to wait due to supply constraints during the chip crunch, and those who did so looked elsewhere. AMD reported a record $12.6 billion in data center revenue through 2024, up 94% year over year. This was the first time in modern memory that the line that defined Intel’s identity had been surpassed. AMD’s data center business reached $4.3 billion in a single quarter by Q3 2025, surpassing the total revenue reported by Intel’s data center group in some recent comparable periods.
To interpret this as Intel collapsing would be incorrect. The straightforward story is complicated by Intel’s recent comeback, which includes a solid Q1 2026 earnings beat, gross margins rising back toward 41%, and Google’s stated commitment to deploy future-generation Xeons. Granite Rapids deployments have been publicly pledged by two hyperscalers. The battle is still ongoing. Simply put, it’s no longer biased.
This story is unique because of its patience. AMD did not reinvent the category in a single Apple-style keynote. For six years, cloud companies used quarterly earnings calls, made gradual roadmap promises, and tested an EPYC sample one engineer at a time. It’s difficult to ignore how infrequently the most significant changes in the industry occur without much fanfare. It’s easy to forget that ten years ago, AMD’s stock was trading for less than $3 as you watch Lisa Su accept industry awards. The business almost failed. The data center became its own after that.
