How AMD went from Intel’s smaller rival to a $1 trillion AI contender

Advanced Micro Devices, better known as AMD, has come a long way from being Intel’s smaller rival in PC processors.

The chipmaker crossed a $1 trillion market capitalisation for the first time on September 21, as investors bet on its expanding role in AI computing. Its shares were trading 9.6% higher at a record $613.31, making it the fourth US chipmaker to reach the milestone, after Nvidia, Broadcom and Micron. The stock’s gain for 2026 stood at 185%, against the Nasdaq’s 15.8%, Reuters reported. On September 23, AMD shares closed at $614.

But AMD’s $1 trillion moment is more than just an AI story. It is the latest chapter in a transformation that began more than a decade ago.


“AI provided multiple expansion, but its EPYC CPU business built the cash engine for AMD,” Shrish Pant, director analyst at Gartner, told ET AI.

Rebuilding the core

When Lisa Su took over as AMD’s CEO in October 2014, the company was in a difficult position. Revenue fell from $5.5 billion that year to $4 billion in 2015, and AMD reported a net loss of $497 million in 2016, according to its financial results.

One of the decisions that would shape AMD’s future came several years before Su took over. In 2009, AMD spun off its wafer-manufacturing business, creating GlobalFoundries, a dedicated semiconductor manufacturing company, shifting towards a fabless model in which it designed chips and outsourced their production.

The move reduced the burden of owning and operating expensive chip factories. Over time, AMD increasingly relied on external manufacturers such as Taiwan Semiconductor Manufacturing Co. (TSMC), while Intel continued to invest heavily in its own fabrication network.

AMD’s product revival under Su centred on Zen, a processor design used across PCs and servers. The first Zen-based Ryzen desktop processors arrived in March 2017, renewing its challenge to Intel in PCs. EPYC server processors followed that June, taking the fight into data centres.

AMD expanded its use of chiplets with Zen 2 in 2019. This approach combines multiple smaller pieces of silicon in one processor, rather than relying on a single large chip. Pant said chiplets improved manufacturing yields, or the proportion of usable chips, while access to TSMC’s advanced manufacturing technology helped AMD as Intel moved more slowly.

The trust AMD built through selling EPYC over the past decade gave it a “seat at the table” when it began selling its Instinct AI accelerators, Pant said. AMD entered the AI race with an established data-centre business and customer relationships.

By 2025, its annual revenue had reached $34.6 billion. The data centre business generated $16.6 billion, up 32% from $12.6 billion in 2024, driven by demand for EPYC processors and Instinct graphics processing units, or GPUs. Client and Gaming revenue stood at $14.6 billion, according to the company.

Building beyond CPUs

AMD had also been assembling the pieces of a business that went beyond central processing units, or CPUs.

Its acquisition of graphics-chip maker ATI in 2006 gave it a GPU business. The purchase of Xilinx, completed in 2022, added programmable chips that can be reconfigured for different tasks. The all-stock deal had been valued at $35 billion when announced in 2020.

AMD also acquired data centre optimisation startup Pensando Systems in 2022, adding data-processing and networking capabilities, followed by AI software companies Nod.ai in 2023 and Silo AI in 2024.

Together, these deals broadened the range of computing components and software AMD could offer data centre customers. That became increasingly important as AI workloads drove demand for computing infrastructure.

AMD’s Instinct accelerators compete with Nvidia’s chips in the market for training and running AI models. Its ROCm software provides the tools developers need to run AI workloads on AMD GPUs.

But AMD’s ambitions have moved beyond the accelerator itself.

From chips to the AI rack

In March 2025, AMD completed its acquisition of ZT Systems for $3.2 billion in cash and 8.3 million AMD shares, according to its annual report. It retained ZT’s design operations and sold the data centre infrastructure manufacturing business to Sanmina Corporation that October, making the latter a preferred manufacturing partner.

The deal strengthened AMD’s ability to design complex AI systems in which processors, networking and software work together across racks of servers.

“AMD is no longer just an alternate silicon company. They are selling rack-scale architectures,” Pant said.

With ZT for system design, Pensando for networking and ROCm for software, AMD is following a similar path to Nvidia’s expansion into complete data centre systems, he added.

AMD’s Helios platform brings these pieces together, combining Instinct GPUs, EPYC CPUs, Pensando networking and ROCm software. The company said in July that Helios was in production, with OpenAI expecting to bring the systems online from the fourth quarter of 2026.

Parv Sharma, senior research analyst at Counterpoint Research, said the liquid-cooled Helios platform showed how AMD’s role was changing.

“It marks AMD’s transition from being a component supplier to a full-system provider for hyperscale data centres,” he told ET AI, referring to the large facilities run by cloud companies.

Moving from individual accelerators to full racks gives AMD an opportunity to earn more revenue from each deployment and deepen customer relationships, Sharma said.

He also sees a fresh opportunity for the CPU business as AI agents take on tasks across applications.

“Agents orchestrate virtual machines, browsers and data flows, and those workloads run mostly on CPUs. AMD’s Venice CPU ships inside every Helios rack and also goes into general-purpose servers,” Sharma said.