READ MORE
AMD’s market cap briefly passed $1 trillion on Monday, making the perpetual underdog one of a handful of chip designers to ever hold the distinction.
Make no mistake, AMD is still the underdog here, at least relative to Nvidia, which holds the distinction of being the most valuable company in the world with a market cap of nearly $5.5 trillion.
But while AMD still trails Nvidia by a wide margin, the House of Zen has taken the lead over its long-time rival turned frenemy Intel, which is currently valued at $640 billion.
Riding the AI wave
Much of AMD’s success comes on the back of the AI boom.
This didn’t happen overnight. AMD was late to the AI party having invested most of its GPU development resources in more traditional high-performance computing applications and national supercomputing projects. That changed in 2023 when it launched the Instinct MI300A — a part that, at least on paper, delivered higher performance, more memory, and greater bandwidth than Nvidia’s then new H100 and H200-series GPUs.
But while better on paper, AMD’s products were dogged by un-optimized software, much of which had been designed from the ground up for Nvidia accelerators.
AMD’s GPU team has spent the better part of three years changing the narrative around ROCm, helped by large-scale deployments of MI300- and MI350-series GPUs by the likes of Microsoft, OpenAI, Oracle, Anthropic, and Meta, with lower prices and higher memory capacity being key differentiators.
Key wins with Neoclouds and smaller cloud service providers like TensorWave and Vultr helped to expose the open source community to AMD’s accelerators.
By mid-2026, AMD claimed it had not only closed the performance gap with Nvidia but was also on track to launch a new rack-scale compute platform in the third quarter that was bigger and faster than Nvidia’s existing Blackwell-based rack systems and Vera Rubin.
Compared to Nvidia’s own bleeding edge AI systems, AMD’s Helios rack boasts 50 percent more HBM4 and scale-out bandwidth and between 15 and 25 percent higher performance for AI training. And, if AMD is to be believed, Helios will deliver a 30 percent performance per dollar lead over the competition. That’s a hard number to validate, but for those on Wall Street, it’s a clear indication as to why the largest and most influential AI labs and hyperscalers have flocked to the platform.
Not just Instinct
However, GPUs aren't the only reason AMD's valuation has skyrocketed in recent years. They're just the hardest to ignore. In February, Aaron Rakers at Wells Fargo Securities estimated that Instinct GPU sales accounted for between $2.5 billion and $2.6 billion out of $10.3 billion in revenues during the fourth quarter of 2025, roughly matching Epyc CPU revenues.
The rise of agentic AI workloads like OpenClaw and Claude Code has fueled demand for CPUs. While the models themselves still run on GPUs, the code they generate and tools they execute still need CPUs to run on. Here there’s a lot more competition. In addition to its long-time rival Intel, AMD is competing against Arm, Qualcomm, Nvidia, and the cloud providers themselves.
But just like we saw with Instinct, AMD’s Epyc processors, which have consistently offered high core counts per dollar and, in more recent generations, higher performance per dollar than the competition, have helped the chip designer grow its market share by a considerable margin.
In August, AMD’s share of the desktop CPU market crossed 35 percent. As we wrote in our review earlier this year, AMD now makes the world’s most powerful desktop CPU, and is largely uncontested in the workstation arena thanks in part to rebadging unlocked Epycs as high-core-count Threadripper CPUs.
In the datacenter, AMD’s share jumped to 34.5 percent, while Arm-based CPUs from Ampere, Amazon, Microsoft, Google, and others accounted for another 13.6 percent of the market. As Mercury Research, the analyst firm behind the numbers, noted, AMD’s share rises to 46.4 percent when the comparison is limited to Epyc and Intel Xeon SP processors.
These gains have been aided by an Intel that’s struggled to put out competitive Xeon processors. Ever since Chipzilla whiffed it on its Sapphire Rapids processors and gave AMD nearly two years to catch up, Intel has been trying to close the gap.
Intel came close in 2024 with its Xeon 6 processors, which remain extremely popular as host CPUs for Nvidia’s smaller GPU systems, but an executive shake up and the cancellation of its mainstream Xeon 7 SP products mean Intel is unlikely to have a competitive datacenter CPU, potentially until 2028.
AMD is facing some pressure from Nvidia’s Vera and Arm’s AGI CPUs, but largely in AI-centric applications like GPU head nodes, agent sandboxes and KV cache offload servers. AMD is able to address a much broader market thanks to a modular design that can support a wider array of target applications.
But AMD’s share gains don’t stop at CPUs and GPUs. Over the past decade, AMD has amassed a large portfolio of technologies ranging from smartNICs and DPUs to FPGAs, and even system design through acquisitions. AMD’s acquisition of Xilinx is particularly notable as while not a particularly large revenue driver, it established the company as the largest supplier of FPGAs in the market and armed the chip designer with a mountain of IP, which has repeatedly helped it stay competitive. For example, the NPUs, which assured AMD’s qualification for Microsoft’s Copilot+ moniker, were lifted directly from Xilinx’s IP holdings.
Where AMD goes from here
Just as AMD became a useful tool for keeping pricing pressure on Intel CPUs, the same is now playing out for datacenter GPUs with Nvidia. Alongside the launch of Helios in July, AMD announced Anthropic had agreed to deploy up to 2 gigawatts of Instinct accelerators to support AI training and inference. OpenAI, Oracle, Microsoft and Meta also plan to deploy the rack systems at scale.
“In datacenter AI, the growing number and scale of Helios and MI450-series deployments position the [datacenter] business for significant growth in the second half of the year, with growth accelerating in 2027," CEO Lisa Su told investors last month.
"We now expect datacenter segment revenue to more than double year over year in 2027,” she added.
Yet, despite the enduring optimism of AMD’s top executives, Wall Street’s feelings on the company remain mixed with the main concern being that AMD may have too many AI eggs in too few baskets.
On the flip side, with such an expansive portfolio spanning general purpose compute, HPC, AI, embedded systems, desktop, and gaming, the case can also be made that while nowhere near Nvidia’s size, AMD's exposure to an AI bubble is far, far smaller. ®
