AMD's Data Center Revenue Doubles as It Becomes AI's Second Engine

AMD's data center revenue reached $8.546 billion in the latest quarter, up 97% year over year, and the segment now produces 64% of total company revenue, up from 47% a year earlier. Segment operating margin climbed to 34% from 25%. Those numbers explain why one 24/7 Wall St. columnist keeps buying the stock, and why the market has started treating AMD as the second engine of the AI buildout rather than a permanent understudy.

The share price reflects the shift. AMD stock is up 158.17% year to date, while Nvidia is down 6.63% over the same period and down 9.84% across the past year. The divergence is not simply a Nvidia problem, the columnist argues. It rests on four structural advantages that hyperscalers and neoclouds keep paying for.

What are the four advantages?

• Cost. CEO Lisa Su frames Instinct accelerator sales around "tokens-per-dollar" advantages, a metric that speaks directly to buyers running inference at scale.
• Memory. AMD says its Helios rack-scale platform offers "an even larger lead in memory bandwidth and higher scale-up bandwidth than the competition."
• Openness. AMD contributed the Helios rack design to the Open Compute Project, letting buyers avoid lock-in to a single supplier.
• Breadth. AMD sells both the accelerator and the CPU, and the rise of AI agents lifts demand for both.

Five lead customers, including Anthropic and OpenAI, already run workloads on Helios racks. That is a confirmed deployment figure, not a roadmap promise.

Are profits keeping pace?

Yes, and then some. Non-GAAP EPS reached $2.09, up 74%, while revenue grew 45% — the fourth consecutive EPS beat. The CFO expects 2027 operating expenses to grow at less than half the top-line revenue growth rate, which implies continued margin leverage.

Guidance points the same direction. Fourth-quarter revenue is pegged at about $17 billion, roughly 66% growth at the midpoint. AMD expects 2027 revenue to grow more than 80%. Early demand for Venice, the newest EPYC generation, is "the strongest we have seen for any new EPYC generation," according to the company.

Why does Nvidia look vulnerable?

Nvidia still runs an elite business with a 75.0% gross margin. The strain is showing in its balance sheet, though. Supply obligations rose to $279.00 billion, partner guarantees run up to $108.5 billion, and customers now take 60 days to pay, compared with 45 days before. Shares trade at about 35 times earnings.

Intel presents a different contrast. Its data center unit grew 78%, but the foundry business lost $1.85 billion last quarter. Intel guides to a 38.0% non-GAAP gross margin against AMD's roughly 55%, and carries borrowings at 0.41 times equity versus AMD's 0.071.

What could hurt the position?

The risks are concrete. AMD pays no dividend. The stock trades at 135 times free cash flow, and free cash flow fell 70.86% to $554 million as inventory rose to $11.463 billion. A $1,097 million customer warrant charge reduced GAAP EPS to $1.46. Sales depend heavily on a few large customers, and export controls remain a live threat to the business.

Su offers a defense of the inventory build: "demand is higher than supply," and AMD stocked up to support fourth-quarter data center growth of about 50% sequentially. Helios production is increasing, Venice is shipping, and AMD is already working with customers on the MI500 and MI600 generations.

Hyperscalers want a second supplier for AI chips. On these numbers, AMD has become the one they are actually paying for.