AMD's Data-Centre Sales Nearly Double, Just as Politics Enters the AI Trade

On the evening the AI trade got its first serious political scare, it also got one of its strongest earnings reports. AMD reported its third-quarter results after Tuesday’s close, and the numbers, from the company’s release, were striking:

• Revenue was a record $13.4 billion, up 45% from a year earlier and above analysts’ forecasts of about $13.1 billion.
• Data-centre revenue was about $8.5 billion, up 97%. It now makes up 64% of the company’s sales, up from 47% a year ago.
• Adjusted earnings per share were $2.09, against expectations of $1.95. (On a GAAP basis, EPS was $1.46.)
• Guidance for the fourth quarter was about $17 billion in revenue, give or take 3%. Analysts had expected about $16.2 billion.

That forecast implies revenue jumping by roughly a quarter in a single quarter, which is rare for a company of AMD’s size. Management said data-centre revenue should grow about 50% from the third quarter to the fourth.

“Faster than we projected”

Chief executive Lisa Su said data-centre AI growth is accelerating faster than the company projected just 90 days ago. She expects AMD’s server revenue to double in 2027, and said AMD plans to “substantially increase our supply” next year, having worked to secure wafers, packaging and memory across its supply chain. “Adoption is just starting,” she said, “and we expect demand to grow sharply as agents are deployed more broadly.”

The products driving this are AMD’s Instinct AI accelerators, its answer to Nvidia’s chips, and its Helios rack-scale systems, which package many of those chips into complete server racks. AMD named Meta, Microsoft, OpenAI, Oracle and Anthropic among the companies beginning to use Helios.

The stock rose about 8% in after-hours trading, then pared that gain as election results came in. On Wednesday morning it was up about 7% before the open, but nearly flat by 9 a.m. as data-centre stocks sold off on the Texas election result. The shares had already more than doubled this year.

The weaker parts

Not everything grew. Revenue from PC processors and gaming chips fell 8% combined, to $3.7 billion, and gaming alone was down about 40%. Fourth-quarter gross margin guidance of 55% was slightly below expectations. Part of the reason is the same one Apple gave this week: memory has become much more expensive, and that hurts consumer devices most.

AMD is now mainly an AI data-centre company with a shrinking consumer business attached, rather than the PC and console chipmaker many investors still think of.

Follow the money

One detail deserves attention. AMD isn’t only selling to AI companies. It’s also investing in them. It has committed to invest up to $5 billion in Anthropic, one of its customers, and agreed to put $1 billion into Meta’s recent share offering.

Arrangements like this have become common across the industry. Chipmakers invest in AI labs and cloud companies, which then buy their chips. Some of the money that shows up as a supplier’s revenue started out as the supplier’s own investment. That doesn’t make the demand fake, since these customers are spending far more than they receive. But it does mean the companies in the AI supply chain are more financially tied to each other than their separate share prices suggest. If one part of the chain stumbles, the effects travel.

Two signals, one morning

AMD’s report and the Texas election pull in opposite directions:

• Demand for AI computing is still rising, and faster than the industry’s own forecasts. Super Micro, which builds AI servers, also reported revenue well ahead of expectations on Tuesday.
• Capacity to house that computing is becoming constrained, by power supply, local opposition and now state politics.

When demand outruns the ability to build, the companies with scarce resources gain pricing power. Today that means chipmakers with secured supply and data-centre operators with approved sites and grid connections. For investors, the AI trade is turning from a bet on whether demand exists into a bet on who controls the bottlenecks.