Unlock the Editor’s Digest free of charge
Roula Khalaf, Editor of the FT, selects her favorite tales on this weekly publication.
AMD has agreed to accumulate synthetic intelligence infrastructure group ZT Methods in a $4.9bn money and inventory transaction, extending a run of AI investments by the chip firm because it seeks to problem market chief Nvidia.
The California-based group mentioned the acquisition would assist speed up the adoption of its Intuition line of AI knowledge centre chips, which compete with Nvidia’s standard graphics processing models.
ZT Methods, a non-public firm based three a long time in the past, builds customized computing infrastructure for the most important AI “hyperscalers”. Whereas the corporate doesn’t disclose its prospects, the AI hyperscalers embody the likes of Microsoft, Meta and Amazon.
It marks AMD’s greatest acquisition because it purchased Xilinx for $35bn in 2022.
“It brings a thousand world-class design engineers into our group, it permits us to develop silicon and programs in parallel, and most significantly get the most recent AI infrastructure up and operating in knowledge centres as quick as potential,” AMD’s chief govt Lisa Su informed the Monetary Instances.
“It actually helps us deploy our expertise a lot quicker as a result of that is what our prospects are telling us [they need],” Su added.
The transaction is predicted to shut within the first half of 2025, topic to regulatory approval, after which New Jersey-based ZT Methods will probably be folded into AMD’s knowledge centre enterprise group. The $4.9bn valuation contains as much as $400mn contingent on “sure post-closing milestones”.
Citi and Latham & Watkins are advising AMD, whereas ZT Methods has retained Goldman Sachs and Paul, Weiss.
The transfer comes as AMD seeks to interrupt Nvidia’s stranglehold on the AI knowledge centre chip market, which earlier this yr noticed Nvidia quickly change into the world’s most beneficial firm as huge tech corporations pour billions of {dollars} into its chips to coach and deploy highly effective new AI fashions.
A part of Nvidia’s success stems from its “programs” strategy to the AI chip market, providing end-to-end computing infrastructure that features pre-packaged server racks, networking gear and software program instruments to make it simpler for builders to construct AI purposes on its chips.
AMD’s ZT Methods acquisition exhibits the chipmaker constructing out its personal “programs” providing. The corporate rolled out its MI300 line of AI chips final yr, and says it can launch its next-generation MI350 chip subsequent yr to compete with Nvidia’s new Blackwell line of GPUs.
In Might, Microsoft was one of many first AI hyperscalers to undertake the MI300, constructing it into its Azure cloud platform to run AI fashions equivalent to OpenAI’s GPT-4. AMD’s quarterly income for the chips surpassed $1bn for the primary time within the three months to June 30.
However whereas AMD has feted the MI300 as its fastest-ever product ramp, its knowledge centre income nonetheless represented a fraction of the $22.6bn that Nvidia’s knowledge centre enterprise raked in for the quarter to the top of April.
In March, ZT Methods introduced a partnership with Nvidia to construct customized AI infrastructure utilizing its Blackwell chips. “I believe we actually consider ZT as a part of AMD will considerably speed up the adoption of AMD AI options,” Su mentioned, however “we’ve got buyer commitments and we’re actually going to honour these”.
Su added that she anticipated regulators’ evaluate of the deal to deal with the US and Europe.
Along with growing its analysis and improvement spending, AMD says it has invested greater than $1bn over the previous yr to broaden its AI {hardware} and software program ecosystem.
In July the corporate introduced it was buying Finnish AI start-up Silo AI for $665mn, the biggest acquisition of a privately held AI start-up in Europe in a decade.
