AMD → ZT Systems
AMD paid $4.9bn in cash and shares for ZT Systems, a New Jersey builder of servers for the largest cloud companies, to gain the rack-scale design skills it needed to sell complete AI systems — then sold ZT's factories to Sanmina for up to $3bn and kept the design teams.
A cash-and-stock purchase of the hyperscale server builder for rack-scale AI systems; AMD sold the manufacturing arm to Sanmina in October 2025.
- Value
- $4.9B
- Announced
- August 2024
- Completed
- March 2025
- Sector
- Technology
- Deal type
- acquisition
- Acquirer
- AMD (United States)
- Target
- ZT Systems (United States)
Why it mattered
AMD sold Instinct AI accelerators and EPYC processors, and argued that its customers would gain from AI systems delivered as complete racks, ready to deploy at scale. ZT Systems, based in Secaucus, had spent more than 15 years designing and deploying compute and storage infrastructure for the biggest cloud companies. AMD agreed on 19 August 2024 to buy it for $4.9bn, including up to $400m payable if post-closing milestones were met, and said at the outset that it would look for a strategic partner to take over ZT's US manufacturing business. A month earlier AMD had agreed to buy the Finnish AI start-up Silo AI for $665m.
AMD put the data-centre AI accelerator opportunity at $400bn in 2027. The deal closed on 31 March 2025, by which time it spoke of $500bn in 2028. ZT's president Doug Huang joined AMD to run the design and customer-enablement teams; its founder and chief executive Frank Zhang was to help lead the sale of the factories.
What it left behind
On 19 May 2025 Sanmina agreed to buy ZT's manufacturing business — plants in New Jersey, Texas and the Netherlands with revenue running at about $5bn to $6bn a year — for $2.25bn in cash for the net assets, a $300m premium paid half in cash and half in Sanmina shares, and up to $450m more depending on performance over three years. Sanmina became AMD's preferred US manufacturing partner for new AI rack products, and the sale completed on 27 October 2025, ahead of schedule. AMD kept the design teams, and its Helios rack-scale system, unveiled in 2025, won large commitments: in late 2025 OpenAI agreed to deploy 6 gigawatts of AMD GPUs and Oracle announced a Helios-based cluster starting with 50,000 GPUs, and by August 2026 AMD said Helios was being deployed by customers including Meta, Microsoft, OpenAI and Oracle. The structure is worth copying: buy the whole company to get its know-how, then sell the capital-heavy part to a specialist that becomes your supplier, so the divestiture recovers cash and secures capacity at the same time.