AMD is committing more than $10 billion to Taiwan's semiconductor ecosystem, funding advanced chip packaging, substrates, and manufacturing capacity through 2029. The spending targets packaging bottlenecks that have constrained AI chipmakers, including rival Nvidia, as AMD works toward its data-center growth targets.
AMD plans to invest more than $10 billion in Taiwan, spreading the money across the island's broader chip ecosystem rather than to TSMC alone. The spending covers advanced packaging, chip substrates, and manufacturing capacity for complete AI systems, and it is set to run through 2029. CEO Lisa Su is investing now so AMD can manufacture enough hardware if its AI demand turns into large-scale deployments.
AMD widens its packaging supply chain
AMD's new Venice EPYC server CPU is already ramping production on TSMC's 2-nanometer process, and the company also uses TSMC's SoIC-X and CoWoS-L packaging for some AI and data-center chips. Beyond TSMC, AMD is developing Elevated Fanout Bridge (EFB) packaging with ASE Technology and Siliconware Precision Industries, and it has completed testing of a panel-based EFB version with Powertech Technology. Crypto Briefing reported that the EFB partner roster also includes Sanmina, an established semiconductor assembly and packaging player.
A bottleneck AMD shares with Nvidia
In July 2026, TSMC CEO C.C. Wei said tight advanced-packaging capacity was limiting customers' growth, meaning AMD could win AI customers but still miss sales if it cannot package and assemble enough chips. Nvidia has leaned heavily on TSMC's CoWoS technology, and Crypto Briefing reported that limited capacity for that process has been a genuine constraint on the broader AI chip market. AMD's EFB push builds capacity in a comparable but distinct packaging architecture.
Growth targets raise the stakes
AMD's data-center revenue rose 107% year-over-year to $6.7 billion in the second quarter, about 58% of total revenue. Management expects that segment to grow at a compound annual rate of more than 60% over the next three to five years, including more than 80% annually for data-center AI. A Helios AI rack packs 72 Instinct MI455X GPUs and 18 Venice CPUs, and OpenAI, Meta Platforms, and Anthropic have announced AMD deployments that could total as much as 14 gigawatts over several years.
AMD spent only $1.2 billion on property and equipment in the first half of 2026, yet it exited the second quarter with $30.3 billion of broader unconditional commitments covering wafers, substrates, components, cloud capacity, software, and technology licenses. Its non-GAAP operating margin stood at 27% in the second quarter, below management's target of more than 35% over the next three to five years. Su is spending billions now to make sure AMD can produce enough AI hardware if customer demand grows as expected.
Sources: The Motley Fool, Crypto Briefing
Trading involves risk.
Risk Disclaimer
