AMD plans to substantially increase chip supply in 2027 as demand for CPUs, GPUs and AI infrastructure continues to exceed available capacity. CEO Lisa Su says the company is now planning manufacturing requirements as far as five years ahead — a sign that the AI infrastructure boom is changing not only what AMD builds, but how far into the future it must secure wafers, packaging, substrates and memory.
AMD is preparing a significant manufacturing expansion for 2027 as demand continues to run ahead of supply across several of the company’s most important product categories.
Speaking with reporters in Taipei on October 6, AMD Chair and CEO Lisa Su said the company had steadily increased supply throughout 2026 but still needs additional manufacturing capacity.
“We’ve been able to increase our supply as we’ve gone through 2026, and we’re going to substantially increase our supply in 2027,” Su said, according to Reuters.
The expansion reflects demand not only for AMD’s Instinct AI accelerators, but also for EPYC server processors, client CPUs and other computing products.
More importantly, AMD is changing the way it plans its supply chain.
AMD Is Planning Chip Capacity Three to Five Years Ahead
Su said AMD and its manufacturing partners historically planned capacity roughly one or two years in advance. The company is now looking three, four and even five years ahead.
That longer planning cycle reflects the increasingly complicated manufacturing chain behind modern processors.
Producing an advanced CPU or AI accelerator requires more than securing wafer starts at a foundry. Wafer fabrication, advanced packaging, substrates, memory and backend assembly capacity must all be available at roughly the right time and in sufficient volume.
If one part of that chain falls short, additional capacity elsewhere may provide little benefit.
According to Focus Taiwan, Su said TSMC, AMD’s backend manufacturing partners and substrate suppliers are all working to expand capacity as the company prepares for greater demand.
She also said additional advanced wafer capacity will be required over the next several years.
AMD Is Expanding Its $10 Billion-Plus Taiwan Investment
Taiwan sits at the center of that expansion.
AMD announced in May that it would invest more than $10 billion across Taiwan’s technology ecosystem, targeting strategic partnerships and advanced packaging capacity for next-generation AI infrastructure.
That investment is proceeding as planned, according to Su, but AMD now expects to invest even more as demand has continued to climb.
The company has not disclosed a revised dollar figure.
During her October visit — her third trip to Taiwan in 2026 — Su met with companies including Acer, Asus, Foxconn and Quanta before traveling to Hsinchu for meetings with TSMC and other suppliers.
AMD’s Taiwan relationships extend across nearly every stage of its computing portfolio, from silicon fabrication and advanced packaging to server design and final system manufacturing.
“There’s no question, Taiwan is absolutely critical to the semiconductor supply chain and very, very critical for AMD,” Su said, according to Focus Taiwan.
Helios, MI450 and EPYC Venice Are Driving the Ramp
The timing of the supply expansion corresponds with some of AMD’s most ambitious data-center launches to date.
AMD said in May that its 6th Gen EPYC processor, codenamed Venice, had begun its production ramp using TSMC’s advanced 2nm process.
Venice is also a core component of AMD’s Helios rack-scale AI platform, which combines EPYC processors with the company’s next-generation Instinct MI450X accelerators.
AMD originally said Helios deployments using Venice and MI450X were scheduled to begin during the second half of 2026. Su said this week that the company began shipping Helios during the third quarter and expects shipments to continue increasing over the coming quarters.
That puts 2027 in an important position for AMD.
Rather than preparing for a single product launch, the company is attempting to scale an entire AI infrastructure platform while continuing to supply its traditional CPU and graphics businesses.
Advanced Packaging Is Becoming as Important as Wafer Capacity
One of the biggest changes created by modern AI accelerators is that manufacturing capacity can no longer be measured simply by how many wafers a foundry can process.
High-end AI processors combine multiple pieces of silicon with large amounts of high-bandwidth memory, requiring increasingly sophisticated packaging technologies.
TrendForce reports that AMD is expanding its MI400 and MI450 families as demand for advanced packaging continues to climb. TSMC’s CoWoS-L packaging technology is expected to remain one of the industry’s primary solutions for large AI processors through 2028.
AMD is also exploring additional packaging technologies with its Taiwanese supply-chain partners rather than relying on a single packaging method.
Su indicated during her visit that AMD expects both CoWoS and other advanced packaging technologies to remain important as AI systems become larger and more complex.
That diversification matters because adding additional wafer capacity does not solve an AI-chip shortage if there is insufficient packaging capacity available to assemble the finished accelerator.
Does More AI Production Mean Fewer Ryzen CPUs?
For PC builders, AMD’s aggressive AI expansion raises an obvious question: will data-center products consume manufacturing capacity that could otherwise be used for Ryzen or Radeon hardware?
The answer is not as simple as one product directly replacing another on the manufacturing line.
AMD’s largest AI accelerators rely on packaging technologies and high-bandwidth memory configurations that are very different from those used by mainstream desktop CPUs and consumer graphics products.
A Ryzen desktop processor does not require the same large CoWoS package and HBM integration used by a high-end Instinct accelerator.
That means additional AI packaging capacity does not automatically translate into fewer Ryzen processors.
However, the two markets are not completely isolated.
AMD still competes for finite leading-edge wafer capacity, substrates, backend manufacturing resources and other components across its product portfolio. As client processors migrate onto newer process nodes, AMD must decide how manufacturing capacity is distributed across server, AI and consumer products.
For enthusiasts, that is the more important long-term consideration.
Memory May Be the Bigger Problem for PC Builders
The clearest connection between the AI infrastructure boom and consumer PCs may actually be memory rather than processors.
High-bandwidth memory has become one of the most important components in modern AI accelerators, and memory manufacturers are devoting an increasing portion of their DRAM production capacity to it.
TrendForce estimates that HBM wafer input among the three largest memory manufacturers will rise from approximately 22% of total DRAM wafer input in 2026 to around 30% in 2027.
Samsung has similarly indicated that HBM could consume close to 30% of industry DRAM wafer capacity next year.
That matters because HBM and conventional DRAM ultimately compete for manufacturing resources.
HBM also consumes substantially more wafer area than conventional DRAM. Industry estimates cited by TrendForce suggest an HBM die can require roughly three times the wafer area needed for an equivalent amount of standard DDR5 memory.
The result is a capacity tradeoff: rapidly increasing HBM production can reduce how much manufacturing headroom remains for conventional DRAM.
TrendForce expects DRAM supply to remain tight through 2027 as AI servers absorb growing quantities of HBM and server memory.
That dynamic has a much clearer path to affecting ordinary PC builders through DDR5 pricing and availability than AMD simply ordering additional Instinct accelerators.
ITD Insight
AMD’s 2027 expansion should not be interpreted as the company choosing AI accelerators over Ryzen processors. The bigger story is that AI has made the semiconductor supply chain interconnected in ways that extend far beyond the GPU itself. Leading-edge wafers, advanced packaging, substrates and especially memory all have to scale together. For PC builders, the most visible consequence may therefore show up in memory and component pricing rather than an outright shortage of Ryzen CPUs.
AMD’s Growth Has Changed the Scale of the Problem
The urgency behind AMD’s expansion also reflects how quickly the company itself has grown.
AMD’s market capitalization passed $1 trillion for the first time in September 2026 as investors continued betting on the company’s ability to win a larger share of the rapidly expanding AI infrastructure market.
That puts AMD in a dramatically different position than it occupied only a few years ago.
The company is no longer simply challenging Intel in CPUs or Nvidia in discrete graphics. It is attempting to become a major supplier of complete rack-scale AI infrastructure, combining processors, accelerators, networking, memory connectivity, software and system-level designs.
That requires a manufacturing footprint capable of supporting customers planning data centers several years before those systems are actually deployed.
AMD’s shift to three-to-five-year capacity planning illustrates just how much that business has changed.
HBM Supply Is Also Sending AMD to South Korea
Taiwan is only one piece of the supply chain.
Reuters reported that Su is also expected to travel to South Korea, where AMD plans to meet with major memory suppliers including Samsung Electronics and SK Hynix.
The purpose is particularly relevant to AMD’s AI ambitions.
Modern Instinct accelerators depend heavily on HBM, and the memory industry’s capacity is already under substantial pressure from competing accelerator and AI-ASIC programs.
TrendForce expects the supply-demand environment for HBM to remain tight through 2027, with HBM4 becoming increasingly important for next-generation AI processors.
Securing memory allocation several years in advance is therefore becoming just as strategically important as reserving wafer or packaging capacity.
Why AMD’s 2027 Supply Expansion Matters
The most important part of Lisa Su’s comments is not simply that AMD intends to manufacture more chips next year.
It is that the company no longer believes a traditional one- or two-year manufacturing plan is sufficient.
AMD is attempting to coordinate wafers, advanced packaging, substrates, HBM, backend manufacturing and system assembly across a three-to-five-year window because the infrastructure required to build tomorrow’s AI hardware cannot be created overnight.
That strategy should give AMD a better chance of avoiding supply constraints as its Instinct and EPYC businesses expand.
It also illustrates why the AI boom can affect parts of the technology market that appear unrelated to AI.
A shortage of HBM capacity can influence conventional DRAM production. Increased demand for advanced-node wafers can tighten foundry allocation. Larger accelerators create additional demand for packaging and substrates. Each layer interacts with the others.
For PC buyers, that does not mean Ryzen processors are about to disappear from shelves.
But it does mean the cost and availability of consumer hardware will increasingly be shaped by a much larger contest for the manufacturing resources needed to feed the global AI buildout.
Sources: Reuters, Focus Taiwan, AMD, and TrendForce.