On October 21, 2026, NVIDIA announced a plan with six suppliers, including Japan's Ibiden, to expand manufacturing capacity for AI infrastructure. Over the next three years, the company will increase its package substrate capacity for NVIDIA products 2.5-fold, while also boosting semiconductor testing and optical communications components. The aim is to strengthen the entire supply chain, not just secure GPU chips: packaging, testing, and the optical components that interconnect AI systems. However, the timing of the effects differs by process, and Ibiden will not begin ramping up its additional capacity until 2028.
Six NVIDIA suppliers to expand four manufacturing processes
The plan, announced by NVIDIA at GTC Berlin, includes three package substrate suppliers: Ibiden, Samsung Electro-Mechanics, and Unimicron Technology. They are joined by Amkor Technology, which handles advanced packaging; King Yuan Electronics (KYEC), which handles semiconductor testing; and Siliconware Precision Industries (SPIL), which makes optical communications components. The plan expands four processes in parallel: substrates, advanced packaging, testing, and optical communications. NVIDIA CEO Jensen Huang said, "Demand is extraordinary. Capacity is the constraint," indicating that securing supply capacity is urgent.
Package substrates are used to mount semiconductor chips such as GPUs and connect them to external circuits. Through agreements with Ibiden, Samsung Electro-Mechanics, and Unimicron, NVIDIA will increase its total substrate capacity 2.5-fold. Amkor, meanwhile, is expanding in South Korea the wafer capacity used to make interposers, the intermediate substrates that connect chips. The company aims to increase this capacity by more than 300% by the end of 2028. That means an increase of more than three times current capacity, so total capacity after the expansion would be more than four times today's level.
In optical communications, SPIL is increasing production in Taiwan of optical engines for co-packaged optics (CPO), which integrate optical components and semiconductors. It plans to expand capacity 15-fold by mid-2027 and double it again by the end of 2028. NVIDIA's overall semiconductor testing capacity is also expected to double between 2026 and 2027. However, the 2.5x for substrates, more than 4x for interposers, and 15x for optical engines are capacity targets for different manufacturing processes. Baseline absolute output and per-product allocations have not been disclosed, so multiplying these figures together cannot yield GPU supply or total AI system production capacity.
Ibiden's expansion starts in 2028; testing equipment begins volume production in 2027
Ibiden aims to triple the capacity of its Ono plant from the current level, with additional capacity starting up in the first quarter of 2028. Under a contract Ibiden announced earlier in 2026, NVIDIA will make an advance payment worth billions of dollars to secure capacity. The exact amount of the advance payment and the timing for the Ono plant to reach its target capacity have not been disclosed.
South Korea's Samsung Electro-Mechanics will begin supplying NVIDIA immediately from existing production sites, newly joining the company's supply chain. Equipment upgrades and new plant construction are also under way in Sejong and Busan, and by 2028 NVIDIA products are expected to account for a substantial portion of its annual substrate capacity. Taiwan's Unimicron will expand its KF site and begin supplying additional capacity from the second quarter of 2027, reaching its post-expansion target capacity in mid-2028.
In semiconductor testing, NVIDIA will make an $1.1 billion advance payment to KYEC to support construction of a new US facility and equipment installation. The target for starting volume production is April 2027. A testing capacity expansion in Singapore is expected to begin volume production earlier, in February 2027. The schedule for each process announced by NVIDIA is summarized below.
| Process / Company | Region | Ramp-up schedule and target |
|---|---|---|
| Substrates / Samsung Electro-Mechanics | South Korea | Supply begins immediately from existing sites; equipment upgrades and new plants also planned |
| Testing / Singapore expansion | Singapore | Volume production expected in February 2027 |
| Testing / KYEC | United States | Targeting volume production in April 2027 |
| Substrates / Unimicron | Taiwan | Additional supply from Q2 2027; expected to reach target capacity in mid-2028 |
| Optical engines / SPIL | Taiwan | 15x by mid-2027, then a further 2x by the end of 2028 |
| Substrates / Ibiden Ono plant | Japan | Additional capacity starts up from Q1 2028; target is 3x current level |
| Interposers / Amkor | South Korea | Target of more than 300% wafer capacity increase by the end of 2028 |
The schedule shows that even though testing equipment begins volume production in 2027, supply capacity for substrates and other items will not all expand at the same time. For example, the first quarter of 2028 given for Ibiden is when additional capacity begins ramping up, not when capacity reaches 3x. It can take time for new equipment to go from the start of volume production to its target capacity. To judge the actual supply outlook, it is necessary to look separately at when volume production starts and when target capacity is reached for each process.
The $279 billion is not total investment in this expansion plan
According to NVIDIA's 10-Q filing with the US Securities and Exchange Commission (SEC) for the second quarter of fiscal 2027, its supply and capacity commitments reached $279 billion as of July 26, 2026. That is an increase of $160 billion from $119 billion in the prior quarter. However, these are mainly multi-year supply-securing agreements for memory and manufacturing equipment, and do not mean NVIDIA is newly investing $279 billion in the expansion plan with the six companies announced on October 21. Nor has the full amount necessarily been paid.
StockTitan's analysis points out that it has not been disclosed whether advance payments such as the $1.1 billion to KYEC are included in the $279 billion total. Total commitments and individual advance payments may cover different transactions and be measured at different points in time, so the two cannot simply be added together to calculate NVIDIA's total investment. Even for advance payments in this announcement, spending to secure contracted capacity should be distinguished from the total investment required for items such as plant construction.
According to NVIDIA's 10-Q, some supply agreements may allow cancellation, postponement, or changes before firm orders are placed. However, changing a contract may incur additional costs, and if demand or business plans differ from expectations, the contractual burden may not be sufficiently reduced. Securing capacity in advance is effective for responding to growing AI demand and avoiding shortages. On the other hand, if demand falls short of expectations, reserved capacity and contractual obligations could become a financial burden.
What matters is the path from volume production to target capacity
The collaboration between Amkor and NVIDIA has also been under way in the US, before this capacity expansion in South Korea. According to Amkor's July 23, 2026 announcement, the two companies signed a multi-year advanced packaging and development agreement worth $1.5 billion. The plan is to use NVIDIA's advance payment to support capacity expansion in Arizona. The latest announcement adds expanded interposer wafer capacity in South Korea, so the region and process are different.
| Announcement date | Initiative with Amkor | Disclosed amount / capacity target |
|---|---|---|
| July 23, 2026 | Expansion of advanced packaging capacity in Arizona, US, and more | Multi-year agreement worth $1.5 billion |
| October 21, 2026 | Expansion of interposer wafer capacity in South Korea | Capacity increase of more than 300% by the end of 2028 |
The $1.5 billion contract announced in July cannot be regarded as the cost of the expansion in South Korea. Taken together, the two initiatives show NVIDIA securing capacity for different manufacturing processes in the US and South Korea to strengthen its advanced packaging supply chain. Again, though, it is important to distinguish between signing contracts or starting capital investment and actually expanding usable capacity. Even if plant construction and equipment installation proceed as planned, additional time is needed between starting volume production and reaching target capacity.
Moreover, more semiconductor and component supply capacity does not by itself mean more AI infrastructure going into operation. In the 10-Q, NVIDIA explains that, on the customer side, securing land, power, and buildings, as well as financing, may also constrain AI infrastructure deployment. Even if production capacity for GPUs and related components grows, the timing of final system deployments will slip if data center construction and power supply cannot keep up. How well manufacturing-side expansion plans and customer-side facility preparation line up is key.
The next milestones are the start of volume production at testing facilities in February and April 2027, and SPIL's target of 15x optical engine capacity by mid-2027. In 2028, substrate expansion at Ibiden and others and Amkor's interposer capacity expansion are expected to get fully under way. Whether volume production starts as scheduled in each process and whether capacity reaches target levels will need to be monitored. How much NVIDIA's expansion plan contributes to the supply of AI infrastructure will depend not only on how individual plants ramp up, but also on how well overall supply capacity across the manufacturing chain meshes with customers' deployment readiness.