A $2 Billion Advanced Packaging Agreement

GlobalFoundries announced the agreement on October 8, 2026. Under the deal, GF will provide manufacturing services to TSMC and expand capacity at its Malta facility specifically to support silicon interposers used in TSMC’s CoWoS advanced packaging ecosystem.

GlobalFoundries says the Malta site is expected to become the first U.S.-based source of silicon interposers supporting advanced packaging technologies. The planned capability also includes manufacturing support for embedded deep-trench-capacitor components used in advanced packaging structures.

Why Silicon Interposers Matter for AI Chips

A silicon interposer is a thin piece of processed silicon that sits between high-performance logic dies, high-bandwidth memory and the package substrate. It contains very fine wiring that allows large numbers of signals to move between chips over short distances.

This is increasingly important in AI accelerators and high-performance computing systems. The compute die alone is no longer the complete product. Modern AI processors must be connected to multiple HBM stacks and, in some cases, additional chiplets. The interposer makes these high-density connections possible while helping provide the bandwidth and power efficiency required by advanced systems.

Advanced packaging has therefore become strategically important. Reuters reported that packaging capacity has become a significant constraint on AI chip production as demand has grown faster than available manufacturing capacity.

How the Deal Fits into TSMC CoWoS

TSMC’s CoWoS — Chip-on-Wafer-on-Substrate — platform is one of the key packaging technologies behind today’s AI and HPC processors. It allows multiple logic and memory dies to be integrated into one high-performance package.

TSMC currently offers three principal CoWoS variants:

• CoWoS-S uses a silicon interposer and is designed for very high-density interconnect between logic and HBM.
• CoWoS-L combines an RDL-based interposer with embedded local silicon interconnect structures.
• CoWoS-R uses redistribution-layer interconnect technology rather than a full silicon interposer.

The GlobalFoundries agreement was announced broadly as support for TSMC’s CoWoS ecosystem. Because GF will manufacture silicon interposers, the new capability most directly aligns with silicon-interposer-based packaging such as CoWoS-S, although neither company has publicly limited the agreement to one specific CoWoS variant.

TSMC Is Scaling CoWoS to Larger AI Packages

The timing is important. TSMC continues to increase the size and complexity of its advanced packaging platforms to support more compute dies and additional HBM capacity.

TSMC says it certified a CoWoS solution using interposers as large as 5.5 times the conventional reticle size in 2025, with volume production beginning in 2026. Larger interposers allow AI accelerator designers to integrate more logic and memory inside a single package, but they also increase the amount of interposer manufacturing capacity required.

As AI accelerators become larger and HBM content increases, interposer manufacturing is therefore becoming a meaningful capacity requirement rather than a small supporting step in the semiconductor supply chain.

Why GlobalFoundries?

The agreement is notable because GlobalFoundries is not being asked to manufacture TSMC’s leading-edge logic wafers. Instead, it will use its U.S. manufacturing footprint to supply a critical component of the advanced packaging flow.

This demonstrates how semiconductor manufacturing is becoming more specialized. Advanced AI systems combine leading-edge logic, HBM, interposers, package substrates and complex assembly processes. Not every part of that system requires the same transistor technology.

GlobalFoundries’ Malta site provides an established U.S. semiconductor manufacturing base that can be expanded for interposer production, while TSMC retains control of its CoWoS ecosystem and customer relationships.

A New U.S. Link in the CoWoS Supply Chain

For the United States, the agreement addresses a specific weakness in the domestic semiconductor supply chain. Significant investment has already been directed toward front-end wafer fabrication, but advanced packaging remains another critical part of building AI hardware locally.

Establishing silicon-interposer manufacturing in New York creates a domestic source for an important CoWoS component and gives TSMC additional geographic flexibility. It does not mean that the entire CoWoS process will move to GlobalFoundries or that GF becomes a substitute for TSMC’s advanced packaging operations. Instead, GF becomes a manufacturing supplier for a key element within TSMC’s packaging ecosystem.

What the Deal Means for GlobalFoundries

For GlobalFoundries, the agreement opens another path into the fast-growing AI infrastructure market. GF is traditionally associated with specialty semiconductor manufacturing in areas such as RF, power management, automotive, connectivity and embedded technologies rather than leading-edge AI processors themselves.

Interposer manufacturing gives the company a way to participate in the AI value chain without competing directly with TSMC in advanced logic nodes. It also allows GF to use its process and manufacturing expertise in a component where wafer-level precision, yield and capacity are becoming increasingly important.

The five-year agreement provides an anchor customer for the planned expansion, while the capacity-expansion framework creates the possibility of additional investment if CoWoS demand continues to increase.

What the Deal Means for TSMC

For TSMC, the partnership adds another source of silicon-interposer capacity and begins building a more geographically diversified packaging supply chain.

TSMC continues to expand its own CoWoS capability, but AI demand is growing rapidly and advanced packaging requires enormous capital investment. Using GlobalFoundries as a manufacturing partner for selected interposer production can provide additional scale without requiring every component to be manufactured inside TSMC’s own facilities.

It also fits with the semiconductor industry’s broader move toward supply-chain resilience. Customers increasingly care not only about process technology but also about where wafers, interposers, substrates and packages are produced.

Advanced Packaging Is Becoming as Strategic as the Wafer Fab

The semiconductor industry historically focused heavily on transistor scaling and leading-edge wafer capacity. AI is changing that balance.

A state-of-the-art GPU or custom AI ASIC can only deliver its intended performance if the package can connect it efficiently to HBM and other chiplets. As systems become larger, advanced packaging increasingly determines memory bandwidth, power delivery, signal integrity, thermal performance and total system size.

The GlobalFoundries–TSMC deal is another sign that the packaging layer is moving from a backend manufacturing consideration to a strategic part of semiconductor capacity planning.

Volume Production Expected in 2028

The new Malta capacity is not expected to affect CoWoS supply immediately. GlobalFoundries expects volume production to begin ramping during the first half of 2028.

That timeline reflects the complexity of adding and qualifying new semiconductor manufacturing capability. Equipment must be installed, processes developed or transferred, customer specifications qualified, and manufacturing yield brought to production levels.

The agreement is therefore best viewed as a long-term capacity move aimed at the next phase of AI and HPC growth rather than a short-term solution to today’s packaging shortage.

Conclusion

The $2 billion agreement between GlobalFoundries and TSMC is significant beyond the contract value itself. It creates the first planned U.S. source of silicon interposers for TSMC’s CoWoS ecosystem, expands GlobalFoundries’ role in AI infrastructure, and gives TSMC another path to scale a critical advanced-packaging component.

More broadly, the deal illustrates an important change in semiconductor manufacturing: competitive advantage is no longer determined only by who can fabricate the smallest transistor. As AI systems depend increasingly on HBM, chiplets and heterogeneous integration, the ability to manufacture interposers and assemble complex packages is becoming just as strategic.