GF’s TSMC deal could add U.S. interposer supply; volume ramp expected in first half of 2028

That value is not a disclosed capital-spending budget, guaranteed revenue for GlobalFoundries or a stated purchase commitment by TSMC. The public material does not explain how the $2 billion is calculated. It also omits planned output, capacity-expansion milestones, yields and delivery schedules. The agreement has a stated term, but the disclosed details do not show how much interposer supply TSMC will receive over it.

GlobalFoundries describes Malta as intended to become the first U.S.-based source of silicon interposers for advanced packaging. That is the company’s characterization, repeated in reporting; the sources do not independently survey other U.S. suppliers to verify the distinction. The company also describes the planned supply as secure and scalable, but the announcement provides no production record or output data to demonstrate those qualities.

The scope is component manufacturing, not an agreement for GlobalFoundries to fabricate TSMC’s leading-edge processors or produce complete AI systems. The reporting does not include a direct TSMC statement on the deal. Tom’s Hardware raised the possibility that the arrangement could help TSMC customers describe products as made and packaged in the United States, but the disclosed scope does not establish where the other components will be made or where final packaging will take place.

How could interposers affect AI-chip supply?

TSMC describes CoWoS as a packaging platform that can combine multiple system-on-chips with high-bandwidth memory, or HBM. The interposer is one way to connect components in a package. TSMC’s technical material describes several CoWoS approaches, including silicon-interposer and redistribution-layer designs, so the GlobalFoundries agreement should not be read as supplying a component used in every CoWoS package. TSMC says CoWoS has been in production since 2012; that history concerns the platform, not GlobalFoundries’ planned Malta capacity. TSMC’s technical description also describes the platform’s use for AI and high-performance computing.

The potential benefit is narrower than adding a new source of finished processors: it is another planned manufacturing source, located in the United States, for interposers serving TSMC’s packaging ecosystem. Reuters reported that demand for advanced packaging exceeds available manufacturing capacity. But neither that report nor the company announcement quantifies the constraint, the capacity GlobalFoundries will add or the share of TSMC’s interposer demand it could eventually supply. Reuters’ report, carried by Investing.com, therefore provides context for why another source could matter, not a measurement of this deal’s effect.

That gap matters because the agreement’s value and its manufacturing impact are different questions. A $2 billion headline does not tell readers how many interposers will be made, how many will enter TSMC packages or how much additional packaged hardware could result. The sources do not establish those figures, and they do not show that the agreement will resolve packaging constraints.

For the announced supply to translate into more available AI hardware, the planned capacity must produce components in volume and those components must enter the packaging supply chain. The reporting does not provide the milestones or output data needed to assess that process. Nor does the agreement establish that every additional interposer would produce an additional finished package; the other production steps and customer allocations are not detailed.

A U.S. production location could broaden the geography of supply for this particular component. It is not evidence that a complete AI system—or even every part of a TSMC customer’s package—will be made in the United States. The most supportable claim is a planned U.S. source for silicon interposers, with the amount of supply and its downstream effect still unknown.

Does the Xanadu photonics partnership add AI capacity?

GlobalFoundries and Xanadu announced a separate multi-year collaboration on Tuesday. Their initial work is focused on silicon nitride photonics and superconducting nanowire single-photon detectors, using GlobalFoundries’ 300 mm production line at Malta. Xanadu identifies the components as intended for its photonic quantum-computing systems, not TSMC’s CoWoS supply chain. The companies’ partnership announcement and DatacenterDynamics’ report describe a project distinct from the interposer agreement.

Xanadu presents the photonics work as a route to performance and chip-overhead benefits for its quantum systems, but the announcement supplies no independent measurements of those effects or production volumes. It also points to future fault-tolerant computing as an aim, not an outcome demonstrated in the material. The partnership may advance manufacturing of components for quantum systems; the sources do not establish that it will increase the supply of conventional AI accelerators.

The two agreements share a manufacturer and a facility, but not a stated product line. Treating the photonic components as part of the TSMC deal would conflate a quantum-computing collaboration with a separate supply arrangement for silicon interposers.

For the TSMC agreement, the meaningful test is whether GlobalFoundries begins the expected volume ramp in the first half of 2028 and later reports how much it produces. Until output and delivery figures are available, the agreement establishes a planned U.S. source for one CoWoS component—not a quantified change in AI-computing hardware supply.