Summary

  • GlobalFoundries and Monolithic Power Systems announced a long-term agreement on 9 September to bring MPS process technology to a 300mm Singapore fab, with capacity expansion targeted for early 2027.
  • Expected products include AI/cloud smart power stages. MPS already has Singapore wafer sources; the new issue is another manufacturing relationship for its own technology, not a first entry into the country.

More than a chip design crosses the factory door

The AI power-supply chain is getting another planned manufacturing route. Under the GlobalFoundries–MPS agreement, the customer brings proprietary process technology to the foundry’s Singapore equipment. That is different from treating a factory as an interchangeable source of generic wafers. A process must produce the required device characteristics repeatedly before additional manufacturing space becomes dependable component supply.

The announcement identifies smart power stages for AI and cloud infrastructure among expected products, alongside automotive power solutions and industrial robotics and automation. It does not allocate the whole agreement to AI, name specific components or publish wafer quantities. Early 2027 is the stated expansion target, not evidence that the new output is already shipping.

For data-centre buyers, these are components in the electrical delivery chain, not new processors or additional building power. As a description of the component class, onsemi’s compute-server overview places smart power stages with multiphase controllers and converters at the point of load serving processors and ASICs. That explains why manufacturing such parts matters to server supply. It is not a performance comparison or a description of the MPS products that will emerge from Singapore.

Fabless does not mean process-free

MPS’s 2025 annual filing makes its operating model unusually relevant to this announcement. The company uses third parties for manufacturing, assembly and testing, but develops proprietary process and packaging technologies and works with foundries to install its technologies on their equipment. It contrasts this with using a foundry’s standard process and design rules.

The distinction changes what has to travel between the two companies. A circuit design alone does not explain the relationship. The manufacturing method and the knowledge needed to reproduce it also matter. For MPS, the attraction of outside production can coexist with close involvement in how production works. A fabless model is not a claim that the company has no engineering infrastructure, testing activities or capital expenditure.

This gives the agreement a practical test beyond nominal capacity. The parties need results that can be reproduced, with a workable way to handle changes in the process or equipment. These are implications of the model, not disclosed milestones or evidence of an existing problem at GF. The release supplies neither qualification results nor yield, pricing or cost-allocation terms.

MPS also says it generally protects manufacturing-process technologies as trade secrets, while seeking patents for aspects of circuit and device designs. That is a description of company policy. It does not reveal the ownership, licensing, access or confidentiality provisions of the new agreement. The relevant commercial point is whose technology has to work on whose equipment—not an assumption that rights have been sold or remain governed by a particular unpublished clause.

A new relationship is not a new country

The same filing already lists Singapore among MPS’s wafer-manufacturing locations, together with China, Taiwan and South Korea. GF’s announcement therefore adds a named relationship and a production plan; it does not establish MPS’s first Singapore wafer source or first supply outside China.

Nor does the wafer location describe the entire route to a finished component. The filing places sorting, assembly, packaging and final testing across several Asian locations and combinations of company and partner facilities. The new release does not announce that all those stages will move to Singapore. A geographically more varied wafer supply may still depend on other stages elsewhere.

The 300mm figure needs an equally narrow reading. It describes wafer diameter, not a process node, annual output or a disclosed cost advantage. Without the product mix, production terms and manufacturing results, it cannot be converted into an estimate of available AI power components.

The opportunity is real but specific: another foundry relationship could extend the reach of MPS’s technology into growing demand for power management. Its value will become clearer when the process, later production stages and customer requirements join into a repeatable supply route. A new name on the factory list starts that assessment; it does not finish it.