When people talk about GPU shortages, they almost always think of the chip’s silicon. But a lesser-known part, the interposer, is one of the bottlenecks. On October 8, 2026, TSMC and GlobalFoundries announced a $2 billion deal to make them in the United States.
According to The Register and other trade outlets (DatacenterDynamics, AnySilicon), GlobalFoundries will make silicon interposers for TSMC at its Malta, New York plant under a five-year, $2 billion contract, with extension options and a framework for expanding capacity. Volume production would not start before the first half of 2028.
An interposer is a sheet of silicon placed beneath the compute logic and memory in an advanced assembly. According to The Register, it provides a faster, lower-power communication path between chips than conventional organic packaging, which is key in multi-die GPUs and accelerators with high-bandwidth memory (HBM). TSMC’s standard that uses them is CoWoS (chip-on-wafer-on-substrate). The line that sums up the problem: practically every chip with HBM today depends on some kind of interposer or interconnect bridge.
That’s why CoWoS capacity has been described as one of the bottlenecks for AI accelerator supply: it’s not enough to make the chip, you have to assemble it.
The Register recalls a telling detail: when TSMC began producing Nvidia Blackwell GPUs in Arizona, final assembly still required shipping the wafers back to Taiwan. Producing interposers in Malta tackles part of that problem, and the companies present it as the first U.S. source of silicon interposers for advanced packaging.
Cloud compute capacity remains tied to a physical supply chain with few suppliers. For anyone buying AI: (1) don’t assume falling inference prices in the short term just because of "more chips"; packaging limits supply; (2) diversify cloud providers and regions so you don’t depend on a single reserved capacity; (3) choose efficient models and architectures, because every GPU-hour will remain a scarce resource. If you need to size AI compute costs for your company, write to info@guatemalia.com.
Carlos Montiel is an enterprise AI solutions architect. He implements LLMs, Agents, RAG, and orchestrators for companies across Guatemala and Latin America. Reach out for a consultation.
Contact Carlos Montiel