By Grace Kim

GlobalFoundries has landed a $2 billion contract from TSMC to manufacture silicon interposers for AI processors, in what Tech Times calls the first dedicated US interposer supply line for AI chips.

Interposers are thin silicon slabs that carry thousands of microscopic electrical paths between a processor die (the main rectangular chip) and adjacent memory stacks. Without them, today's AI accelerators can't deliver the bandwidth large language models demand. The agreement pulls one step in advanced packaging — the post-fabrication stage that joins multiple chips into one functional unit — onto US soil at commercial scale for the first time.

The headline is big. The details are thinner.

What the deal covers

Tech Times didn't name the receiving GlobalFoundries fab, disclose the technology node (the process generation used to make the interposer — older nodes are cheaper, newer ones carry higher-density routing), or list contract start and end dates in its initial report. The $2 billion figure is the headline value attached to the multi-year pact. Neither GlobalFoundries nor TSMC had responded to confirmation requests at publication time.

Why the interposer step matters

Every top AI accelerator shipped today relies on TSMC's chip-on-wafer-on-substrate, or CoWoS, flow — a 2.5D packaging technique that mounts a GPU alongside stacks of high-bandwidth memory on a single interposer. Until now, every commercial CoWoS interposer came from Taiwan.

That monopoly has consequences. CoWoS capacity is the gating factor on NVIDIA's H100 and H200 deliveries, and analysts expect the same constraint to hold for the Blackwell-generation parts (NVIDIA's 2024 GPU family succeeding Hopper) shipping through 2025. A US-made interposer source doesn't add new CoWoS output — TSMC still runs the final packaging step — but it removes one Taiwan-bound link from a constrained chain.

What this changes for chip buyers

It gives US AI accelerator vendors, hyperscalers (cloud platforms running AI training and inference at scale), and federal procurement programs a domestic input that previously had no alternative supplier. That's politically and operationally significant, and it gives Washington a tangible onshoring win on a packaging layer nobody else has replicated outside Taiwan.

The deal also slots into a broader onshoring push. The CHIPS and Science Act of 2022 (federal legislation allocating roughly $52 billion in semiconductor incentives) has already directed grants to packaging investments from Micron and Intel. GlobalFoundries itself has received CHIPS funding for its Malta, New York expansion. A $2 billion TSMC contract adds the interposer layer to that footprint.

What to watch next

Three signals will tell readers whether the headline becomes real capacity.

• Site selection: confirmation of a GlobalFoundries fab — likely Malta, NY, or Essex Junction, VT — running the line.
• Customer disclosure: any US AI accelerator vendor confirming it will receive GF-built interposers inside production parts.
• Integration path: whether finished interposers route back to TSMC's Arizona advanced-packaging facility, itself still ramping through 2025, or to a third-party OSAT (outsourced semiconductor assembly and test vendor).

What does this change for the AI chip supply chain?

One link moves onshore. The chokepoint — TSMC's CoWoS output in Taiwan — does not. The deal's real impact hinges on whether GlobalFoundries can match the yields TSMC runs in Hsinchu today, and on how quickly US packaging capacity scales alongside it. Watch the site announcement first. Everything else follows from there.