Intel Eyes Google and Amazon to Challenge TSMC in Advanced AI Packaging with EMIB

For some time, Intel has been trying to convince the market that its future isn't solely about re-competing in traditional chip manufacturing, as it has done for most of its history, but also about gaining influence in a much more discreet segment of the business: advanced packaging. This is where today's rumor, published by WIRED, comes into play, pointing to active discussions with Google and Amazon to utilize Intel's advanced packaging services, specifically Intel EMIB or EMIB-T.

Is this a genuine, silent overtaking of TSMC, or a limitation on Taiwan's ability to produce more? While no confirmed contracts exist, the mere mention of these two names speaks volumes about the type of clients Intel needs for its Foundry to transition from a promise to a serious, and above all, profitable business.

Intel in Negotiations with Google and Amazon for Advanced Packaging Based on EMIB or EMIB-T

This thesis makes considerable sense from Intel's perspective, driven by the profitability sought for IFS. AI now requires more than just fabricating a more powerful GPU; it demands uniting logic, chiplets, interconnections, and HBM memory within the same package with minimal loss, lowest possible power consumption, and optimal space utilization. Furthermore, it needs to be as cost-effective as possible, complemented by open manufacturing capacity that isn't already saturated.

TSMC undeniably dominates this field with CoWoS, but Intel has been building its own path for years with EMIB and Foveros. Now it aims to sell this capacity to third parties in volume, and this is where Amazon and Google fit quite naturally, as both design their own chips and outsource critical parts of the process.

EMIB is important because it allows connecting different dies within the same package without requiring a complete, much more expensive, full-size interposer. Simply put, Intel proposes a more localized and selective connection between components.

If the Rumor Becomes Reality, it's a Giant Leap for IFS

WIRED even suggests that, within this context, EMIB and EMIB-T are viewed as a more "surgical" approach than TSMC's proposal, with the idea of saving space, improving efficiency, and containing long-term costs for the AI race. Additionally, manufacturing on U.S. soil would complete the circle, especially for clients designing increasingly large and complex accelerators. This promise ceases to be a technical detail and becomes a strategic issue that would satisfy all parties. That is, on paper, it makes sense; now, the actual figures need to be on the table.

EMIB-T further strengthens this interpretation. Intel announced it in May 2025, and according to the company, this evolution improves signal integrity and power efficiency between components, with factory deployment planned for — curiously — 2026. This is precisely where the rumor gains significant strength: Amazon and Google wouldn't need to commit to Intel for all their chips; entrusting a specific part of advanced assembly to Intel would suffice if they perceive it resolves a genuine bottleneck with TSMC.

The logical conclusion of this story lies in 2024, and indeed, it reinforces it. Intel and AWS announced in September of that year an expansion of their collaboration with a multi-year, multi-billion-dollar framework for custom chips, including an AI fabric chip on Intel 18A and a custom Xeon 6 on Intel 3.

This wasn't a packaging agreement per se, but it was proof that an industrial relationship already existed between the two parties and could scale with Amazon and Google for EMIB or EMIB-T within Intel Foundry Services. Seen this way, the current rumor doesn't emerge from nowhere; it builds upon a real precedent. If confirmed, if Intel manages to secure another major agreement... keep an eye on the blues; they could return to the global top 10 FAB list before the year ends.