Google, Amazon prepay tens of billions

- Intel is in talks with Google and Amazon to package custom AI chips, seeking multibillion-dollar annual foundry revenue from advanced assembly starting in 2026. - Intel executives said customers are willing to prepay billions for scarce packaging capacity, as Google rolls out TPU 8 and Amazon expands Trainium deployments. - The race has shifted to packaging, where capacity constraints now shape AI chip supply. (trendforce.com)

Intel is in talks with Google and Amazon to package their custom artificial intelligence chips, turning a back-end manufacturing step into a new battleground for cloud computing. (finance.yahoo.com) Chip packaging is the stage where separate pieces of silicon are connected into one working processor, and advanced packaging lets companies stitch together more compute and memory without building one giant chip. Intel is pitching its Embedded Multi-die Interconnect Bridge, or EMIB, as that connector layer for hyperscaler AI processors. (trendforce.com) The customers in those talks are Google, which builds Tensor Processing Units, and Amazon, which builds Trainium and Inferentia chips for Amazon Web Services. Multiple reports said the potential Intel work centers on packaging rather than wafer fabrication. (finance.yahoo.com) (www.idnfinancials.com) Intel has told investors that packaging demand is now arriving before major outside wafer-manufacturing revenue. In Intel’s April 23, 2026 first-quarter materials, chief executive Lip-Bu Tan listed advanced packaging as one of the company’s three strategic assets for the artificial intelligence buildout. (newsroom.intel.com) (download.intel.com) The money at stake is large, but the public record does not support the claim that Google and Amazon have already prepaid “tens of billions.” The reporting tied to these talks says Intel executives described customers as willing to prepay billions of dollars for capacity, with potential deals worth billions per year. (trendforce.com) (electronics360.globalspec.com) That distinction matters because packaging has become one of the tightest choke points in the AI supply chain. Reports on the Intel talks point to limited capacity at Taiwan Semiconductor Manufacturing Co., especially for the advanced packaging used in high-performance AI systems. (trendforce.com) (www.techspot.com) Google’s side of the story became more concrete last week, when it introduced its eighth-generation TPU family at Cloud Next on April 22. Google said TPU 8t is aimed at large-scale training and TPU 8i at inference and post-training, with availability later this year. (blog.google) (cloud.google.com) Amazon’s public roadmap is less aligned with the rumor mill. AWS’s current Trainium page lists Trainium1, Trainium2 and Trainium3, not Trainium4, and Amazon said on April 20 that Anthropic had secured up to 5 gigawatts of current and future Trainium capacity, including significant Trainium3 capacity expected this year. (aws.amazon.com) (www.aboutamazon.com) Intel is also building out the physical footprint to chase this business. Reports say its advanced packaging operations in Malaysia are slated to come online in 2026, while its Rio Rancho, New Mexico facilities are already part of its 3D packaging production network. (trendforce.com) So the cleaner version of the story is narrower than the social posts: Intel is trying to win Google and Amazon as advanced-packaging customers, and executives have signaled possible prepayments in the billions, not confirmed prepaid commitments in the tens of billions. (finance.yahoo.com) (trendforce.com)

Get your own daily briefing

Scout delivers personalized news, insights, and conversations tailored to your role and industry.

Download on the App Store

Shared from Scout - Be the smartest in the room.