Intel made several big announcements about its 7nm tech at today's Intel Unleashed: Engineering the Future event and divulged that it expects that the majority of its products in 2023 to still be produced in-house using its own manufacturing technology. But there's a caveat: Intel CEO Pat Gelsinger said the company will also release "leadership CPU products" in 2023 with CPU cores that are fabricated with an unspecified process node from third-party foundry TSMC, and those CPUs will come to both the client and data center markets.
This development comes on the heels of Intel's announcement last year that its 7nm process was delayed, possibly forcing it to do the unthinkable — turn to external foundries to produce its core logic, like CPUs and GPUs, for the first time in the company's history.
The newest announcements mean that, in addition to the 7nm Meteor Lake desktop chips and Granite Rapids data center processors that Intel will produce with its own process technology in 2023, the company will also release other lines of CPUs in 2023 that will use CPU cores with an as-yet-unspecified process node from TSMC. Intel noted that the chips that utilize TSMC's third-party process tech will power Intel's "CPU leadership" products for both the client and data center markets, suggesting a split product stack.
Intel says that the majority of its products in 2023 will come manufactured with its own process technology. Still, it's important to note that Intel hasn't specified that the majority of the newly-released 2023 products will come with its own 7nm process. Naturally, Intel will still have plenty of chip production volume centered on its 14nm and 10nm process tech in that timeframe, and even older nodes that still ship in large volumes.
Intel's distinction that its line of "leadership CPUs" with outsourced cores will come for both the client and data center markets could be telling. TSMC plans to begin high volume manufacturing of its 3nm process in the second half of 2023, meaning that Intel's 7nm could be outmatched by competitors, such as AMD, Apple, and ARM-based designs, with CPU designs etched on a more advanced node than Intel's. Samsung will also have more advanced 3nm technology than Intel in the same time frame.
TSMC and Samsung's newer process nodes might not be faster than Intel's 7nm; clock rates on newer nodes are expected to remain static or even possibly regress, but CPU architectural modifications can offset those issues. However, the denser processes will absolutely be more power-efficient and cost-effective (even though per-transistor costs are rising with denser nodes), leaving Intel at a disadvantage against competing chip designers that have access to more advanced process technology.
As such, Intel's secondary lineup of leadership CPU products with outsourced cores could be comprised of the highest-end halo products built with the latest process node from TSMC, providing the company with more performance than it can achieve on its own process tech.
As we see with most semiconductor manufacturers, Intel's highest-end parts comprise a smaller percentage of its overall chip production. In fact, the mid-range and lower-end chips often comprise the overwhelming majority of sales, while halo products offer the highest-end performance at a premium and reinforce performance leadership and the brand.
Setting aside its highest-end products for an external process node makes a lot of sense — Intel currently produces more x86 processors than any other company, producing in the neighborhood of one million dies per day. That means that no single outside foundry can satisfy its production requirements. For instance, Intel has more than twice the semiconductor revenue of TSMC. Given that TSMC is already capacity constrained, it obviously wouldn't have enough output to satisfy all of Intel's incredible sales volume.
Intel will also likely leverage its packaging tech to reduce the number of externally-sourced components required to build a full chip. It's feasible that Intel could 'simply' swap out its own 7nm CPU tiles found in its Meteor Lake and Granite Rapids chips in exchange for compute tiles based on TSMC's smaller process node to build more competitive variants. Or we could see entirely different models based on the third-party process tech.
As the highest-end products typically comprise the smallest amount of overall sales, outsourcing portions of its high-end CPUs would allow Intel to satisfy its larger production volume requirements with its own 7nm tech, while remaining competitive with its lower-volume leadership products fabbed on an external process node.
Intel also isn't divulging whether it will fab the chips with an external process in its factories via a technology licensing agreement or if the chips will be built at TSMC. Outgoing CEO Bob Swan told us that technology licensing was a possibility.
Intel has a long history of producing low-value simple chips at several foundries, like TSMC, Samsung, GlobalFoundries, and UMC. Intel has also traditionally used third-party fabs, currently to the tune of ~20% of its production (opens in new tab), for low-margin, non-CPU products, like chipsets and Wi-Fi chips, built on trailing-edge nodes. Intel hasn't shared how much it expects to expand manufacturing with third parties.