Intel CEO: 18A Yield Reaches 70%, 14A Targets Tape-Out in 2027 and Mass Production in 2028

nashnova research
今天发布阅读约 14 分钟

Intel CEO Lip-Bu Tan disclosed that 18A process yield has reached 70% and the first external foundry customer has signed on. 14A targets tape-out in 2027 and volume production in 2028 — if delivered on schedule, the gap with TSMC's leading node narrows to 'slightly behind.'

01

18A yield at 70% — what does that number actually mean?

When Lip-Bu Tan took over, 18A yield was at a low level. After sustained improvement it has reached 70% and is heading toward mass production.
This means → 70% is the dividing line between "it works in the lab" and "factories can produce it profitably." Above that threshold, enough chips on each wafer are usable to make volume economics work.
The first product to use 18A is Intel's own Diamond Rapids — the strategy is to prove the process on internal silicon before opening it to outside customers.
The first external foundry customer has signed; external volume production is expected to begin between 2028 and 2029.
02

The 14A timeline is on the table — can Intel catch TSMC?

14A internal products target tape-out in 2027 and volume production in 2028; external customer production follows in 2028–2029.
In plain terms = tape-out — printing a design onto silicon for the first time — is "submitting the exam"; volume production is "shipping the product." Intel is giving itself roughly a one-year tuning window.
Lip-Bu Tan said the gap with TSMC's comparable leading node would narrow to "slightly behind" by that point.
This reflects a narrative shift from "far behind" to "comparable" — but "slightly behind" is Intel's own assessment and still needs real production data to back it up.
03

What does it take to earn foundry customers' trust?

Lip-Bu Tan described foundry as "a trust-based service business" and laid out three technical thresholds for customer entry: process performance at a specified level, SRAM yield — the pass rate of on-chip cache memory — meeting customer requirements, and a defect density target of 0.1.
Intel released PDK 0.9 (the design toolkit customers use to lay out their chips) for 14A in October 2024; the production-ready version is due in Q1.
Tan noted his nearly four-decade personal relationship with TSMC founder Morris Chang, saying he understands how long it takes to build foundry trust and will set "reasonable expectations."
This means → Intel is managing market expectations — foundry revenue will not spike overnight; the cadence is "prove first, then scale."
04

Why are PCs sluggish while data centers can't get enough supply?

PC side: inventory built up in Q3 2024, partly because memory prices surged roughly six- to sevenfold, dragging down demand for phones and lower-end PCs.
Data-center side: demand remains strong — Tan said "existing supply simply can't keep up" — but Intel 3 process capacity limits constrained Intel's supply pivot to data centers to about 78%, below AMD's roughly 97% in the same period.
In plain terms = Intel wanted to shift production from sluggish PCs to undersupplied data centers, but its own factory capacity ceiling and certain product designs — which use more memory and packaging substrate than competitors — limited how far it could shift.
Tan said the next-generation products will address the excess-material problem at the architecture level.
05

How does Intel plan to regain CPU leadership — and what is Tick-Tock?

Lip-Bu Tan named restoring CPU leadership as his top priority. The internal team is advancing Diamond Rapids and Coral Rapids, adding SMT — simultaneous multithreading, a technique that lets one CPU core handle multiple tasks at once — to boost performance.
He has also assembled a separate 141-person team, formed just two months ago, responsible for an independent next-round iteration that will merge back into the main organization at the right time.
Intel will return to the Tick-Tock cadence. Put simply = one generation upgrades the manufacturing process (Tick), the next upgrades the chip architecture (Tock), alternating so neither track stalls.
06

Where is the verification window for all of this?

2028 is the critical checkpoint: only if 14A reaches volume production on schedule and closes the timing gap with TSMC can Intel's IDM vertical integration — designing and manufacturing its own chips without relying on an outside foundry — deliver tangible results.
This means → the window for the market to reassess Intel's dual-track repair — foundry and products — is locked around 2028.
Until then, every timeline is a promise, not a result. The market will score Intel on production data, not roadmaps.

市场有风险,内容仅供研究参考,不构成投资建议。