JPMorgan Pours Cold Water on Intel's AI Orders: Google and NVIDIA Partnerships Lack Meaningful New Volume
N.R. Finch
JPMorgan says the market has read too much into Intel's reported AI chip deals — what's visible from Google, Nvidia, and SK Hynix points almost entirely to packaging and early testing, not an advanced foundry breakthrough.
The "Intel comeback" narrative — how much is real?
The Information reported that TSMC capacity constraints are creating an opening for Intel, with Google, Nvidia, and others eyeing it as a backup manufacturer.
JPMorgan pushed back directly: the report looks more like "making a big deal out of very little" and provides no clear new information.
This means → the market narrative around "Intel replacing TSMC" lacks hard evidence, in JPMorgan's view.
Google's 3-million-TPU order — what does Intel actually get?
Google plans to deliver over 3 million TPUs by 2028 — on the surface, a major win for Intel.
But the actual split: compute chiplets (the cores that do the math) stay on TSMC's N2 process; IO chiplets (the parts that move data in and out) use TSMC's N3 process.
Intel's role is limited to EMIB-T advanced packaging — assembling different chiplets together — and this work is already baked into JPMorgan's model, so it adds no new revenue.
In plain terms = TSMC still *makes* the chips; Intel only helps *assemble* them.
Nvidia and SK Hynix — the same story?
Reports say Nvidia is testing Intel's capabilities for its next-gen Feynman architecture, but JPMorgan believes the tests target interconnect or packaging, not core compute chip manufacturing.
The reports do mention Nvidia running early wafer trials on Intel's 18A process node, but JPMorgan stresses this is still at the early-test stage — not a production-volume commitment.
SK Hynix's engagement falls into the same bucket — the focus is on testing packaging capability, not wafer fabrication.
This means → all three customers point to one conclusion: Intel is winning packaging and test opportunities, but the door to advanced foundry work hasn't truly opened.
TSMC is capacity-constrained — how big is Intel's real opportunity?
JPMorgan acknowledges that TSMC's tight capacity does give Intel more chances to enter customer supply chains.
But every visible form of engagement — packaging, interconnect, early testing — falls short of proving Intel has secured a decisive advanced foundry order.
This reflects a critical gap: the distance from "being tested by a customer" to "winning a production contract" remains vast.
In plain terms = Intel has made the shortlist, but it hasn't received the offer letter.
Content is for reference only, not financial advice.