Google's 3 Million TPU Order Gives Intel Foundry Its First Whale

The Foundry Finally Has a Customer Worth Naming

Intel Foundry has spent two years as the hardest part of the bull case to underwrite: a capital sink with no anchor tenant. That framing took a hit when The Information reported that Google placed an order for more than 3 million of its custom Tensor Processing Units to be built by Intel for 2028 production — the largest external manufacturing win in the foundry's history.

The timing is the tell. The order lands as TSMC's leading-edge capacity is sold out well into the AI buildout, pushing hyperscalers to qualify a second source whether they want to or not. Intel is the only Western fab with a credible advanced node, and 18A is the bet. Nvidia is reportedly testing Intel as a backup manufacturer too, though no order from Jensen Huang has been placed — that's still in the qualification phase. One whale and a tire-kicker is not a book of business, but it's the first hard evidence the foundry can win work it doesn't have to subsidize itself.

Why BofA's Double Upgrade Suddenly Has Teeth

The Google headline doesn't sit in a vacuum. A day earlier, BofA's Vivek Arya ran a rare double upgrade — Underperform straight to Buy, skipping Neutral, with the target jacked from $96 to $135. The note models earnings power above $6 per share by 2030, up from a prior $3–$4 range, and leans on two engines: a server-CPU TAM lifted to more than $170 billion (from $125 billion) on agentic-AI workloads, and a foundry business BofA sees scaling from $1.1 billion in 2026 to $47.1 billion by 2030.

That $47 billion foundry number was the part most worth fading 48 hours ago — it's a forecast built on wafers, packaging, and rumored Apple and MediaTek work. The Google TPU order is the first data point that makes the curve look less like a spreadsheet and more like a backlog. Sell-side conviction plus a named customer is a more durable combination than either alone.

What's Actually In Play

Intel is up roughly 210% year-to-date, tracking its best year since 1975, and still sits in only about 16% of S&P 500 funds — among the least-owned names in its cohort. That combination is the live positioning trade: a stock this far up that this many funds still don't own is a magnet for chase-and-cover, which is part of what the foundry headline is feeding.

The bear rebuttal is intact and worth respecting. The Google order is for 2028 — three years out, contingent on 18A hitting volume yields it hasn't proven, and orders that distant can shrink. After a move this size, the near-term risk is mechanical: profit-taking into thin books. HIP-3 perp volume on this market ran about $91.6 million over the last 24 hours, so there's real two-way flow to absorb it. What invalidates the move isn't a tweet — it's any 18A yield wobble or a hint the Google commitment is softer than the headline reads.